A multi-source anti-interference fusion navigation method and system for a body space-time intelligent agent

Abstract

The application discloses a body space intelligent agent multi-source anti-interference fusion navigation method and system, the method comprises the following steps: according to the key features screened from the historical flight data by using the random forest algorithm in advance, real-time multi-modal perception information of the aircraft in the current scene is obtained, and standardized processing is performed to obtain a body state feature vector; the body state feature vector is input into a pre-trained navigation cognitive model to obtain a predicted navigation performance, a navigation scene risk potential field is constructed, and a navigation risk potential value is calculated based on the predicted navigation performance and the navigation scene risk potential field; according to the navigation risk potential value and a preset decision style, an elastic navigation strategy is generated and executed, the elastic navigation strategy comprises switching between a satellite navigation dominant mode, a ground-based ranging dominant mode and a combined navigation mode. The application improves the continuity of aviation navigation in a complex scene, thereby adapting to a variable flight environment.

Description

Technical Field

[0001] This application belongs to the field of aviation navigation technology, and in particular relates to a multi-source anti-interference fusion navigation method and system for an embodied spatiotemporal intelligent agent. Background Technology

[0002] Aviation flight is susceptible to the combined effects of complex weather, electromagnetic interference, terrain and building obstructions, and dynamic changes in the aircraft's own flight attitude and maneuvering. This directly leads to signal quality degradation in satellite navigation and ground-based ranging navigation, deterioration of spatial geometry, and anomalies in observation data, causing navigation service interruptions, operational instability, or even complete failure. Current aviation navigation systems mostly employ fixed empirical rules, static weighted fusion of multiple navigation systems, or hard threshold switching after navigation system failure. These methods fail to characterize the intrinsic relationship between environmental factors, aircraft motion states, and navigation performance. They cannot adapt to multi-scenario and multi-interference environments to achieve adaptive adjustments to navigation strategies. Furthermore, they do not consider the impact of sensor perception errors and data uncertainties on navigation reliability assessment and decision boundaries, and lack the ability to predict navigation performance and make flexible decisions for long-term environmental adaptation. Aircraft possess the inherent attribute of spatiotemporal interaction of embodied intelligent agents, which supports their perception and decision-making closed loop. However, existing embodied navigation systems mostly rely on a single satellite navigation system, which is susceptible to obstruction and interference in mountainous areas, urban canyons, and complex electromagnetic environments. This can lead to spatiotemporal instability, disrupt the perception and decision-making closed loop of the intelligent agent, and cause flight planning and execution results to deviate from the actual environment, resulting in a significant reduction in mission performance. There is an urgent need for a navigation solution that can achieve multi-scenario situational awareness, dynamic assessment of the reliability of multiple navigation systems, and adaptive strategy switching. Summary of the Invention

[0003] This application provides a spatiotemporal intelligent agent multi-source anti-interference fusion navigation method and system. In flight scenarios with complex weather, electromagnetic interference, terrain obstruction and dynamic changes in aircraft operating status, it can not only extract key features of multimodal perception information, accurately predict navigation performance and quantify failure risk under perception uncertainty, but also adapt navigation modes to match different decision styles and complete model closed-loop self-update.

[0004] This application discloses a multi-source anti-interference fusion navigation method for embodied spatiotemporal intelligent agents, including: Based on key features pre-selected from historical flight data using the random forest algorithm, real-time multimodal perception information of the aircraft in the current scenario is obtained and standardized to obtain an embodied state feature vector. The embodied state feature vector is input into a pre-trained navigation cognition model to obtain the predicted navigation performance, a navigation scenario risk potential energy field is constructed, and the navigation risk potential energy value is calculated based on the predicted navigation performance and the navigation scenario risk potential energy field. Based on the navigation risk potential value and the preset decision-making style, a flexible navigation strategy is generated and executed. The flexible navigation strategy includes switching between satellite navigation-dominated mode, ground-based ranging-dominated mode and combined navigation mode.

[0005] Optionally, the key features include carrier-to-noise ratio, interference-to-signal ratio, multipath delay, accuracy factor, and aircraft motion state parameters.

[0006] Optionally, the navigation cognitive model is a nonlinear mapping model trained by a multilayer perceptron, with the key features as input and navigation positioning error and availability as output.

[0007] Optionally, the training steps of the navigation cognitive model include: Historical flight data is acquired, and the average information gain of multi-dimensional scene features at the decision tree splitting node is calculated using the random forest algorithm. Key features are then selected based on the average information gain. A multilayer perceptron model is constructed, with the selected key features as input and navigation positioning error and availability as output. The model is trained using the backpropagation algorithm to obtain the navigation cognitive model.

[0008] Optionally, constructing the navigation scenario risk potential energy field includes: Set the physical failure boundary values ​​for each navigation system; Define a risk potential energy function such that the value of the risk potential energy function increases exponentially as the navigation system reliability state variable is within a preset threshold as the difference between it and the physical failure boundary value is within a preset threshold. The perception error is modeled using a Gaussian distribution function, and the perception uncertainty is superimposed on the risk potential energy function to obtain the risk potential energy field of the navigation scenario.

[0009] Optionally, calculating the navigation risk potential energy value based on the predicted navigation performance and the navigation scenario risk potential energy field includes: The predicted navigation performance is mapped to the navigation system reliability state variable; The navigation risk potential value is calculated based on the navigation system reliability state variable and the risk potential function superimposed with perceived uncertainty. The navigation risk potential value represents the probability density of navigation system failure.

[0010] Optionally, generating and executing a flexible navigation strategy based on the navigation risk potential value and a preset decision-making style includes: When the preset decision-making style is low-risk tolerance, when the navigation risk potential value of any navigation system reaches the warning threshold, the corresponding navigation mode is restricted or turned off, and other available navigation modes are switched. When the preset decision-making style is high-risk tolerance, the navigation mode is switched when the navigation risk potential value exceeds the failure threshold.

[0011] Optionally, generating and executing the elastic navigation strategy includes: The measurement matrix and measurement noise covariance matrix of the combined navigation filter are reconstructed based on the generated elastic navigation strategy. When switching to ground-based ranging-dominated mode, increase the covariance of satellite navigation measurement noise or remove satellite navigation measurement dimensions; when multipath delay of a specific ground-based ranging channel is detected, perform information detection and isolation on that channel.

[0012] Optionally, the aforementioned embodied spatiotemporal intelligent agent multi-source anti-interference fusion navigation method further includes: During flight or after a mission, post-processed or extrapolated data are used as reference values ​​to calculate navigation residuals. The navigation residuals are fed back to the navigation cognitive model, the measurement noise covariance matrix of the combined navigation filter is adjusted online adaptively, the navigation residual information is stored, and the model weights are updated offline after the aircraft finishes its flight mission. Incremental feature importance assessment is initiated periodically. The average information gain of each feature is recalculated based on the cumulative flight data, the feature weights are dynamically adjusted, and the elimination of low-contribution features and the inclusion of new related features are completed, thereby achieving dynamic adaptive updating of the feature space.

[0013] This application also discloses an embodied spatiotemporal intelligent agent multi-source anti-interference fusion navigation system, including: The scene perception module is used to obtain the aircraft's real-time multimodal perception information in the current scene based on key features pre-selected from historical flight data using the random forest algorithm, and to perform standardization processing to obtain the embodied state feature vector. The evaluation calculation module is used to input the embodied state feature vector into a pre-trained navigation cognitive model to obtain the predicted navigation performance, construct the navigation scenario risk potential energy field, and calculate the navigation risk potential energy value based on the predicted navigation performance and the navigation scenario risk potential energy field. The flexible decision-making module is used to generate and execute flexible navigation strategies based on navigation risk potential values ​​and preset decision-making styles. The flexible navigation strategies include switching between satellite navigation-dominated mode, ground-based ranging-dominated mode, and combined navigation mode.

[0014] As can be seen from the above technical solution, this application first relies on random forest to screen parameters from massive scene features, extracts and standardizes the real-time multimodal perception information of the aircraft to form a regular embodied state feature vector, providing a unified input basis for subsequent model calculations; then, it constructs a nonlinear mapping relationship between features and navigation performance through a multilayer perceptron, realizing the early prediction of navigation operation status and making up for the shortcomings of traditional methods in lacking predictive capabilities; at the same time, it introduces physical failure boundaries to build a risk potential energy field, combines Gaussian modeling to incorporate perception measurement uncertainty, and uses continuous potential energy values ​​to quantify the failure trend of the navigation system, avoiding strategy jumps caused by hard threshold judgment. Subsequently, based on the quantified risk potential value, high and low risk decision-making styles are matched, and the system can flexibly switch between satellite navigation, ground-based ranging navigation, and combined navigation modes as needed. The navigation filter matrix is ​​reconstructed simultaneously, and abnormal ranging channels are isolated to ensure that the navigation solution is not affected by abnormal data. Finally, the model weights are updated iteratively using the feedback of actual navigation residuals. Each link is linked and coordinated in sequence according to the logic of perception regularization, performance prediction, risk quantification, style decision-making, and closed-loop update. The entire process is free from the constraints of fixed experience rules and can autonomously adjust the navigation strategy according to changes in the flight environment, continuously adapt to complex and ever-changing aviation operation scenarios, and steadily improve the continuity and reliability of navigation operation. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is an overall flowchart of a multi-source anti-interference fusion navigation method for an embodied spatiotemporal intelligent agent according to an embodiment of this application; Figure 2 This is a flowchart of the training steps of the navigation cognitive model in an embodiment of a spatiotemporal intelligent agent multi-source anti-interference fusion navigation method according to an embodiment of this application; Figure 3 This is a flowchart illustrating the construction of a navigation scenario risk potential energy field in a multi-source anti-interference fusion navigation method for an embodied spatiotemporal intelligent agent according to an embodiment of this application. Figure 4 This is a flowchart illustrating the multi-source anti-interference fusion navigation method for an embodied spatiotemporal intelligent agent in this application embodiment, which calculates the navigation risk potential energy value based on the predicted navigation performance and the navigation scenario risk potential energy field. Figure 5 This is a flowchart illustrating the generation and execution of an elastic navigation strategy based on the navigation risk potential value and a preset decision style in an embodiment of the present application's spatiotemporal intelligent agent multi-source anti-interference fusion navigation method. Figure 6 This is a flowchart illustrating the execution of an elastic navigation strategy in an embodiment of a spatiotemporal intelligent agent multi-source anti-interference fusion navigation method according to an embodiment of this application; Figure 7 This is a closed-loop update flowchart of a multi-source anti-interference fusion navigation method for an embodied spatiotemporal intelligent agent according to an embodiment of this application; Figure 8 This is a schematic diagram of the structure of an embodied spatiotemporal intelligent agent multi-source anti-interference fusion navigation system according to an embodiment of this application. Detailed Implementation

[0017] In the following description, specific details such as particular system architectures and technologies are set forth for illustrative purposes and not limiting, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without such specific details.

[0018] This application discloses, on the one hand, a multi-source anti-interference fusion navigation method for embodied spatiotemporal intelligent agents, such as... Figure 1 As shown, it includes: S100: Based on key features pre-selected from historical flight data using the random forest algorithm, obtain real-time multimodal perception information of the aircraft in the current scenario, and perform standardization processing to obtain the embodied state feature vector.

[0019] S200: Input the embodied state feature vector into a pre-trained navigation cognitive model to obtain the predicted navigation performance, construct a navigation scenario risk potential energy field, and calculate the navigation risk potential energy value based on the predicted navigation performance and the navigation scenario risk potential energy field.

[0020] S300: Generate and execute a flexible navigation strategy based on the navigation risk potential value and the preset decision style. The flexible navigation strategy includes switching between satellite navigation-dominated mode, ground-based ranging-dominated mode and combined navigation mode.

[0021] In optional implementations, the key features include carrier-to-noise ratio, interference-to-signal ratio, multipath delay, accuracy factor, and aircraft motion state parameters.

[0022] In this embodiment of the application, the airspace environment during flight is complex and changeable. Fluctuations in weather conditions, electromagnetic radiation interference, and terrain and building obstructions will all have varying degrees of impact on the navigation signal transmission link. Various scenario parameters will indirectly change the service performance of satellite navigation and ground-based ranging navigation. Selecting characteristic parameters that are suitable for aviation navigation status assessment can fully characterize the relationship between flight scenarios and navigation operation status.

[0023] The carrier-to-noise ratio (CNR) is used to characterize the ratio between the effective navigation signal power captured by the navigation receiving device and the background ambient noise power. Terrain obstruction in the airspace weakens the strength of the direct navigation signal, while various natural electromagnetic noises and artificial radiated noises increase the background noise level. Both of these changes will cause the CNR value to change. The fluctuation of the CNR can intuitively reflect the operating status of the navigation signal receiving link.

[0024] The interference-to-signal ratio (ISR) is used to characterize the power ratio between the effective navigation signal and the airspace interference signal. Adjacent-channel electromagnetic radiation, suppression interference signals, and scattering interference signals present in the airspace of aviation operations will all change the value level of the ISR. A continuously decreasing ISR will cause distortion and offset in navigation observation data, which cannot truly reflect the spatial position information of the aircraft. It can serve as an important reference for identifying the navigation operation status in complex electromagnetic environments.

[0025] Multipath delay refers to the difference in propagation time between the reflected propagation signal formed after the navigation signal is reflected by mountainous terrain and urban high-rise buildings and the straight-line propagation signal. As the number of signal reflection propagation paths increases in mountainous airspace and urban canyon airspace, multipath delay will increase significantly. The additional observation bias introduced will affect the operational performance of satellite navigation positioning calculation and ground-based ranging navigation distance calculation.

[0026] The accuracy factor includes two commonly used parameters: position accuracy factor and horizontal accuracy factor. The value of the parameter is determined by the spatial geometric distribution configuration of available navigation satellites in orbit. When terrain obstruction reduces the number of available navigation satellites and deteriorates the spatial arrangement configuration of satellites, the accuracy factor value will show an upward trend, which can reflect the degree of change in the navigation spatial geometric configuration and indirectly reflect the fluctuation range of positioning solution.

[0027] Aircraft motion state parameters encompass multiple parameters, including flight rate of travel, pitch angle, roll angle, yaw angle, and attitude maneuver rate of change. When an aircraft is in a highly dynamic maneuvering flight condition, its attitude and flight state change rapidly, causing changes in the navigation antenna signal reception angle and signal propagation path, indirectly affecting the quality of navigation signal reception and the stability of observation data. Including these parameters in the feature selection scope comprehensively covers the influence of the external environment and the flight platform's own operating conditions on navigation operations, providing comprehensive parameter support for subsequent feature modeling and performance prediction. Aircraft inherently possess the characteristic of continuous interaction between physical entities and the real-world spatiotemporal environment. In typical airspace conditions such as urban canyons, mountainous areas, and complex electromagnetic interference, navigation signals are highly susceptible to multipath effects and electromagnetic suppression interference. Including these feature parameters aligns with the aircraft's inherent spatiotemporal interactive operating characteristics and adapts to the needs of multi-scenario aviation navigation state assessment.

[0028] The standardization process employs the Z-score standardization algorithm, which independently processes each dimension of key features in the real-time multimodal sensing information. The specific processing formula is as follows:

[0029] in Real-time acquisition of raw values ​​for single-dimensional features. This is the statistical mean of this dimension feature in historical flight training data. The standard deviation of this dimension feature in historical flight training data is given. After processing, each dimension feature satisfies a standard normal distribution with a mean of 0 and a variance of 1, eliminating the differences in dimensions and numerical magnitudes between different features, and obtaining a dimensionally unified embodied state feature vector, which can be directly used as input data for the navigation cognitive model.

[0030] For five key features—carrier-to-noise ratio, interference-to-signal ratio, multipath delay, accuracy factor, and aircraft motion state parameters—the mean and standard deviation parameters of the corresponding dimensions are pre-stored, and the real-time standardization calculation is completed by directly calling them during airborne operation.

[0031] In an optional implementation, the navigation cognitive model is a nonlinear mapping model trained by a multilayer perceptron, with the key features as input and navigation positioning error and availability as output.

[0032] It should be noted that the multilayer perceptron has a network architecture with multiple interconnected neurons, which can establish complex correlation mapping relationships between the input parameter set and the output parameter set. Unlike the limitations of traditional linear fitting methods, which can only adapt to simple linear change laws, it can adapt to the nonlinear change characteristics brought about by the coupling of multiple feature parameters in aviation flight scenarios.

[0033] The input port of the multilayer perceptron only accepts various feature parameters that have been filtered and processed, and no longer accepts irrelevant scene parameters with high redundancy and no significant impact on the navigation operation status. This simplifies the network input architecture and avoids the interference of redundant parameters on the model fitting and mapping effect. The output port of the multilayer perceptron synchronously corresponds to the navigation positioning error representation parameter and the navigation service availability representation parameter. The positioning error can reflect the degree of deviation between the spatial position calculated by navigation and the actual spatial position of the aircraft. The availability can reflect the ability of the corresponding navigation system to continuously output effective navigation services in the current flight airspace scenario.

[0034] The multilayer perceptron, which completes iterative training and convergence based on massive historical flight sample data, can autonomously learn the inherent correlation between changes in feature parameters and navigation performance in various scenarios. It does not need to rely on manually setting fixed corresponding rules to adapt to various flight conditions. It can adapt to complex and ever-changing airspace scenarios that cannot be exhausted in advance during aviation flights, thus addressing the application shortcomings of existing fixed experience rules for aviation navigation that cannot cover all scenario conditions.

[0035] After the multilayer perceptron completes offline iterative training, the connection weights and bias parameters of the neurons inside the network remain fixed. After being deployed inside the aircraft's onboard computing equipment, it only performs forward inference operations during flight. The overall computational logic is simple and clear, adapting to the real-time operation conditions of the onboard computing equipment. The navigation performance-related representation data output by the model can be smoothly updated following the real-time changes of the input key feature parameters, without any sudden changes in values.

[0036] The nonlinear mapping model built upon the multilayer perceptron serves as an intermediate carrier connecting front-end scene feature acquisition and processing with back-end navigation risk assessment calculations. It enables the automated transformation from multi-dimensional scene feature data to navigation operation performance status, aligning with the closed-loop operation logic of perception, understanding, and planning actions under the aircraft's embodied spatiotemporal interaction architecture. This provides reliable predictive data support for the subsequent quantitative assessment of navigation failure risks. Furthermore, this model architecture can adapt to the synchronous performance prediction requirements of both satellite navigation systems and ground-based ranging navigation systems, unifying the performance assessment modeling standards for the two types of navigation systems.

[0037] The embodied state feature vector is a standardized set of features that characterizes the current spatiotemporal interaction state of an aircraft. It covers two dimensions: external navigation signal environment parameters and aircraft motion state parameters. Unlike traditional feature vectors that only contain navigation signal parameters, it can fully reflect the coupling effect between the aircraft's own operating state and the external environment, providing a full-dimensional input basis for subsequent navigation performance prediction.

[0038] In alternative implementations, such as Figure 2 As shown, the training steps of the navigation cognitive model include: S110: Obtain historical flight data, use the random forest algorithm to calculate the average information gain of multi-dimensional scene features at the decision tree split nodes, and select key features based on the average information gain.

[0039] S120: Construct a multilayer perceptron model, using the selected key features as input and navigation positioning error and availability as output, and train it through the backpropagation algorithm to obtain the navigation cognitive model.

[0040] For example, historical flight data integrates long-term airborne data collected from different airspace routes, weather conditions, electromagnetic interference environments, terrain obstruction scenarios, and maneuvering flight states. The data dimensions cover various potential scenario feature parameters and the actual navigation operation performance indicators of the aircraft at the corresponding time. The sufficient sample data volume can meet the data requirements for iterative training and pattern learning of machine learning models. At the same time, the samples cover multiple typical aviation flight conditions, which can improve the adaptation performance of the model in unknown airspace scenarios after training.

[0041] Random forests consist of multiple independent decision trees forming an overall computational architecture. They can statistically analyze the average information gain of various multi-dimensional scene feature parameters during the splitting operation of decision tree nodes. This quantitative value reflects the degree of influence of each scene feature parameter on navigation performance changes. The higher the average information gain, the more prominent the influence of the corresponding scene feature parameter on the changes in navigation operation status. This quantitative statistical result serves as the basis for selecting key scene feature parameters. The entire selection process is automatically completed based on the statistical patterns of massive flight data, without relying on subjective judgments based on human experience, thus avoiding the one-sidedness and limitations of manual selection.

[0042] After screening, feature parameters whose mean information gain meets the preset evaluation criteria are retained, while redundant feature parameters with low mean information gain and almost no impact on navigation performance are removed. This simplifies the dimensionality of subsequent model input data, reduces the airborne computing resource overhead during neural network operations, and adapts to the current deployment situation where aircraft onboard equipment has limited computing power. The multilayer perceptron is built according to a standard network architecture of input layer, multiple hidden layers, and output layer. The hidden layer is responsible for abstracting and extracting input feature information and performing associated feature mapping operations. The input layer connects to the screened key feature parameter dataset, and the output layer corresponds to two types of output targets: navigation and positioning error parameters and navigation service availability parameters.

[0043] The backpropagation algorithm uses the deviation between the inference output of the multilayer perceptron model and the historical sample real labeled data as the adjustment basis. It corrects the connection weight parameters between neurons layer by layer from the network output layer to the input layer. Combined with an adaptive optimization strategy, it completes multiple rounds of iterative training operations until the model prediction deviation converges to a stable range, thus completing the training and solidification of the navigation cognitive model.

[0044] Once trained, the navigation cognitive model can adapt to the feature mapping requirements of various flight scenarios, such as plateau mountains, ocean airspace, and urban canyons, and has good generalization and adaptation capabilities. At the same time, the model training process can introduce multi-source environmental simulation data to supplement the sample size, make up for the lack of measured samples in special extreme flight conditions, and further improve the model's operational adaptability in complex and extreme scenarios. The trained model can be batch-deployed to various airborne platforms of the same type of aircraft without having to carry out the training process for each aircraft individually, and has good engineering versatility and convenient deployment.

[0045] In alternative implementations, such as Figure 3 As shown, the construction of the navigation scenario risk potential energy field includes: S210: Set the physical failure boundary values ​​for each navigation system.

[0046] S220: Define a risk potential function such that the value of the risk potential function increases exponentially as the navigation system reliability state variable is within a preset threshold as the difference between it and the physical failure boundary value is within a preset threshold.

[0047] S230: The perception error is modeled using a Gaussian distribution function, and the perception uncertainty is superimposed on the risk potential energy function to obtain the risk potential energy field of the navigation scenario.

[0048] Specifically, for satellite navigation systems and ground-based ranging navigation systems, corresponding physical failure boundary values ​​are set based on the technical specifications of navigation equipment manufacturers, general guidelines for aviation navigation operations, and critical operating condition data obtained from long-term flight tests. These boundary values ​​represent the extreme operating conditions under which the corresponding navigation system can maintain effective navigation service output. When navigation operating parameters exceed these boundary values, the navigation service output effect cannot meet the actual requirements of aviation flight. The navigation scenario risk potential energy field is a one-dimensional scalar field constructed separately for each type of navigation system, with the navigation system reliability state variable as the independent variable and the risk potential energy value as the dependent variable. It is used to continuously quantify the degree of risk of the navigation system's operating state approaching the physical failure boundary. Satellite navigation systems and ground-based ranging navigation systems independently construct their corresponding potential energy fields, without interference, and each outputs its own risk potential energy value.

[0049] In practical applications, a position accuracy attenuation factor of five can be set as the risk judgment point for satellite navigation failure boundary. When the position accuracy attenuation factor is greater than five or the horizontal accuracy attenuation factor is in the range of three to eight, the satellite navigation operation status is judged to be deteriorating, and the ground-based ranging and navigation-dominated operation mode can be switched. At the same time, a ground-based ranging and navigation pulse delay of one microsecond is set as the multipath effect tolerance boundary. When the pulse delay is detected to exceed this value range, the corresponding ground-based ranging and navigation observation source is eliminated. Furthermore, a spatial interference power of -110 decibels and milliwatts can be set as the satellite navigation rejection boundary, and -40 decibels and milliwatts as the failure boundary of the entire navigation system, so as to define the critical operating range of various navigation conditions with clear physical parameters.

[0050] The risk potential function uses the navigation system's reliability state variable as the independent variable and the risk potential energy value as the dependent variable to construct a corresponding operational relationship. As the operational value of the reliability state variable gradually approaches the preset physical failure boundary value, the output value of the risk potential function shows a steady upward trend, and the rate of increase gradually accelerates within the preset value range near the boundary. This can delicately reflect the degree of risk evolution as the navigation operation state gradually approaches the failure condition. The risk potential function is constructed using a continuous function operation form, without using a piecewise discrete value setting method. It can uninterruptedly represent the entire process of the navigation system's state change from a safe and stable operating condition to a critical failure condition, avoiding the operational state jump problem that is prone to occur in discrete fixed threshold judgment methods, and adapting to the usage requirements of smooth switching of aviation navigation operation modes.

[0051] The Gaussian distribution function is used to characterize the random fluctuation characteristics of measurement data from airborne sensing and acquisition equipment. Airborne sensors are affected by the inherent operating noise of hardware devices and instantaneous electromagnetic disturbances in the external airspace. Multiple sampling values ​​of the same physical parameter will show regular fluctuation distribution characteristics around the true value. Using the Gaussian distribution function can fit the distribution variation law of this type of measurement error.

[0052] By incorporating the uncertainty generated during the sensing and measurement process as a dispersion characteristic into the basic risk potential energy function, the final navigation scenario risk potential energy field no longer relies on the ideal, error-free measurement assumptions. Instead, it incorporates the numerical fluctuations and deviations that objectively exist in the actual airborne sensing and acquisition process. This allows the navigation risk state represented by the potential energy field to closely match the data acquisition characteristics during actual aircraft flight. Satellite navigation systems and ground-based ranging navigation systems adopt a unified potential energy field construction logic architecture, differing only in the configuration of boundary parameters and Gaussian error distribution variance parameters. This simplifies the overall modeling architecture, facilitating modular development, deployment, and maintenance of airborne programs. The navigation scenario risk potential energy field formed after superimposing perceived uncertainty can output smooth and continuous risk assessment results, avoiding the numerical jitter defects present in traditional single-point numerical comparison and judgment methods.

[0053] In a specific example, the logic for constructing the potential energy field model is as follows: Let... For the reliability state variable of the current navigation system, the failure boundary value is defined as follows: Constructing the risk potential energy field function , characterization Approaching The level of risk at that time. A Gaussian function is used. Perception error Modeling is performed to construct a risk potential energy field with superimposed uncertainties. This allows for the acquisition of smooth, continuous risk assessment results.

[0054] To address the issue of random errors in airborne sensor measurements, this invention abandons simple numerical comparisons and introduces a navigation scenario evaluation potential energy field model. The physical failure boundaries of various navigation systems (such as the interference power value of the Global Navigation Satellite System and the multipath delay value of the rangefinder) are set as high potential energy barriers. A risk potential energy function is constructed, whose value increases exponentially as navigation performance approaches the failure boundary.

[0055] The sensor measurement error is modeled using a Gaussian distribution function, and the uncertainty of the sensed data is mapped to the dispersion of the potential energy field.

[0056]

[0057] in, For the index subscript of the navigation system, These correspond to two different navigation systems and are used to distinguish the independent calculation parameters for different navigation systems. Let be the credibility state variable of the i-th type of navigation system, representing the coordinate value of the current operating state of the corresponding navigation system in the potential energy field; σ is the baseline value of the confidence state variable corresponding to the physical failure boundary value; σ is the velocity parameter characterizing the change in the potential energy field strength value near the physical failure boundary value. For the first The calculation results of the basic field strength corresponding to the navigation class.

[0058] The probability distribution function of the navigation risk potential value t is as follows:

[0059] in For navigation risk potential energy values, a random variable represents the overall set of risk values; The specific observed value of the navigation risk potential energy is a random variable. A specific value; Let be the conditional probability density function, representing the condition on the th... Navigation confidence position is Under the premise that the navigation risk potential energy value is obtained The probability density; The mean position parameter of the probability distribution corresponds to the basic field strength value of the potential energy field of the i-th type of navigation system calculated above. For the scaling factor, satisfying It is used to adjust the overall dispersion of the probability distribution and control the width of the distribution curve; For morphological coefficients, satisfying Used to adjust the credibility position The strength of the influence on the degree of dispersion of the distribution, and the pattern of how the distribution shape changes with the degree of confidence; The variance term of the probability distribution, determined by the scaling coefficient, confidence position, and morphology coefficient, characterizes the range of fluctuation in the navigation risk potential energy value. The acquired embodied state feature vector is input into the trained multilayer perceptron model, and combined with the potential energy field function, the navigation risk potential energy value of each navigation system at the current moment is calculated.

[0060]

[0061] in, This represents the navigation risk potential energy value corresponding to the low-risk tolerance style, characterizing the cumulative risk level of navigation failure under the low-risk tolerance style; The navigation risk potential value corresponding to the high-risk tolerance style is obtained by interval integration, which characterizes the cumulative risk of navigation failure under the high-risk tolerance style; The risk threshold under the low-risk tolerance style is the lower limit of the risk score of the low-risk tolerance style. This represents the lower risk threshold under a high-risk tolerance style, and is the starting boundary of the integration interval for a high-risk tolerance style. This represents the upper risk threshold under a high-risk tolerance style, and is the boundary of the integration interval for that style. The navigation risk potential value smoothly characterizes the probability density of the navigation system's impending failure, avoiding numerical fluctuations caused by hard threshold determination.

[0062] In alternative implementations, such as Figure 4 As shown, the calculation of the navigation risk potential energy value based on the predicted navigation performance and the navigation scenario risk potential energy field includes: S310: Map the predicted navigation performance to the navigation system reliability state variable.

[0063] S320: Calculate the navigation risk potential value based on the navigation system reliability state variable and the risk potential function superimposed with perceived uncertainty. The navigation risk potential value represents the probability density of navigation system failure.

[0064] In addition, the navigation performance parameters output by the navigation cognitive model include quantitative representations such as the magnitude of changes in positioning deviation and the availability of navigation services. According to the preset unified parameter mapping rules, multiple types of navigation performance parameters are transformed into navigation system reliability state variables with unified dimensional standards. This completes the unified format conversion of navigation performance data into input parameters for the risk assessment stage, unifies the numerical standards and calculation methods for subsequent potential energy field calculations, and avoids calculation deviations caused by differences in the dimensionality of different types of parameters.

[0065] The risk potential energy function, superimposed with perceived uncertainty, integrates two types of influencing factors: the boundary constraints of navigation physical failure and the random fluctuations in sensor measurements. By substituting the standardized navigation system reliability state variable into the function calculation process, the navigation risk potential energy value corresponding to a single navigation system at the corresponding flight time can be obtained. The navigation risk potential energy value is presented in continuous numerical form, and its magnitude reflects the probability of navigation system service failure. It can quantitatively characterize the operational safety margin of the navigation system in the current flight airspace scenario, replacing the traditional method of coarsely judging navigation operation status with a single fixed threshold, and refining the granularity of navigation risk level differentiation. This type of value can characterize the probability density distribution characteristics corresponding to the failure development trend of the navigation system, rather than directly giving a definite failure judgment conclusion. Instead, it quantifies the strength of the failure development trend, which is in line with the actual operational needs of aviation navigation to predict operational risks in advance and smoothly transition navigation working modes, and will not trigger unnecessary mode switching operations due to instantaneous parameter fluctuations.

[0066] In actual calculations, risk potential energy values ​​are calculated independently for both satellite navigation and ground-based ranging navigation systems. The risk assessment results for the two systems are determined independently, allowing for the separate identification of their respective real-time operating conditions. This provides data support for flexible switching between multiple navigation modes. The entire risk potential energy value calculation process relies on an offline, fixed function model on the airborne terminal, eliminating the need for a ground-based platform to perform data processing. This meets the timing requirements for autonomous real-time decision-making during aircraft flight. The update frequency of the risk potential energy values ​​is synchronized with the airborne sensor data acquisition frequency, ensuring that the risk assessment results match the changes in flight scenario conditions in real time, preventing any lag between assessment results and changes in operating conditions.

[0067] The process of mapping predicted navigation performance to navigation system reliability state variables specifically includes: First, normalizing the dimensions of navigation performance parameters: For the navigation positioning error parameter, the maximum error normalization method is used, dividing the positioning error value by a preset upper error limit, mapping it to the [0,1] interval. The larger the error, the higher the normalized value. For the navigation availability parameter, it is directly represented as a value in the [0,1] interval, with higher availability corresponding to a higher normalized value. Second, calculating the reliability state variables using a linear weighted summation method: The normalized positioning error parameter and availability parameter are multiplied by preset weights and then summed to obtain a navigation system reliability state variable with a value range of [0,1]. The sum of the weights for positioning error and availability is 1. The weight values ​​are preset using the analytic hierarchy process (AHP) combined with the aviation navigation industry's emphasis on the two types of parameters, or determined through statistical fitting of historical flight data. The third step is to unify the physical meaning of the credibility state variables: the closer the value is to 1, the more stable the corresponding navigation system is and the lower the risk of failure; the closer the value is to 0, the closer the operating state is to the physical failure boundary and the higher the risk of failure.

[0068] In alternative implementations, such as Figure 5 As shown, the step of generating and executing a flexible navigation strategy based on the navigation risk potential value and a preset decision-making style includes: S410: When the preset decision-making style is low-risk tolerance, when the navigation risk potential value of any navigation system reaches the warning threshold, the corresponding navigation mode is restricted or turned off, and other available navigation modes are switched.

[0069] S420: When the preset decision-making style is high-risk tolerance, the navigation mode is switched when the navigation risk potential value exceeds the failure threshold.

[0070] Specifically, during the approach and landing phase of low-risk tolerance-adapted aircraft, the flight phase has high control requirements for the integrity of navigation services and does not allow the navigation system to have any potential failure trend. Correspondingly, a low warning threshold is set, and a high sensitivity is maintained to the dynamic changes of navigation risk potential energy value. In practical applications, the airspace interference power of -120 dBmW can be set as the warning judgment threshold for low-risk tolerance.

[0071] When the navigation risk potential value corresponding to any type of navigation system reaches the preset warning threshold standard, even if the current operating parameters of the navigation system have not yet reached the physical failure boundary, the system will restrict or shut down the observation data access link of the navigation mode in advance, and actively switch to other stable navigation working modes in other operating states. By intervening in advance, potential navigation failure hazards are avoided, and the safety and stability of the flight operation process are prioritized.

[0072] The high-risk tolerance type is suitable for the cruise and tactical maneuver flight phases of aircraft. These phases prioritize continuous and uninterrupted navigation service output and can tolerate low-probability potential failures. Correspondingly, a higher failure threshold is set, and the allowable range for normal fluctuations in navigation risk potential energy values ​​is relaxed. In practical applications, an airspace interference power of -100 dB / mW can be set as the failure judgment threshold for the high-risk tolerance type. The system only triggers a navigation operating mode switch operation when the navigation risk potential energy value exceeds the preset failure threshold or the navigation system's operating state approaches the physical failure limit. This fully utilizes the inherent anti-disturbance capability of the integrated navigation filtering algorithm to maintain continuous and stable navigation service output, avoiding unnecessary fluctuations in navigation calculation results caused by frequent navigation mode switching.

[0073] The two decision-making styles can be automatically matched and adapted based on the flight mission phase markers, without the need for manual parameter configuration and mode switching. The airborne system autonomously determines the current adaptive decision logic based on flight altitude values ​​and route phase marker information, achieving unmanned autonomous adaptation of decision-making styles. When the aircraft performs flight missions in complex airspace, it can flexibly switch between a single satellite navigation operation mode, a single ground-based ranging navigation operation mode, and a multi-source combined navigation operation mode based on the real-time matching decision style and risk potential value comparison results. This adapts to the differentiated usage requirements of navigation integrity and continuity in different flight phases, meeting the adaptation requirements of planning and decision-making under the aircraft's own spatiotemporal interactive closed-loop operation architecture.

[0074] In alternative implementations, such as Figure 6 As shown, the generation and execution of the elastic navigation strategy includes: S510: Reconstruct the measurement matrix and measurement noise covariance matrix of the combined navigation filter based on the generated elastic navigation strategy.

[0075] S520: When switching to ground-based ranging-dominated mode, increase the satellite navigation measurement noise covariance or remove satellite navigation measurement dimensions. When multipath delay is detected in a specific ground-based ranging channel, perform innovation detection and isolation on that channel.

[0076] Specifically, the integrated navigation filter, as a functional unit for aviation navigation calculation, is internally configured with two types of operation matrices: a measurement matrix and a measurement noise covariance matrix. These two types of matrices directly determine the weight allocation ratio and data reliability classification standard of observation data from various navigation systems in the filtering and fusion calculation process.

[0077] After the flexible navigation strategy is generated, the airborne navigation computing unit synchronously responds to the strategy control commands and completes the reconstruction, adjustment, and adaptation of the internal matrix parameters of the filter in real time according to the command requirements. It dynamically matches the solution logic requirements of the current navigation operation mode, without using fixed matrix parameters to adapt to all flight airspace conditions. When the system determines to switch to the ground-based ranging-dominated operation mode, the system gradually increases the value of the measurement noise covariance matrix of the corresponding satellite navigation position, gradually weakening the weight of satellite navigation observation data in the filtering and fusion solution process. It can also directly remove the measurement dimension corresponding to satellite navigation based on the severity of airspace interference and obstruction conditions, and complete the aircraft spatial position calculation entirely based on ground-based ranging and navigation observation data.

[0078] During aircraft flight, the system continuously monitors the changes in multipath delay parameters corresponding to various ground-based ranging channels in real time. It uses a sliding window statistical method to track the fluctuation range of delay sampling data. When the multipath delay of a specific ground-based ranging channel exceeds the normal tolerance range, it is determined that the observation data of that channel has been severely interfered with and contaminated by multipath effects. The system initiates a novelty detection operation for that channel, compares the deviation between the actual observed novelty and the theoretical standard novelty, and after confirming the existence of abnormal observation deviation, completes the isolation processing operation of the corresponding channel to prevent contaminated observation data from entering the navigation filtering solution process, thus avoiding abnormal data from affecting the overall stability and accuracy of navigation solution.

[0079] The integrated navigation filter can adopt an extended Kalman filter architecture to adapt to the solution and operation requirements of nonlinear motion models and nonlinear observation models in aviation navigation. Filter reconstruction and channel anomaly isolation operations are automatically completed in real time by onboard software. The reconstruction transition process is smooth without abrupt step changes, ensuring the continuity and stability of the navigation output position, velocity, and attitude parameters. Under normal flight conditions without significant navigation failures, the system also compares the spatial distance between the satellite navigation solution position and the ground-based ranging multi-station intersection solution position in real time. A consistency statistic is constructed using the chi-square test to determine whether there are significant systematic deviations between the two types of navigation solution results. This serves as an auxiliary basis for judging sudden navigation failures, further improving the rationality of navigation mode switching and abnormal channel isolation.

[0080] In alternative implementations, such as Figure 7 As shown, the embodied spatiotemporal intelligent agent multi-source anti-interference fusion navigation method further includes: S610: During flight or after a mission, use post-processed data or extrapolated data as reference values ​​to calculate navigation residuals.

[0081] S620: Feed the navigation residuals back to the navigation cognitive model, adaptively adjust the measurement noise covariance matrix of the combined navigation filter online, store the navigation residual information, and perform offline incremental updates of the model weights after the aircraft finishes its flight mission.

[0082] S630: Periodically initiates incremental feature importance assessment, recalculates the average information gain of each feature based on cumulative flight data, dynamically adjusts feature weights, and eliminates low-contribution features and incorporates new related features, thereby achieving dynamic adaptive updating of the feature space.

[0083] In this embodiment, the parameter update and optimization process of the navigation cognitive model can be initiated both throughout the aircraft's flight and after landing following a single flight mission. To meet the deterministic and safety requirements of airworthiness, during flight, the system adaptively adjusts the measurement noise covariance matrix and state noise covariance matrix of the Kalman filter online based on the real-time residual information. After the flight mission, the airborne system updates the network weights using an incremental learning algorithm based on the stored data, serving as the update model for subsequent flight missions and completing the agent evolution loop.

[0084] The post-processing data uses navigation reference data generated by high-precision post-processing calculations, while the extrapolated data relies on stable position and attitude data obtained from long-term pure inertial calculations by the inertial navigation component as a reference benchmark. Both types of reference data have a stable degree of reliability and can be used as objective reference standards to calculate navigation residual values, thus avoiding the bias limitations of single reference data.

[0085] The navigation residuals are calculated by statistically analyzing the mean deviation between the real-time navigation solution results and the reference baseline data over a continuous flight period. Instead of using instantaneous deviations at a single moment as the basis for residual determination, this avoids distortion of the residual calculation results caused by instantaneous random disturbances in the airspace. The residual values ​​can intuitively reflect the degree of matching between the navigation cognitive model's inference accuracy and actual flight conditions. The calculated navigation residuals are fed back into the internal computational link of the navigation cognitive model. Incremental learning is used to make small online adjustments and optimizations to the network connection weights of the multilayer perceptron model. This eliminates the need for a full-scale iterative training process based on massive historical samples, resulting in a smaller overall computational load and making it suitable for the current hardware operating conditions and computing power configurations of airborne terminals requiring real-time online updates.

[0086] As residual data from multiple flight missions gradually accumulates, the model can slowly adapt to the feature correlation patterns under specific routes, specific airspace electromagnetic environments, and specific terrain occlusion scenarios, progressively optimizing the model's inference and prediction performance. Simultaneously, during the online adjustment process, constraints are set on the magnitude of network weight changes to prevent large fluctuations in network weights caused by single abnormal residual feedback, maintaining the stability of the navigation cognitive model's long-term operation. The model update frequency can be autonomously set according to the cumulative number of flights, cumulative flight duration, or residual exceedance threshold standards, balancing the relationship between model adaptation and optimization effects and airborne computing resource consumption, adapting to the usage requirements of different flight mission frequencies and route operating characteristics.

[0087] This closed-loop feedback update mechanism enables aircraft to gradually adapt to the inherent characteristics of specific airspace and electromagnetic environment as flight time increases, achieving continuous adaptation and evolution of navigation decision-making capabilities. This aligns with the operational characteristics of embodied intelligent agents that rely on real-world spatiotemporal environments for continuous interactive iterative optimization, eliminating the need for manual, periodic offline model retraining and reducing the workload of later maintenance and optimization of aviation navigation systems.

[0088] It should be noted that, from the perspective of the overall development and evolution of aviation navigation technology, most existing aviation navigation-related operational solutions focus on localized technical improvements and designs such as single navigation system signal filtering and optimization, static weighted fusion of multiple navigation systems under fixed empirical rules, and single discrete hard threshold triggering navigation mode switching. The application scenarios of these technologies are mostly limited to conventional and stable airspace flight conditions, and can only cope with simple scenarios of slight signal attenuation and low-intensity electromagnetic interference. For complex flight scenarios with the combined effects of complex weather conditions, mountainous urban canyon terrain obstruction, strong electromagnetic suppression interference, and high-dynamic maneuvering conditions of aircraft, the shortcomings of existing operational solutions will gradually become apparent.

[0089] Current aviation navigation operation schemes generally use discrete fixed numerical thresholds as the sole criterion for navigation risk assessment, without considering the parameter fluctuations caused by inherent measurement errors in airborne sensor hardware. When the sensor measurement parameters fluctuate slightly within the threshold critical range, navigation mode switching commands are repeatedly triggered, affecting the smoothness and continuity of navigation service output. At the same time, there is a lack of ability to predict the development trend of navigation system failure in advance. Mode switching operations can only be passively triggered after navigation performance has completely deteriorated and failed. The switching action has obvious time lag characteristics, which makes it difficult to meet the strict control requirements for the integrity of navigation services during high-safety-level flight phases such as approach and landing.

[0090] The incremental update of the model weights adopts the incremental backpropagation algorithm, setting a sliding time window with a length of 10 flight data. Only the newly added flight residual data within the time window is used as training samples to adjust the connection weights between the output layer and the last hidden layer of the multilayer perceptron, without changing the weight parameters of the bottom feature extraction layer. At the same time, a weight adjustment constraint is set, which stipulates that the adjustment range of a single weight in a single incremental training does not exceed 5% of its original value, so as to avoid large drift of model parameters caused by a single abnormal data and ensure the stability of airborne operation.

[0091] The incremental feature importance assessment employs a sliding window incremental random forest algorithm, setting a sliding time window of 30 days of cumulative flight data. For each assessment period, based on newly added flight data within the window, the average information gain of each feature at the decision tree split node is incrementally calculated. Feature elimination and feature inclusion thresholds are set. When the average information gain of an existing feature is below the elimination threshold of 0.01 for three consecutive assessment periods, the feature is removed from the key feature set. When the average information gain of a newly added scene feature is above the inclusion threshold of 0.1 for three consecutive assessment periods, the feature is included in the key feature set, simultaneously adjusting the corresponding dimensions of the multilayer perceptron input layer.

[0092] It should be noted that the parameters mentioned above in this application are merely illustrative, and those skilled in the art can set them according to actual needs; this application does not limit them in this regard.

[0093] Optionally, the periodic incremental feature importance assessment specifically includes: using cumulative flight time, cumulative flight sorties, or navigation residual overrun frequency as assessment trigger conditions, and employing an incremental random forest algorithm to calculate the average information gain of each feature at the decision tree split node based on newly added flight data within a sliding time window. When the average information gain of an existing feature is lower than a preset elimination threshold for multiple consecutive assessment periods, the feature is removed from the key feature set, and the corresponding dimension of the multilayer perceptron input layer is simultaneously pruned. When the average information gain of newly collected scene features is higher than a preset inclusion threshold for multiple consecutive assessment periods, the feature is included in the key feature set, and the corresponding dimension of the multilayer perceptron input layer is simultaneously expanded, completing the dynamic iterative evolution of the input feature space.

[0094] The technical solution adopted in this application does not require modification to the existing airborne hardware architecture of the aircraft. All functions can be operated by simply deploying and configuring the airborne software program. The engineering implementation cost is low, and it is compatible with the upgrade and modification needs of various existing aircraft as well as the standard deployment needs of newly developed aircraft. It can simultaneously adapt to various application scenarios such as precision approach and landing of manned aircraft, long-endurance cruise flight, long-distance autonomous flight of unmanned aircraft, and tactical maneuver flight in complex airspace. It can maintain stable and continuous operation of navigation services in complex spatiotemporal scenarios such as mountainous airspace, urban canyon airspace, and airspace with strong electromagnetic interference. It effectively improves the problems of insufficient stability, limited adaptability, and lack of autonomous optimization capability of traditional aviation navigation solutions under complex conditions. The entire operation process does not require human intervention. The entire process is completed autonomously by the airborne system, including scene perception, risk handling, assessment and decision-making, strategy generation, and model update. It is compatible with the development trend of unmanned aerial navigation.

[0095] This application is not limited to this. When an aircraft flies in mountainous airspace, the undulating terrain can easily obstruct the transmission paths of some navigation satellite signals, reducing the number of available navigation satellites. Consequently, the values ​​of the satellite geometric accuracy factor and horizontal accuracy factor increase, and the multipath delay parameter also changes abnormally due to signal reflection from the mountains. This application can accurately capture the changing trends of parameters such as accuracy factor, carrier-to-noise ratio, and multipath delay by extracting key feature parameters. It can predict the deterioration trend of satellite navigation performance in advance by relying on the navigation cognitive model, quantify the failure risk level of satellite navigation by using the navigation risk potential energy field, and promptly and autonomously switch to the ground-based ranging and navigation-dominated operation mode to maintain uninterrupted navigation service output.

[0096] When aircraft fly over urban canyon airspace, the dense arrangement of high-rise buildings can simultaneously cause navigation signal blockage and multipath effects. The electromagnetic environment also becomes more complex due to radiation from various electronic devices in the city, and the interference-to-signal ratio and carrier-to-noise ratio parameters fluctuate continuously. This application can simultaneously sense the coupling changes of multiple characteristic parameters, adapt the risk judgment logic according to the preset decision style, flexibly adjust the navigation operation mode, and avoid the service interruption problem caused by the failure of a single navigation system.

[0097] When an aircraft operates in an airspace with strong electromagnetic interference, various suppressive and adjacent-channel interference signals can cause a significant decrease in the navigation interference-to-signal ratio and a decline in the quality of navigation signal reception. This application can quickly detect the changes in key characteristic parameters such as interference-to-signal ratio and carrier-to-noise ratio, quantify the navigation operation risks caused by electromagnetic interference, match the corresponding decision-making style according to different flight phases of approach and cruise, and choose to restrict the operation of the interfered navigation system in advance or maintain the stable operation of the integrated navigation, thus adapting to the flexible operation requirements under complex electromagnetic interference scenarios.

[0098] When an aircraft performs a high-dynamic maneuvering flight mission, the rapid changes in the aircraft's attitude and flight speed will alter the signal reception angle and propagation path of the navigation antenna, indirectly causing multipath delay fluctuations and changes in the signal-to-noise ratio. This application incorporates the aircraft's motion state parameters into the scope of key feature evaluation, which can fully characterize the correlation between maneuvering flight conditions and navigation performance, predict navigation state fluctuations caused by maneuvering actions in advance, complete navigation mode adaptation preparations in advance, and further improve the adaptability of navigation decisions under complex maneuvering conditions.

[0099] During long-term, fixed-route, routine flights, navigation residual data in the airspace environment of that route will be continuously accumulated. Through incremental learning, the network weight parameters of the multilayer perceptron model will be gradually optimized, allowing the navigation cognitive model to gradually adapt to the unique characteristics of the terrain, electromagnetic radiation, and meteorological environment of that route. After long-term operation, the adaptation performance of navigation performance prediction will gradually improve, and the rationality and timeliness of navigation mode switching will also be improved simultaneously. There is no need for manual offline retraining of model parameters on a regular basis, which greatly reduces the workload of later maintenance of the aviation navigation system. The entire operational process perfectly aligns with the closed-loop interactive logic of an aircraft as an embodied intelligent agent, perceiving, understanding, planning, and acting. Starting with the perception and collection of the external airspace environment and its own flight status, it goes through feature modeling, performance prediction, risk assessment, decision generation, and strategy execution. Finally, through operational data feedback, it achieves autonomous iterative optimization of the model, forming a complete spatiotemporal interactive operational closed loop. Compared to traditional fixed-rule aviation navigation solutions, it can adapt to more complex flight conditions, improve the stability and continuity of navigation service operation, and break through the limitations of traditional solutions that can only optimize model parameters. It has the operational characteristics of long-term environmental adaptive evolution, realizing the co-evolution of the embodied intelligent agent and the flight environment. It aligns with the future development and application direction of intelligent aviation navigation systems and has good application and promotion value in multiple fields such as civil aviation transport flight, general aviation operation flight, military aircraft tactical flight, and autonomous navigation of unmanned aerial vehicles.

[0100] This application also discloses an embodied spatiotemporal intelligent agent multi-source anti-interference fusion navigation system, such as Figure 8 As shown, it includes: The scene perception module 11 is used to obtain the real-time multimodal perception information of the aircraft in the current scene based on the key features pre-selected from historical flight data using the random forest algorithm, and to perform standardization processing to obtain the embodied state feature vector.

[0101] The evaluation calculation module 12 is used to input the embodied state feature vector into a pre-trained navigation cognitive model to obtain the predicted navigation performance, construct the navigation scenario risk potential energy field, and calculate the navigation risk potential energy value based on the predicted navigation performance and the navigation scenario risk potential energy field.

[0102] The flexible decision module 13 is used to generate and execute a flexible navigation strategy based on the navigation risk potential value and the preset decision style. The flexible navigation strategy includes switching between satellite navigation-dominated mode, ground-based ranging-dominated mode and combined navigation mode.

[0103] Based on the same principle, specific embodiments of the system in this application can be found in the specific implementation of the method, and this application does not limit them.

[0104] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0105] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A multi-source anti-interference fusion navigation method for embodied spatiotemporal intelligent agents, characterized in that, include: Based on key features pre-selected from historical flight data using the random forest algorithm, real-time multimodal perception information of the aircraft in the current scenario is obtained and standardized to obtain an embodied state feature vector. The embodied state feature vector is input into a pre-trained navigation cognition model to obtain the predicted navigation performance, a navigation scenario risk potential energy field is constructed, and the navigation risk potential energy value is calculated based on the predicted navigation performance and the navigation scenario risk potential energy field. Based on the navigation risk potential value and the preset decision-making style, a flexible navigation strategy is generated and executed. The flexible navigation strategy includes switching between satellite navigation-dominated mode, ground-based ranging-dominated mode and combined navigation mode.

2. The method according to claim 1, characterized in that, The key features include carrier-to-noise ratio, interference-to-signal ratio, multipath delay, accuracy factor, and aircraft motion state parameters.

3. The method according to claim 1, characterized in that, The navigation cognitive model is a nonlinear mapping model trained by a multilayer perceptron, with the key features as input and navigation positioning error and availability as output.

4. The method according to claim 3, characterized in that, The training steps of the navigation cognitive model include: Historical flight data is acquired, and the average information gain of multi-dimensional scene features at the decision tree splitting node is calculated using the random forest algorithm. Key features are then selected based on the average information gain. A multilayer perceptron model is constructed, with the selected key features as input and navigation positioning error and availability as output. The model is trained using the backpropagation algorithm to obtain the navigation cognitive model.

5. The method according to claim 1, characterized in that, The constructed navigation scenario risk potential energy field includes: Set the physical failure boundary values ​​for each navigation system; Define a risk potential energy function such that the value of the risk potential energy function increases exponentially as the navigation system reliability state variable is within a preset threshold as the difference between it and the physical failure boundary value is within a preset threshold. The perception error is modeled using a Gaussian distribution function, and the perception uncertainty is superimposed on the risk potential energy function to obtain the risk potential energy field of the navigation scenario.

6. The method according to claim 5, characterized in that, The calculation of the navigation risk potential energy value based on the predicted navigation performance and the navigation scenario risk potential energy field includes: The predicted navigation performance is mapped to the navigation system reliability state variable; The navigation risk potential value is calculated based on the navigation system reliability state variable and the risk potential function superimposed with perceived uncertainty. The navigation risk potential value represents the probability density of navigation system failure.

7. The method according to claim 1, characterized in that, The step of generating and executing a flexible navigation strategy based on the navigation risk potential value and a preset decision-making style includes: When the preset decision-making style is low-risk tolerance, when the navigation risk potential value of any navigation system reaches the warning threshold, the corresponding navigation mode is restricted or turned off, and other available navigation modes are switched. When the preset decision-making style is high-risk tolerance, the navigation mode is switched when the navigation risk potential value exceeds the failure threshold.

8. The method according to claim 1, characterized in that, The generation and execution of the elastic navigation strategy includes: The measurement matrix and measurement noise covariance matrix of the combined navigation filter are reconstructed based on the generated elastic navigation strategy. When switching to ground-based ranging-dominated mode, increase the covariance of satellite navigation measurement noise or remove satellite navigation measurement dimensions; when multipath delay of a specific ground-based ranging channel is detected, perform information detection and isolation on that channel.

9. The method according to claim 4, characterized in that, Also includes: During flight or after a mission, post-processed or extrapolated data are used as reference values ​​to calculate navigation residuals. The navigation residuals are fed back to the navigation cognitive model, the measurement noise covariance matrix of the combined navigation filter is adjusted online adaptively, the navigation residual information is stored, and the model weights are updated offline after the aircraft finishes its flight mission. Incremental feature importance assessment is initiated periodically. The average information gain of each feature is recalculated based on the cumulative flight data, the feature weights are dynamically adjusted, and the elimination of low-contribution features and the inclusion of new related features are completed, thereby achieving dynamic adaptive updating of the feature space.

10. A multi-source anti-interference fusion navigation system for an embodied spatiotemporal intelligent agent, characterized in that, include: The scene perception module is used to obtain the aircraft's real-time multimodal perception information in the current scene based on key features pre-selected from historical flight data using the random forest algorithm, and to perform standardization processing to obtain the embodied state feature vector. The evaluation calculation module is used to input the embodied state feature vector into a pre-trained navigation cognitive model to obtain the predicted navigation performance, construct the navigation scenario risk potential energy field, and calculate the navigation risk potential energy value based on the predicted navigation performance and the navigation scenario risk potential energy field. The flexible decision-making module is used to generate and execute flexible navigation strategies based on navigation risk potential values ​​and preset decision-making styles. The flexible navigation strategies include switching between satellite navigation-dominated mode, ground-based ranging-dominated mode, and combined navigation mode.

Images