Radio navigation system state evaluation method based on multi-dimensional data-augmented learning

Abstract

The present invention relates to the field of radio navigation system state evaluation. Provided is a radio navigation system state evaluation method based on multi-dimensional data-augmented learning. The method comprises: acquiring function operation monitoring data of a radio navigation system, and selecting data associated with system health; performing sliding window processing on the selected data, so as to obtain sample data including time information; dividing the sample data into a training set, a first test set and a second test set, and constructing positive and negative samples; establishing and training a contrastive learning model; establishing and training a deep one-class support vector machine; and using output results of the contrastive learning model and deep one-class support vector machine for the test sets to complete evaluation. In the present invention, deep features of measured data are learned by means of contrastive learning, and the deep one-class support vector machine is used to construct a state evaluation model. The present invention has the advantages of high detection precision and interpretability, has important theoretical and engineering practice significance, and realizes precise quantitative evaluation of the health state of an airborne radio navigation system.

Description

A radio navigation system state evaluation method based on multi-dimensional data enhanced learning TECHNICAL FIELD

[0001] The present application relates to the field of radio navigation system state evaluation, and particularly relates to a radio navigation system state evaluation method based on multi-dimensional data enhanced learning. BACKGROUND

[0002] In the process of aircraft performing flight tasks, the radio navigation system provides the relative distance, position and other information of the aircraft through real-time communication with the tower, realizes real-time positioning and ranging of the aircraft, determines the heading, avoids the no-fly zone and other functions, and is one of the important component systems in the aircraft.

[0003] The radio navigation system generally adopts the secondary radar ranging principle for ranging, and the bearing measurement is realized by rotating the ground station antenna to emit a beacon signal with direction information, and the aircraft determines the azimuth angle relative to the ground station according to the relative intensity and direction of the signal. In the tasks such as in-flight refueling, formation flight, air interception and the like, the radio navigation system plays an important role.

[0004] The radio navigation system generally comprises a receiver, a bearing solution unit, a control unit, a data connection interface and the like. Once the radio navigation system fails, it will lead to navigation failure, difficulty in determining the position, and even flight task failure. It is necessary to carry out abnormal detection research on the radio navigation system, to timely give an early warning when the radio navigation system fails, to improve the auxiliary ground command center to timely modify the flight task, and to reduce the flight or navigation safety risk possibly caused by the radio navigation system.

[0005] The state evaluation of the traditional radio navigation system has the following limitations in the aspects of data acquisition and evaluation method:

[0006] 1) The data for state evaluation is limited. The state evaluation of the radio navigation system is mainly based on test data obtained by two means of BIT (Built In Test) and ground test. The BIT method determines the health status of the equipment by monitoring the working state of the main digital or radio frequency circuit inside the equipment in real time, which has the advantages of real-time and continuity. However, the implementation of the BIT function needs to add additional embedded test circuit or software inside the equipment, which increases the volume, weight and power consumption of the equipment, and reduces the basic reliability of the equipment. Limited by the above factors, the test accuracy and fault coverage of the BIT method are limited, and it is difficult to achieve comprehensive coverage of the fault mode of the equipment. The ground test method, as a supplement to the BIT, can detect the function and performance index of the radio navigation system, and has higher test accuracy and more comprehensive fault coverage. However, the ground test method needs to use ground detection equipment, and the test process is relatively complex, and frequent testing will increase the burden of aircraft maintenance personnel, therefore, the timing of the ground test work is limited, and the test time interval generally depends on the inspection cycle of the aircraft, which is relatively long, and cannot realize real-time and continuous monitoring of the health status of the equipment.

[0007] 2) The traditional navigation system state evaluation is mainly based on the method based on expert knowledge, but due to the characteristics of high-dimensional, massive monitoring data and complex system composition of the radio navigation system, the method based on expert knowledge has the problems of low detection accuracy, insufficient abnormal space mining and the like. SUMMARY

[0008] In view of the problems in the prior art, a radio navigation system state evaluation method based on multi-dimensional data enhanced learning is provided, which learns the deep features of the detection data by contrast learning, and constructs a state evaluation model by using a deep one-class support vector machine, which has the advantages of high detection accuracy and interpretability, has important theoretical and engineering practical significance, and realizes accurate quantitative evaluation of the health status of the airborne radio navigation system.

[0009] The technical scheme adopted by the present application is as follows: A radio navigation system state evaluation method, comprising:

[0010] Obtaining radio navigation system function operation monitoring data, and selecting data related to system health;

[0011] Performing sliding window processing on the selected data to obtain sample data containing time information;

[0012] Dividing the sample data into a training set, a first test set and a second test set; wherein, based on the data type, positive and negative samples are constructed for all sample data;

[0013] Establishing a contrast learning model and training the training set;

[0014] a deep one-class support vector machine is established and trained, and an input of the deep one-class support vector machine is a feature representation extracted by the trained contrastive learning model;

[0015] The output results of the contrastive learning model and the deep one-class support vector machine are used as thresholds by using the first test set, and the output results of the second test set are compared to complete the state evaluation of the radio navigation system.

[0016] As a preferred solution, the data associated with system health selected includes the number of uplink received pulses, the number of downlink received pulses, the number of uplink received AGC responses, the number of downlink received AGC responses, the number of downlink received DSP responses, the time of being locked out, the distance, and the azimuth.

[0017] As a preferred solution, the specific method of obtaining sample data containing time information includes:

[0018] At every dth time point on the selected data associated with system health, a sliding window is taken, that is, at every dth time point, the number of samples + 1, until the sliding window slides to the end, and the total sample data is obtained.

[0019] As a preferred solution, the training set and the test set are divided according to a ratio of 6:4, and the test set is further divided into a first test set and a second test set according to a ratio of 1:1.

[0020] As a preferred solution, the specific method of constructing positive and negative samples includes:

[0021] Positive sample construction: generating positive samples for sample data in a data enhancement manner;

[0022] Negative sample construction: different data samples are negative samples for each other.

[0023] As a preferred solution, the specific method of constructing positive samples includes:

[0024] The sample data obtained by the sliding window is divided into discrete data and continuous data according to the data type, wherein the number of uplink received pulses, the number of downlink received pulses, the number of uplink received AGC responses, the number of downlink received AGC responses, the number of downlink received DSP responses, the time of being locked out, and the distance are discrete data; and the azimuth is continuous data.

[0025] For discrete data, random noise with an amplitude of 0 or 1 is added to the original sample data to generate new sample data, and the new sample data and the original sample data are positive samples for each other.

[0026] For continuous data, data enhancement is performed by adding Gaussian random noise to generate positive samples.

[0027] As a preferred solution, the training method of the contrastive learning model includes:

[0028] The contrast learning model is established, for any sample data of the training set, an enhanced sample pair of the sample data is generated by constructing positive and negative samples, the enhanced sample pair is input into an encoder to obtain a feature representation of the sample data, and then a multi-layer perceptron is used to obtain a feature nonlinear mapping, and the model is trained by using a contrast loss function, wherein the contrast loss function is:

[0029] wherein, is a sample batch, x i is the i-th sample data, and B represents a set of all sample data in a sample batch, is an enhanced sample pair batch representation, x i (1) ,x i (2) are enhanced sample pairs of x i , τ is a temperature coefficient, is a set of all other samples in the sample batch except the positive / negative sample corresponding to x , represents a sample loss for contrast learning with x as an anchor point; represents a sample loss for contrast learning with x as an anchor point, and specifically,

[0030] wherein, is a similarity calculation function for calculating the similarity between u and v, exp is an exponential function, are feature nonlinear mappings of x i (1) , x i (2) , and z j corresponds to x j ,

[0031] As a preferred solution, the specific process of establishing and training the deep one-class support vector machine is as follows:

[0032] The deep one-class support vector machine is established by combining deep learning and one-class support vector machine, which is used to identify and separate abnormal or outlier points; the input is the feature representation obtained by the encoder in the trained contrast learning model; the training of the deep one-class support vector machine is completed by using an optimization objective function, wherein the optimization objective function is defined as:

[0033] wherein, min represents minimization, for a set of weights, c is the hypersphere center, h i = f(x i ) is the feature representation of the i-th sample data, n is the number of samples in the given training set , φ(W) is a neural network, w l represents the weight of the l-th layer hidden layer, l = 1, 2, …, L, λ is a regularization parameter, ||·|| represents the Euclidean norm, ||·|| represents the Frobenius norm. F

[0034] As a preferred solution, the output results of the contrast learning model and the deep one-class support vector machine are used as thresholds, and the output results of the second test set are compared to complete the state evaluation of the radio navigation system, specifically including:

[0035] The first test set and the second test set are input into the contrast learning model to obtain feature representations, and the feature representations are input into the deep one-class support vector machine for state evaluation. For sample data in the first test set and the second test set, the health state index is defined as the distance from the feature nonlinear mapping to the hypersphere center, wherein the feature nonlinear mapping is output by the contrast learning model, and the hypersphere center is output by the deep one-class support vector machine.

[0036] The health state index of all sample data in the first test set is calculated and sorted.

[0037] The sorted health state index is used as a temporary threshold to calculate the Youden index, and the temporary threshold that maximizes the Youden index is used as the hypersphere radius.

[0038] The health state index of all sample data in the second test set is calculated, and if the health state index is greater than the hypersphere radius, the sample data is abnormal.

[0039] As a preferred solution, before the sliding window processing, it also includes a data cleaning process: for missing value data, linear interpolation is used for filling; for repeated value data, repeated data is removed according to timestamp information; for singular value data, it is directly deleted.

[0040] Compared with the prior art, the beneficial effects of the above technical solution are:

[0041] (1) By constructing a contrast learning model, different working states in the data can be captured. The data is divided according to the working state, which provides data support for the subsequent one-class support vector machine for state evaluation, and avoids the problem of hypersphere collapse.

[0042] ​(2) Using the feature representation extracted by the contrastive learning model as input, a deep one-class support vector machine model is constructed. This model aims to find a hyperplane to maximize the separation of normal data points and exclude abnormal points.

[0043] (3) Deep one-class support vector machine focuses on identifying abnormal points by maximizing the separation of normal data points, improving the accuracy of state assessment. This method allows for adaptation to different requirements and data characteristics by adjusting parameters and optimizing model structure, providing flexibility and scalability. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is a flowchart of the radio navigation system state assessment method according to an embodiment of the present application.

[0045] Figure 2 is a sliding window processing diagram according to an embodiment of the present application.

[0046] Figure 3 is a diagram of positive and negative sample construction according to an embodiment of the present application.

[0047] Figure 4 is a diagram of the structure of the contrastive learning model according to an embodiment of the present application.

[0048] Figure 5 is a diagram of the deep support vector machine according to an embodiment of the present application. DETAILED DESCRIPTION

[0049] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numbers represent the same or similar modules or modules with the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as limiting the present application. On the contrary, the embodiments of the present application include all changes, modifications and equivalents falling within the spirit and scope of the appended claims.

[0050] In view of the problems of traditional knowledge-based methods such as difficult abnormal space coverage and low detection accuracy, the embodiments of the present application propose a multi-dimensional data enhanced learning radio navigation system state assessment method, which learns the deep features of the detection data using contrastive learning, and constructs a state assessment model using deep one-class support vector machine, has the advantages of high detection accuracy and explainability, has important theoretical and engineering practical significance, and realizes precise quantitative assessment of the health state of airborne radio navigation system. Please refer to Figure 1, the specific scheme is as follows:

[0051] Step 1, obtaining radio navigation system function operation monitoring data, and selecting data related to system health;

[0052] In this embodiment, the acquired radio navigation system function operation monitoring data can be existing radio navigation system function operation monitoring data, or can be obtained by using the navigation system equipped aircraft to fly m times to obtain m times of radio navigation system function operation monitoring data.

[0053] In fact, the function operation monitoring data of the airborne radio navigation system describes the system operation state, but the internal detection data of the radio navigation system is as high as more than one hundred, including character type, discrete numerical type, continuous numerical type and other types of data, which describes the navigation system work task, performance, health state and other information. These detection data are not all capable of reflecting the state of the radio navigation system, and most of them are unrelated to the health state. If this part of data is added to the state evaluation, it will cause a large waste of resources. Therefore, in this embodiment, the number of uplink antenna received pulses, the number of downlink antenna received pulses, the number of uplink antenna received AGC responses, the number of downlink antenna received AGC responses, the number of downlink antenna received DSP responses, the time of being locked out, the distance and the azimuth are taken as the parameters for evaluating the state of the radio navigation system.

[0054] Since the collected data can contain missing values, repeated values and singular values, data cleaning processing is required. In this embodiment, for missing value data, linear interpolation is used for filling; for repeated value data, repeated data is removed according to the time stamp information; and for singular value data, it is directly deleted.

[0055] Step 2, sliding window processing is performed on the selected data to obtain sample data containing time information.

[0056] The health state of the radio navigation system has continuity in the time dimension, and it is very difficult to judge the health state of the radio navigation system only by single time point data. In this embodiment, the historical data of the detection time stamp segment is extracted as the detection time stamp feature data by sliding window processing. Through sliding window processing, the statistical information of the detection time stamp is contained in the feature data. Compared with single time point data, the sliding window data contains more information and can describe more health state information.

[0057] Please refer to FIG. 2, in this embodiment, a sliding window is constructed, which slides on the cleaned multi-dimensional health representation parameters to obtain window data one by one, which represents the state data of the radio navigation system in the segment time. The total sample data number, the total time length T, the window width s and the sliding window step d are related. On the selected data related to the system health, every d time points, a sliding window is taken, that is, every d time points, the sample number + 1, until the sliding window slides to the end, to obtain the total sample data.

[0058] Step 3, divide the sample data into a training set, a first test set, and a second test set; wherein, based on the data type, positive and negative samples are constructed for all sample data respectively.

[0059] In this embodiment, the obtained sample data is divided into a training set and a test set according to a ratio of 6:4, wherein the test set is further divided into a first test set and a second test set according to a ratio of 1:1.

[0060] Since the sample data needs to be used for contrastive learning subsequently, in this embodiment, a positive and negative sample construction method is further provided, as shown in FIG. 3, wherein the positive sample refers to a sample pair that is similar or relevant under a certain metric, and the negative sample refers to a sample pair that is not similar or relevant under a certain metric. In contrastive learning, the division of positive and negative samples is crucial, because it determines the direction and effect of model learning.

[0061] Positive sample construction: generate positive samples for sample data in a data enhancement manner; in this embodiment, the sample data obtained by sliding window is divided into discrete data and continuous data according to the data type, wherein the number of uplink received pulses, the number of downlink received pulses, the number of uplink received AGC responses, the number of downlink received AGC responses, the number of downlink received DSP responses, the time of being locked out, and the distance are discrete data; and the azimuth is continuous data.

[0062] For discrete data, add random noise with an amplitude of 0 or 1 to the original sample data to generate new sample data, which is a positive sample of the original sample data, that is:

[0063] wherein, X c is the discrete original sample data, is the new sample data, and N (0-1) is random noise with an amplitude of 0 or 1.

[0064] For continuous data, data enhancement is performed by adding Gaussian random noise to generate positive samples, that is:

[0065] wherein, X n is the continuous original sample data, is the new sample data, and N Gaussian is Gaussian random noise.

[0066] Negative sample construction: in actual application, different data samples have different working conditions, so the data correlation is not high. In this embodiment, different data samples directly serve as negative samples of each other.

[0067] Step 4, establish a contrastive learning model and train it using the training set.

[0068] The contrastive learning task is an unsupervised learning method aiming to learn effective data representation by comparing the similarities and differences between data samples. The contrastive learning model established in this embodiment is shown in FIG. 4. In the context of state evaluation of time series data, the goal of the contrastive learning task is to enable the model to distinguish data under different working conditions and avoid the super-sphere collapse of the support vector machine, thereby providing a foundation for subsequent state evaluation tasks.

[0069] For the training set The sample batch is represented as For one of the samples x i , the enhanced sample pair x i (1) ,x i (2) is generated by the above method of constructing positive and negative samples through data enhancement, and the samples x i are positive or negative samples to each other. The feature representation h i is obtained by the encoder f(·) to obtain the feature representation h i , and then the multi-layer perceptron g(·) is used to obtain the feature nonlinear mapping z i = g(f(x i )). During the training process, when the samples are positive samples to each other, the distance between the samples is reduced, and when the samples are negative samples to each other, the distance between the samples is increased.

[0070] If the enhanced sample pair is represented as The sample loss is:

[0071] wherein, are the feature nonlinear mappings of x i (1) , x i (2) , z j corresponds to x j , is a set of enhanced sample pairs that do not contain x i in the sample batch; represents contrastive learning with as the anchor point; represents contrastive learning with as the anchor point;

[0072] The contrastive loss function formula is represented as:

[0073] wherein, is a similarity function, τ is a temperature coefficient, B represents a set of all samples in a sample batch, is a set of samples in the sample batch except Corresponding positive / negative samples All other sample sets except the above.

[0074] The model is trained using a contrastive loss function. During training, the model adjusts its parameters to minimize the loss function, learning to distinguish between different operating conditions.

[0075] Step 5, establish and train deep one-class support vector machine.

[0076] In this embodiment, please refer to Figure 5, combine deep learning and one-class support vector machine to establish deep one-class support vector machine for state evaluation task, that is, identify and separate abnormal or outlier points without explicit labeled normal data. Deep one-class support vector machine combines the advantages of deep learning and one-class support vector machine. Specifically, it uses a deep learning model instead of a kernel function to automatically extract and learn complex features of the data, and then uses these features for state evaluation. At the same time, the optimal connection weight of the neural network and the minimum hypersphere surrounding the learned feature representation are calculated. When all training data are normal data, the training goal of deep one-class support vector machine is to find a minimum hypersphere so that all samples are within the sphere.

[0077] For sample data, the input is the feature representation h i =f(x i ) obtained by the trained contrastive learning model encoder f(·)

[0078] Hypersphere center:

[0079] where min denotes minimization, is a set of weights, n is the number of samples in the given training set c is the hypersphere center, h i =f(x i ) is the feature representation of the i-th sample data, φ(·;W) is the neural network, w l represents the weight of the l-th hidden layer, l=1,2,…,L, λ is the regularization parameter, ‖·‖ is the Euclidean norm, ‖·‖ F is the Frobenius norm, B represents the set of all samples in a sample batch, and the feature nonlinear mapping z i =g(f(x i )).

[0080] Step 6, using the first test set to compare the output results of the contrast learning model and the deep one-class support vector machine as the threshold value, and comparing them with the output results of the second test set, completing the state evaluation of the radio navigation system.

[0081] After the training of the contrast learning model and the deep one-class support vector machine is completed, the first test set and the second test set are used to complete the test evaluation. Specifically:

[0082] The first test set and the second test set are respectively input into the contrast learning model to obtain feature representation, and then the feature representation is input into the deep one-class support vector machine for state evaluation. For sample data in the first test set and the second test set, the health state index is defined as the distance of the nonlinear mapping of the feature to the center of the hypersphere: S(x) = ‖z-c‖ 2

[0083] Where S(x) is the health state index, z is the nonlinear mapping of the feature, which is output by the contrast learning model, and c is the center of the hypersphere, which is output by the deep one-class support vector machine.

[0084] The health state index is used to determine the classification threshold, and then the Youden index is calculated. The specific steps are as follows:

[0085] (1) Calculate the health state index of each sample data in the first test set and sort it {S1, S2... S i , S n};

[0086] (2) For each value in the sorted health state index list, use the value as a temporary threshold R temp =S i , calculate the TPR and TNR under the current threshold value:

[0087] Where TPR is the true class rate (sensitivity), i.e. the proportion of all actual abnormal samples {x1, x2... x i , x n} abnormal, TP+FN=n, correctly identified as abnormal; TNR is the true negative class rate (specificity), i.e. the proportion of all actual normal samples {x1, x2... x i , x m} normal, TN+FP=m, correctly identified as normal; TN is the number of true positives, i.e. the number of samples correctly predicted as normal by the model; FP is the number of false positives, i.e. the number of normal samples incorrectly predicted as abnormal by the model; FN is the number of false negatives, i.e. the number of abnormal samples incorrectly predicted as normal by the model; TP is the number of true positives, i.e. the number of samples correctly predicted as abnormal by the model.

[0088] (3) Calculate the Youden index J of the current threshold according to TPR and TNR, and the specific formula is as follows: J = TPR + TNR - 1

[0089] (4) Select the threshold that makes the Youden index maximum as the final classification threshold, and this threshold defines the radius of the hypersphere, that is, the dividing line between normal and abnormal samples.

[0090] That is, the radius of the hypersphere is: R = Max{J} R = R temp while Max{J}

[0091] The temporary threshold that makes the Youden index J maximum is the final classification threshold, that is, the radius of the hypersphere R.

[0092] Calculate the health status index S(x) = ‖z-c|| of all sample data in the second test set by the foregoing formula 2 If the health status index is greater than the radius of the hypersphere S(x) > R, it indicates that the sample data is abnormal, and if the health status index is less than or equal to the radius of the hypersphere S(x) ≤ R, it indicates that the sample data is normal.

[0093] In particular, according to the embodiments of the present application, the processes described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments of the present application include a computer program product comprising a computer program carried on a computer readable medium, and the computer program comprises program code for executing the method shown in the flowchart.

[0094] It should be noted that the computer-readable medium in the embodiments shown in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (Compact Disc Read-Only Memory, CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component. In the present application, the computer-readable signal medium can include a data signal carrying computer-readable program code in a baseband or as a part of a carrier wave. Such a propagated data signal can take various forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or component. The program code contained in the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wired, or the like, or any suitable combination of the above.

[0095] The flowcharts in the drawings show the possible implementation architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In the flowcharts or block diagrams, each block can represent a module, a program segment or a part of code, which contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different order from that shown in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and the combination of blocks in the block diagram or flowchart, can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0096] The units described in the embodiments of the present application can be implemented in the form of software, or can be implemented in the form of hardware, and the described units can also be arranged in a processor. In some cases, the names of the units do not constitute a limitation on the units themselves.

[0097] As another aspect, the present application also provides a computer program product or computer program, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to enable the computer device to perform the method for evaluating the state of a radio navigation system based on multi-dimensional data enhanced learning described in the above embodiments.

[0098] As another aspect, the present application also provides a computer readable medium, which can be included in the electronic device described in the above embodiments, or can exist separately without being assembled into the electronic device. The computer readable medium carries one or more programs, which, when executed by the electronic device, enable the electronic device to implement the method for evaluating the state of a radio navigation system based on multi-dimensional data enhanced learning described in the above embodiments.

[0099] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, such division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into several modules or units.

[0100] From the above description of the embodiments, those skilled in the art can easily understand that the example embodiments described herein can be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solutions according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, U disk, mobile hard disk, etc.) or network, and includes several instructions to enable a computing device (which can be a personal computer, server, touch terminal, or network device, etc.) to perform the methods according to the embodiments of the present application.

[0101] The above terms can be understood in the specific meaning in the present application by the person of ordinary skill in the art according to the specific circumstances; the drawings in the embodiments are used to clearly and completely describe the technical solutions in the embodiments of the present application, and obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0102] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and cannot be understood as limiting the present application, and the person of ordinary skill in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method of radio navigation system state assessment using multi-dimensional data reinforcement learning, characterized by, The method comprises the following steps: obtaining radio navigation system function operation monitoring data, and selecting data related to system health; performing sliding window processing on the selected data to obtain sample data containing time information; dividing the sample data into a training set, a first test set and a second test set; wherein, based on the data type, positive and negative samples are constructed for all sample data; establishing a contrast learning model and training the model using the training set; establishing and training a deep one-class support vector machine, the input of which is the feature representation extracted by the trained contrast learning model; using the output results of the contrast learning model and the deep one-class support vector machine as thresholds, and comparing them with the output results of the second test set to complete the radio navigation system state evaluation.

2. The method of claim 1, wherein, The selected data related to system health includes: the number of uplink received pulses, the number of downlink received pulses, the number of uplink received AGC responses, the number of downlink received AGC responses, the number of downlink received DSP responses, the time of being locked out, the distance and the azimuth.

3. The method of claim 1, wherein, The specific method for obtaining sample data containing time information comprises: Taking a sliding window every d time points on the selected data related to system health, that is, sample number +1 every d time points until the sliding window slides to the end to obtain the total sample data.

4. The method of claim 1, wherein, The training set and the test set are divided according to the ratio of 6:4, and the test set is further divided into the first test set and the second test set according to the ratio of 1:

1.

5. The method of claim 2, wherein, The specific method for constructing positive and negative samples comprises: Positive sample construction: generating positive samples for sample data in the manner of data enhancement; Negative sample construction: different data samples are negative samples of each other.

6. The method of claim 5, wherein, The specific method for constructing positive samples comprises: Dividing the sample data obtained by the sliding window into discrete data and continuous data according to the data type, wherein the number of uplink received pulses, the number of downlink received pulses, the number of uplink received AGC responses, the number of downlink received AGC responses, the number of downlink received DSP responses, the time of being locked out and the distance are discrete data; and the azimuth is continuous data; For discrete data, random noise with an amplitude of 0 or 1 is added to the original sample data to generate new sample data, which are positive samples of each other; For continuous data, data enhancement is performed by adding Gaussian random noise to generate positive samples.

7. The method of claim 1, wherein, The training method of the contrast learning model comprises: The contrast learning model is established. For any sample data in the training set, an enhanced sample pair of the sample data is generated by constructing positive and negative samples, the enhanced sample pair is input into an encoder to obtain a feature representation of the sample data, and then a multi-layer perceptron is used to obtain a nonlinear mapping of the feature. The model is trained by using a contrast loss function. The contrast loss function is For: wherein For a sample batch, x i For the i-th sample data, B represents a collection of all sample data in a sample batch, to enhance the batch representation of the samples, x i (1) x i (2) are the enhanced sample pairs of xi, τ is the temperature coefficient, For the sample batch, except for the samples with corresponding positive / negative samples all other sample sets except for the one with the highest score, denotes sample loss for the anchor point to conduct contrastive learning; denotes The sample loss of contrastive learning as an anchor point is specifically: wherein for the similarity calculation function for calculating the similarity of u and v, exp is an exponential function, respectively i (1) , x i (2) characteristic nonlinear mapping, z j corresponding to x j , 8. The method of claim 7, wherein, The specific process of establishing and training the deep one-class support vector machine comprises: A deep one-class support vector machine is established by combining deep learning and one-class support vector machine, which is used to identify and separate abnormal or outlier points; the input of the deep one-class support vector machine is the feature representation obtained by the encoder of the trained contrast learning model; The training of the deep one-class support vector machine is completed by using an optimization objective function, wherein the optimization objective function is defined as: wherein min denotes minimization, for the weight set, c is the hypersphere center, h i = f(x i ) is the feature representation of the i-th sample data, n is the given training set the number of samples, φ( ; W) is a neural network, w l denotes the weight of the l-th hidden layer, l = 1, 2, …, L, λ is a regularization parameter, ||·|| denotes the Euclidean norm, ||·||F F is the Frobenius norm.

9. The method of claim 1, wherein, The specific process of using the output results of the contrast learning model and the deep one-class support vector machine as thresholds, and comparing them with the output results of the second test set to complete the radio navigation system state evaluation comprises: inputting the first test set and the second test set into the contrastive learning model respectively to obtain feature representations, and inputting the feature representations into the deep one-class support vector machine for state evaluation, for sample data in the first test set and the second test set, defining a health state index as a distance from a feature nonlinear mapping to a hypersphere center, wherein the feature nonlinear mapping is output by the contrastive learning model, and the hypersphere center is output by the deep one-class support vector machine; calculating the health state indexes of all sample data in the first test set and sorting the health state indexes; taking the sorted health state indexes as temporary threshold values in sequence to calculate Youden indexes, and taking the temporary threshold value that makes the Youden index maximum as a hypersphere radius; calculating the health state indexes of all sample data in the second test set, and if the health state index is greater than the hypersphere radius, it indicates that the sample data is abnormal.

10. The method of claim 1, wherein, Before the sliding window processing, a data cleaning process is further included: for missing value data, linear interpolation is used for filling; for repeated value data, repeated data is removed according to timestamp information; for singular value data, direct deletion is performed.

Images