Beidou satellite intelligent positioning and communication method and system based on artificial intelligence

Abstract

The invention discloses a Beidou satellite intelligent positioning and communication method and system based on artificial intelligence. The method comprises the following steps: S1, receiving a satellite signal and a cooperative signal; s2, processing the received satellite signal by using a pseudo-range measurement and carrier phase difference technology, and performing preliminary positioning calculation in combination with the cooperative positioning data; s3, fusion is carried out, and a multi-modal data set is constructed; s4, dynamically adjusting weights of different data sources in the multi-modal data set according to real-time environment change by utilizing a dynamic environment perception model based on a graph neural network; s5, processing the multi-modal data set by adopting a self-adaptive Monte Carlo variational inference algorithm, and optimizing the positioning precision; s6, adjusting a filtering parameter and a signal processing strategy in real time through a self-adaptive signal processing module; and S7, dynamically optimizing the communication frequency and path by using a dominant actor-commentator algorithm. According to the invention, multi-modal data fusion and a dynamic optimization algorithm are utilized, so that the precision and stability of Beidou satellite positioning and communication are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of satellite navigation and communication, and particularly to a Beidou satellite intelligent positioning and communication method and system based on artificial intelligence. Background Art

[0003] Existing satellite positioning systems usually rely on the signals of a single satellite system. Although collaborative positioning of multiple satellite systems has become a trend, in complex environments, a single satellite system is easily affected by environmental factors and cannot guarantee continuous high-precision positioning. For example, multipath effects will occur between high-rise buildings in the city. After the signal is reflected by the building and reaches the receiver, it will cause errors in the measured pseudorange and affect the positioning accuracy. In addition, in areas with severe signal occlusion such as forests, the satellite signal strength will weaken or even be lost, and the positioning system cannot obtain enough satellite signals, resulting in positioning failure or a significant decrease in accuracy. Although some systems have introduced an inertial navigation system (INS) as an auxiliary, traditional inertial navigation accumulates errors during long-term use and it is difficult to compensate for the lack of satellite signals for a long time.

[0004] On the other hand, the problem of signal interference also poses challenges to existing Beidou satellite positioning and communication systems. In a complex electromagnetic environment, factors such as electromagnetic interference, noise, and multipath effects greatly reduce the quality of the received signal. Traditional fixed filters and anti-interference algorithms cannot be adjusted according to the changes in the real-time environment. Therefore, in the case of severe interference, the quality of positioning and communication drops significantly. At the same time, when existing satellite communication systems face concurrent communication of multiple devices, the allocation of communication frequencies and path selection are relatively fixed, and it is easy to cause congestion when the network load increases, resulting in an increase in communication delay and seriously affecting the overall performance of the system.

[0005] To overcome these technical problems, artificial intelligence technology has gradually been introduced into the field of satellite navigation and communication. AI technologies such as deep learning and reinforcement learning have powerful data processing and pattern recognition capabilities and can dynamically optimize the system according to the changes in the real-time environment. For example, AI-driven multimodal data fusion technology can fuse satellite signals, inertial navigation data, geographic information system data, and external auxiliary data (such as meteorological data, terrain data, etc.), so that the system can provide reliable positioning information through other data sources even when satellite signals are occluded or lost in complex environments. In addition, AI can also learn the signal interference patterns in different environments and adjust the filtering parameters and anti-interference strategies in real time to ensure high communication quality in complex electromagnetic environments.

[0006] Although artificial intelligence technology has shown great potential in the fields of satellite positioning and communication, existing AI technologies still face many challenges in practical applications. First, most existing multi-modal data fusion technologies rely on traditional Kalman filtering algorithms. Although the system accuracy can be improved to a certain extent, in complex environments, the performance of Kalman filtering will decline significantly. Especially when there are large noises and errors between multiple data sources, traditional weighted fusion methods are difficult to work effectively. Second, most existing anti-interference technologies use fixed filter parameters and cannot be adaptively adjusted according to real-time interference intensity and type, resulting in unstable signal quality. In addition, traditional communication frequency allocation and path selection methods are relatively fixed and difficult to cope with the dynamic changes of network load. Communication delays and congestion problems are likely to occur when multiple devices are concurrent.

[0007] Therefore, how to provide an intelligent Beidou satellite positioning and communication method and system based on artificial intelligence is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0008] An object of the present invention is to provide an intelligent Beidou satellite positioning and communication method and system based on artificial intelligence. The present invention adopts AI-driven multi-modal data fusion technology and dynamic optimization algorithms. By combining satellite signals, inertial navigation data, geographic information system data, and external auxiliary data, it effectively solves the problems of insufficient positioning accuracy and communication interference in complex environments. Using a dynamic perception model based on graph neural networks and an adaptive Monte Carlo variational inference algorithm, it can adjust the data source weights in real time, correct positioning errors, and dynamically optimize communication frequencies and paths through the Advantage Actor-Critic algorithm. This method significantly improves the accuracy and stability of Beidou satellite positioning and communication, and has the advantages of strong anti-interference, low communication delay, and adaptability to complex environments.

[0009] According to the intelligent Beidou satellite positioning and communication method based on artificial intelligence of the embodiments of the present invention, the following steps are included:

[0010] S1. Receive satellite signals from the Beidou satellite navigation system and at the same time receive cooperative signals from other satellite systems for cooperative positioning of multiple satellite systems;

[0011] S2. Use pseudorange measurement and carrier phase differential technology to process the received satellite signals, and combine cooperative positioning data for preliminary positioning calculation;

[0012] S3. Through an AI-driven multi-modal data fusion module, fuse the received satellite signals, inertial navigation data, geographic information system data, and external auxiliary data to construct a multi-modal data set;

[0013] S4. Use the dynamic environment perception model based on graph neural network to dynamically adjust the weights of different data sources in the multi-modal dataset according to real-time environmental changes. When satellite signals are blocked, weak, or lost, increase the weights of inertial navigation data and external auxiliary data;

[0014] S5. Process the multi-modal dataset using the adaptive Monte Carlo variational inference algorithm to correct the positioning errors caused by multipath effects and signal interference and optimize the positioning accuracy;

[0015] S6. Through the adaptive signal processing module, monitor the interference conditions in communication and positioning signals, identify electromagnetic interference and noise characteristics, and adjust the filtering parameters and signal processing strategies in real time;

[0016] S7. Use the Advantage Actor-Critic algorithm to dynamically optimize the communication frequency and path according to the real-time location, network load, and signal strength of the terminal device.

[0017] Optionally, the other satellite systems include the Global Positioning System, the GLONASS system, and the Galileo positioning system.

[0018] Optionally, the S2 specifically includes:

[0019] S21. Receive satellite signals, calculate and record the pseudorange of each satellite:

[0020]

[0021] where P i represents the pseudorange between the i-th satellite and the receiver, x i , y i and z i represent the known coordinates of the i-th satellite at the time of reception, x r , y r and z r represent the unknown coordinates of the receiver, c represents the speed of light, Δt r represents the receiver clock error, I i represents the ionospheric delay of the i-th satellite, and ε i represents the measurement noise of the i-th satellite;

[0022] S22. Conduct carrier phase observations on each satellite:

[0023]

[0024] where Φ i represents the carrier phase observation value between the i-th satellite and the receiver, λ represents the carrier wavelength, η i represents the phase measurement noise of the i-th satellite, and N i represents the integer ambiguity of the i-th satellite;

[0025] S23. Construct a single-difference carrier-phase observation equation from the received satellite signals:

[0026]

[0027]

[0028] where, ΔΦ ij represents the single-difference carrier-phase observation value between the i-th satellite and the j-th satellite, and N j represents the integer ambiguity of the j-th satellite;

[0029] S24. Construct a double-difference observation model from the single-difference observation values:

[0030] Δ 2 Φ ijk = ΔΦ ij - ΔΦ ik ;

[0031] where, ΔΦ ik represents the single-difference carrier-phase observation value between the i-th satellite and the k-th satellite, and Δ 2 Φ ijk represents the double-difference observation value among the i-th satellite, the j-th satellite, and the k-th satellite;

[0032] S25. Combine the pseudo-range and carrier-phase double-difference observation values to construct a non-linear equation set, and use the extended Kalman filter or adaptive Kalman filter algorithm to calculate the preliminary position coordinates (x r , y r , z r ) of the receiver and the receiver clock deviation Δt r .

[0033] Optionally, the S4 specifically includes:

[0034] S41. Obtain a multi-modal data set D = {S, I, G, A}, where S represents satellite signal data, I represents inertial navigation data, G represents geographic information system data, and A represents external auxiliary data;

[0035] S42. Construct a graph structure G = (V, E) for the obtained multi-modal data set, where V represents the node set of multi-modal data sources, E represents the spatio-temporal dependence relationship between data sources, and the edge weight of the graph is:

[0036] W uv = f(d(u, v), t(u, v))·g(σ u , σ v , θ);

[0037] where, Wuv denotes the edge weight from node u to node v, d(u, v) represents the spatial distance between nodes u and v, t(u, v) represents the temporal correlation between nodes u and v, and σ u represents the environmental state of node u, and σ v represents the environmental state of node v, θ represents the environmental adaptive adjustment coefficient, the function f represents the spatio-temporal coupling relationship between nodes, and the function g represents the correlation of different data sources under the environmental state;

[0038] S43. On the constructed graph structure, a graph neural network is used to update the features of nodes, and the feature of each node is represented as a vector Through the message passing mechanism, the node feature update formula is:

[0039]

[0040] where represents the feature vector of node v at the l + 1 layer, N(v) represents the set of neighbor nodes of node v, represents the edge weight from node u to node v at the l layer, represents the bias vector, and σ represents the activation function;

[0041] S44. Model the environmental perception process, and use the graph neural network model to dynamically adjust the weights of different data sources according to the real-time environmental state:

[0042]

[0043] where w i represents the i-th data source, β i represents the adaptive weight factor of the i-th data source, β j represents the adaptive weight factor of the j-th data source, τ represents the environmental state vector, and α i represents the weight adjustment parameter of the i-th data source, and α j represents the weight adjustment parameter of the j-th data source;

[0044] S45. Through the multi-layer iterative optimization of the graph neural network, generate the final weighted fusion model of data sources, integrate the information of multiple data sources, and generate the fused feature vector F:

[0045]

[0046] where D i represents the original data of the i-th data source, γ uv represents the coupling strength of the edge connecting nodes u and v, h u represents the feature vector of node u, and h vRepresents the feature vector of node v.

[0047] Optionally, S5 specifically includes:

[0048] S51. Define a latent variable K, and the latent variable follows a Gaussian distribution:

[0049]

[0050] where μ i represents the mean of the i-th latent variable, Σ i represents the covariance matrix of the i-th latent variable, and M represents the number of latent variables;

[0051] S52. Use adaptive Monte Carlo sampling to generate an initial sample set and update the initial sample set:

[0052]

[0053] where represents the sample value at the (k + 1)-th iteration, represents the sample value at the k-th iteration, δ represents the adaptive step size, represents the gradient of the latent variable, ξ i represents standard normal distribution noise, and D represents the multimodal dataset;

[0054] S53. Optimize the distribution of the latent variable through variational inference and minimize the KL divergence:

[0055]

[0056] where L(q) represents the objective function, q(K) represents the approximate distribution of variational inference, and p(D, K) represents the joint distribution of the multimodal dataset D and the latent variable K;

[0057] S54. Combine importance sampling and variational inference, and correct the weights of each latent variable sample through the weight resampling formula:

[0058]

[0059] where ω i represents the weight of the i-th sample, p(D|z i ) represents the likelihood function, p(z i ) represents the prior distribution, q(z i |D) represents the variational posterior distribution, χ represents the adaptive correction factor, and f(σ, D) represents the weight correction function caused by environmental noise and system uncertainty;

[0060] S55. Generate a final corrected dataset based on the corrected weights:

[0061]

[0062] Among them, S represents the corrected positioning data, and k i represents the i-th latent variable, and α i represents the correction factor of the external auxiliary data, and A i represents the i-th external data, and N represents the total number of latent variables;

[0063] S56. Through multiple iterations, the adaptive Monte Carlo variational inference algorithm continuously optimizes the latent variable distribution and combines a weighted correction mechanism to finally generate a high-precision corrected data set and optimize the positioning accuracy.

[0064] Optionally, the S6 specifically includes:

[0065] S61. Obtain the original communication and positioning signal data, including the target signal, environmental noise, and electromagnetic interference:

[0066] S raw = S target + N EMI + N thermal ;

[0067] Among them, S raw represents the original signal, S target represents the target signal, N EMI represents the electromagnetic interference, and N thermal represents the thermal noise;

[0068] S62. Perform preliminary processing on the original signal through an adaptive filter, and the filtering parameters of the filter are dynamically adjusted according to the current environment to generate preprocessed signal data;

[0069] S63. Use a convolutional neural network to extract features from the filtered signal, identify the interference patterns and noise characteristics in the signal, and generate an interference feature matrix;

[0070] S64. Based on the extracted interference feature matrix, calculate the power spectral density of the noise and identify the influence of different noise sources on the signal:

[0071]

[0072] Among them, P noise represents the power spectral density of the noise, T represents the signal observation duration, and S filtered (t) represents the value of the filtered signal at time t;

[0073] S65. Dynamically adjust the frequency response of the adaptive filter according to the calculation result of the noise power spectral density:

[0074]

[0075] Among them, W (k+1) represents the parameter of the filter at the (k + 1)-th iteration, and W (k) represents the parameter of the filter at the k-th iteration, and ξ represents the adaptive learning rate. represents the gradient of the noise power spectral density with respect to the filter parameter;

[0076] S66. Online optimize the signal processing result, gradually reduce noise and electromagnetic interference through the adaptive mechanism, and finally output the processed target signal.

[0077] Optionally, the S7 specifically includes:

[0078] S71. Initialize the status information of the terminal device, including the real-time location L t , the network load situation N t and the signal strength S t , and construct the status vector s t :

[0079] s t ={L t , N t , S t};

[0080] S72. Generate the optimal action policy in the current state through the actor network of the Advantage Actor-Critic algorithm. The action policy is expressed as:

[0081] π(a t |s t , θ actor );

[0082] Among them, π(a t |s t ) represents the probability of taking action a t in state s t , a t represents the action vector, and θ actor represents the parameter of the actor network. The actions include the adjustment of the communication frequency and the selection of the path;

[0083] S73. Estimate the value function in the current state through the critic network and update the action policy. The value function is defined as:

[0084]

[0085] Among them, V(s t ) represents the value function of state s t , r t+k+1 represents the immediate reward at time t + k + 1. Denotes the discount factor, θ critic Denotes the parameters of the critic network;

[0086] S74. Based on the feedback from the critic network, improve the policy of the actor network, with the goal of maximizing the advantage function A(s t , a t ):

[0087]

[0088] S75. According to the immediate reward r of the feedback t , update the parameters of the actor network through the gradient descent method, and the update formula is:

[0089]

[0090] where α r Denotes the learning rate, Denotes the policy gradient;

[0091] S76. Update the state vector s after each communication event t+1 , and dynamically adjust the communication parameters through the Advantage Actor-Critic algorithm to optimize the communication frequency and path in real time.

[0092] The Beidou satellite intelligent positioning and communication system based on artificial intelligence according to the embodiment of the present invention includes the following modules:

[0093] Satellite signal receiving module, used to receive satellite signals of the Beidou satellite navigation system, and at the same time receive collaborative signals to achieve collaborative positioning of multi-satellite systems;

[0094] Pseudorange measurement and carrier phase processing module, used to process the received satellite signals through pseudorange measurement and carrier phase differential technology, and perform preliminary positioning calculations in combination with collaborative positioning data;

[0095] AI-driven multi-modal data fusion module, used to receive and fuse satellite signals, inertial navigation data, geographic information system data and external auxiliary data to construct a multi-modal data set;

[0096] Dynamic environment perception module based on graph neural network, used to dynamically adjust the weights of different data sources in the multi-modal data set according to real-time environmental changes;

[0097] Adaptive Monte Carlo variational inference module, used to process the multi-modal data set, correct the positioning errors caused by multipath effects and signal interference through the adaptive Monte Carlo variational inference algorithm, and optimize the positioning accuracy;

[0098] An adaptive signal processing module, which is used to monitor the electromagnetic interference and noise characteristics in communication and positioning signals, and adjust the filtering parameters and signal processing strategies in real time;

[0099] A communication optimization module based on the Advantage Actor-Critic algorithm, which dynamically optimizes the communication frequency and path according to the real-time position, network load and signal strength of the terminal device.

[0100] The beneficial effects of the present invention are as follows:

[0101] First of all, the present invention utilizes the AI-driven multi-modal data fusion technology to organically combine Beidou satellite signals, inertial navigation data, geographic information system data and external auxiliary data, so that even in the case of high-rise buildings in the city, forest coverage or signal occlusion, the system can still maintain high-precision positioning. Compared with traditional single data source or fixed weighting methods, the present invention uses a dynamic environment perception model based on graph neural network to monitor and adjust the weights of different data sources in real time. Especially when the signal is weak or lost, the contribution of inertial navigation and auxiliary data is increased, making the system positioning more stable and reliable.

[0102] In terms of signal processing, the present invention uses the Adaptive Monte Carlo Variational Inference algorithm to efficiently correct the noise, multipath effect and signal interference in the multi-modal data set, significantly improving the positioning accuracy. This adaptive algorithm has stronger robustness and self-adaptability compared with the traditional Kalman filtering algorithm, and can provide more accurate error correction in the face of dynamic and changing environments. In addition, through the adaptive signal processing module, the convolutional neural network is used to monitor the electromagnetic interference and noise in the signal in real time, and the system can dynamically adjust the filter parameters and signal processing strategies to effectively cope with the complex electromagnetic environment and ensure that the quality of communication and positioning signals is always at a high level.

[0103] At the same time, the present invention introduces the Advantage Actor-Critic algorithm in communication optimization, and dynamically adjusts the communication frequency and path according to the real-time position, network load and signal strength of the terminal device. This enables the system to effectively allocate frequency resources when multiple devices communicate concurrently, avoiding channel congestion and communication delay. Compared with traditional fixed frequency allocation and path selection schemes, the dynamic optimization mechanism of the present invention ensures the communication stability and low latency of the system in high-load environments, greatly improving the overall communication performance. Description of the Drawings

[0104] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0105] Figure 1Overall flowchart of the Beidou satellite intelligent positioning and communication method based on artificial intelligence proposed by the present invention;

[0106] Figure 2 Schematic structural diagram of the Beidou satellite intelligent positioning and communication system based on artificial intelligence proposed by the present invention. Detailed implementation manners

[0107] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0108] Refer to Figure 1 , the Beidou satellite intelligent positioning and communication method based on artificial intelligence includes the following steps:

[0109] S1. Receive satellite signals from the Beidou satellite navigation system, and at the same time receive collaborative signals from other satellite systems for collaborative positioning of multi-satellite systems;

[0110] S2. Process the received satellite signals using pseudorange measurement and carrier phase differential techniques, and perform preliminary positioning calculations in combination with collaborative positioning data;

[0111] S3. Through an AI-driven multi-modal data fusion module, fuse the received satellite signals, inertial navigation data, geographic information system data, and external auxiliary data to construct a multi-modal data set;

[0112] S4. Use a dynamic environment perception model based on graph neural networks to dynamically adjust the weights of different data sources in the multi-modal data set according to real-time environmental changes. When satellite signals are blocked, weak, or lost, increase the weights of inertial navigation data and external auxiliary data;

[0113] S5. Adopt an adaptive Monte Carlo variational inference algorithm to process the multi-modal data set, correct positioning errors caused by multipath effects and signal interference, and optimize positioning accuracy;

[0114] S6. Through an adaptive signal processing module, monitor the interference conditions in communication and positioning signals, identify electromagnetic interference and noise characteristics, and adjust filtering parameters and signal processing strategies in real time;

[0115] S7. Use the Advantage Actor-Critic algorithm to dynamically optimize communication frequencies and paths according to the real-time position, network load, and signal strength of the terminal device.

[0116] In this embodiment, the other satellite systems include the Global Positioning System, the GLONASS system, and the Galileo positioning system.

[0117] In this embodiment, the specific content of S2 includes:

[0118] S21. Receive satellite signals, calculate and record the pseudorange of each satellite:

[0119]

[0120] where P i represents the pseudorange between the i-th satellite and the receiver, x i , y i and z i represent the known coordinates of the i-th satellite at the time of reception, x r , y r and z r represent the unknown coordinates of the receiver, c represents the speed of light, Δt r represents the receiver clock error, I i represents the ionospheric delay of the i-th satellite, ε i represents the measurement noise of the i-th satellite;

[0121] S22. Conduct carrier phase observations on each satellite:

[0122]

[0123] where Φ i represents the carrier phase observation value between the i-th satellite and the receiver, λ represents the carrier wavelength, η i represents the phase measurement noise of the i-th satellite, N i represents the integer ambiguity of the i-th satellite;

[0124] S23. Construct a single-difference carrier phase observation equation through the received signals of multiple satellites:

[0125]

[0126] where ΔΦ ij represents the single-difference carrier phase observation value between the i-th satellite and the j-th satellite, N j represents the integer ambiguity of the j-th satellite;

[0127] S24. Construct a double-difference observation model through the single-difference observation values:

[0128] Δ 2 Φ ijk = ΔΦ ij - ΔΦ ik ;

[0129] where ΔΦ ik represents the single-difference carrier phase observation value between the i-th satellite and the k-th satellite, Δ 2 Φ ijkDenote the double-difference observation value between the i-th satellite, the j-th satellite and the k-th satellite;

[0130] S25. Combine the pseudo-range and carrier-phase double-difference observation values to construct a non-linear equation system, and use the extended Kalman filter or adaptive Kalman filter algorithm to calculate the preliminary position coordinates (x r , y r , z r ) of the receiver and the receiver clock bias Δt r .

[0131] In this embodiment, the S4 specifically includes:

[0132] S41. Obtain a multi-modal data set D = {S, I, G, A}, where S represents satellite signal data, I represents inertial navigation data, G represents geographic information system data, and A represents external auxiliary data;

[0133] S42. Construct a graph structure G = (V, E) for the obtained multi-modal data set, where V represents the node set of multi-modal data sources, and E represents the spatio-temporal dependence relationship between data sources. The edge weight of the graph is:

[0134] W uv = f(d(u, v), t(u, v))·g(σ u , σ v );

[0135] Among them, W uv represents the edge weight from node u to node v, d(u, v) represents the spatial distance between node u and node v, t(u, v) represents the time correlation between node u and node v, σ u represents the environmental state of node u, σ v represents the environmental state of node v, θ represents the environmental adaptive adjustment coefficient, function f represents the spatio-temporal coupling relationship between nodes, and function g represents the correlation between different data sources under the described environmental state;

[0136] S43. On the constructed graph structure, use a graph neural network to update the features of nodes. The feature of each node is represented as a vector Through the message passing mechanism, the node feature update formula is:

[0137]

[0138] Among them, represents the feature vector of node v at the l + 1 layer, N(v) represents the set of neighbor nodes of node v, represents the edge weight from node u to node v at the l layer, represents the bias vector, and σ represents the activation function;

[0139] S44. Model the environmental perception process and use the graph neural network model to dynamically adjust the weights of different data sources according to the real-time environmental state:

[0140]

[0141] Among them, w i represents the i-th data source, and β i represents the adaptive weight factor of the i-th data source, and β j represents the adaptive weight factor of the j-th data source, τ represents the environmental state vector, and α i represents the weight adjustment parameter of the i-th data source, and α j represents the weight adjustment parameter of the j-th data source;

[0142] S45. Through the multi-layer iteration optimization of the graph neural network, generate the final weighted fusion model of the data sources, synthesize the information of multiple data sources, and generate the fused feature vector F:

[0143]

[0144] Among them, D i represents the original data of the i-th data source, γ uv represents the coupling strength of the edge connecting node u and node v, h u represents the feature vector of node u, and h v represents the feature vector of node v.

[0145] In this embodiment, the specific content of S5 includes:

[0146] S51. Define the latent variable K, and the latent variable follows a Gaussian distribution:

[0147]

[0148] Among them, μ i represents the mean of the i-th latent variable, Σ i represents the covariance matrix of the i-th latent variable, and M represents the number of latent variables;

[0149] S52. Use adaptive Monte Carlo sampling to generate the initial sample set and update the initial sample set:

[0150]

[0151] Among them, represents the sample value at the (k + 1)-th iteration, represents the sample value at the k-th iteration, δ represents the adaptive step size, represents the gradient of the latent variable, and ξ irepresents standard normal distribution noise, and D represents a multimodal dataset;

[0152] S53. Optimize the distribution of latent variables through variational inference and minimize the KL divergence:

[0153]

[0154] where L(q) represents the objective function, q(K) represents the approximate distribution of variational inference, and p(D, K) represents the joint distribution of the multimodal dataset D and the latent variable K;

[0155] S54. Combine importance sampling and variational inference, and correct the weights of each latent variable sample through the weight resampling formula:

[0156]

[0157] where ω i represents the weight of the i-th sample, p(D|z i ) represents the likelihood function, p(z i ) represents the prior distribution, q(z i |D) represents the variational posterior distribution, χ represents the adaptive correction factor, and f(σ, D) represents the weight correction function caused by environmental noise and system uncertainty;

[0158] S55. Generate the final corrected dataset based on the corrected weights:

[0159]

[0160] where S represents the corrected positioning data, k i represents the i-th latent variable, α i represents the correction factor of external auxiliary data, A i represents the i-th external data, and N represents the total number of latent variables;

[0161] S56. Through multiple iterations, the adaptive Monte Carlo variational inference algorithm continuously optimizes the latent variable distribution and combines the weighted correction mechanism to finally generate a high-precision corrected dataset and optimize the positioning accuracy.

[0162] In this embodiment, the S6 specifically includes:

[0163] S61. Obtain the original communication and positioning signal data, including target signals, environmental noise, and electromagnetic interference:

[0164] S raw = S target + N EMI + N thermal ;

[0165] Among them, S raw represents the original signal, and S target represents the target signal, N EMI represents electromagnetic interference, and N thermal represents thermal noise;

[0166] S62. Perform preliminary processing on the original signal through an adaptive filter. The filtering parameters of the filter are dynamically adjusted according to the current environment to generate preprocessed signal data;

[0167] S63. Use a convolutional neural network to extract features from the filtered signal, identify the interference patterns and noise characteristics in the signal, and generate an interference feature matrix;

[0168] S64. Based on the extracted interference feature matrix, calculate the power spectral density of the noise and identify the influence of different noise sources on the signal:

[0169]

[0170] Among them, P noise represents the power spectral density of the noise, T represents the signal observation duration, and S filtered (t) represents the value of the filtered signal at time t;

[0171] S65. Dynamically adjust the frequency response of the adaptive filter according to the calculation result of the noise power spectral density:

[0172]

[0173] Among them, W (k+1) represents the parameter of the filter at the (k + 1)-th iteration, W (k) represents the parameter of the filter at the k-th iteration, ξ represents the adaptive learning rate, represents the gradient of the noise power spectral density with respect to the filter parameter;

[0174] S66. Perform online optimization on the signal processing result, gradually reduce the noise and electromagnetic interference through an adaptive mechanism, and finally output the processed target signal.

[0175] In this embodiment, the specific content of S7 includes:

[0176] S71. Initialize the status information of the terminal device, including the real-time location L t , the network load condition N t and the signal strength S t , and construct a status vector s t :

[0177] s t ={L t ,N t,S t};

[0178] S72. Generate the optimal action strategy in the current state through the actor network of the Advantage Actor-Critic algorithm. The action strategy is expressed as:

[0179] π(a t |s t , θ actor );

[0180] Among them, π(a t |s t ) represents the probability of taking action a t in state s t . a t represents the action vector, and θ actor represents the parameters of the actor network. The actions include adjusting the communication frequency and selecting the path;

[0181] S73. Estimate the value function in the current state through the critic network and update the action strategy. The value function is defined as:

[0182]

[0183] Among them, V(s t ) represents the value function of state s t . r t+k+1 represents the immediate reward at time t + k + 1, represents the discount factor, and θ critic represents the parameters of the critic network;

[0184] S74. Based on the feedback of the critic network, improve the strategy of the actor network. The goal is to maximize the advantage function A(s t , a t ):

[0185]

[0186] S75. According to the feedback immediate reward r t , update the parameters of the actor network through the gradient descent method. The update formula is:

[0187]

[0188] Among them, α r represents the learning rate, represents the policy gradient;

[0189] S76. Update the state vector s t+1 after each communication event, and dynamically adjust the communication parameters through the Advantage Actor-Critic algorithm to optimize the communication frequency and path in real time.

[0190] Reference Figure 2 , the Beidou satellite intelligent positioning and communication system based on artificial intelligence includes the following modules:

[0191] Satellite signal receiving module, which is used to receive satellite signals of the Beidou satellite navigation system and at the same time receive collaborative signals to achieve collaborative positioning of multi-satellite systems;

[0192] Pseudorange measurement and carrier phase processing module, which is used to process the received satellite signals through pseudorange measurement and carrier phase differential technology and perform preliminary positioning calculations in combination with collaborative positioning data;

[0193] AI-driven multi-modal data fusion module, which is used to receive and fuse satellite signals, inertial navigation data, geographic information system data and external auxiliary data to construct a multi-modal data set;

[0194] Dynamic environment perception module based on graph neural network, which is used to dynamically adjust the weights of different data sources in the multi-modal data set according to real-time environmental changes;

[0195] Adaptive Monte Carlo variational inference module, which is used to process the multi-modal data set, correct the positioning errors caused by multipath effects and signal interference through the adaptive Monte Carlo variational inference algorithm, and optimize the positioning accuracy;

[0196] Adaptive signal processing module, which is used to monitor the electromagnetic interference and noise characteristics in communication and positioning signals and adjust the filtering parameters and signal processing strategies in real time;

[0197] Communication optimization module based on the Advantage Actor-Critic algorithm, which dynamically optimizes the communication frequency and path according to the real-time position, network load and signal strength of the terminal device.

[0198] Embodiment 1:

[0199] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a typical area with high-rise buildings. The building height and density in this area are relatively high, and satellite signals are often reflected and blocked, belonging to a typical "urban canyon" environment. Test vehicles are driven in it to simulate the actual application scenario. During the test, the vehicles drive on main roads, narrow streets and underground passages, collect positioning and communication data in different scenarios, and conduct a comparative analysis with the traditional Beidou satellite navigation system.

[0200] In the traditional Beidou system, when a vehicle travels in a high-rise dense area, the positioning accuracy fluctuates greatly. Especially at locations where the signal is affected by reflection interference, the positioning error of the vehicle once reached 30 meters. In terms of communication, the communication delay of the traditional system increases significantly in this area. Especially during the communication peak period, there is a phenomenon of packet loss in the system, resulting in unstable communication. When facing a complex urban environment, the traditional system is difficult to effectively cope with the multipath effect and signal occlusion problems, resulting in obvious decline in both positioning accuracy and communication quality.

[0201] To solve these problems, our system receives Beidou satellite signals through a satellite signal receiving module and simultaneously receives collaborative signals from other global satellite navigation systems to achieve collaborative positioning of multi-satellite systems. Using an AI-driven multi-modal data fusion module, the system fuses satellite signals, inertial navigation data, geographic information system data, and external auxiliary data. In this high-rise dense environment, where satellite signals are affected by occlusion and reflection, the system dynamically adjusts the weights of data sources through AI, enhancing the contribution of inertial navigation data and external auxiliary data to ensure that the system can maintain high-precision positioning in the case of weak satellite signals.

[0202] Our system also applies an adaptive Monte Carlo variational inference algorithm to deeply process multi-modal data, effectively eliminating errors caused by the multipath effect and signal interference. In the same test environment, the maximum positioning error of the traditional system reached 30 meters, while the system of the present invention reduced the error to within 3 meters. Even when the vehicle enters an underground passage with extremely poor signal or a completely occluded area, the system can still maintain continuous high-precision positioning through inertial navigation data and geographic information system.

[0203] In terms of communication, the system of the present invention uses the Advantage Actor-Critic algorithm to dynamically optimize the communication frequency and path according to the real-time location, network load, and signal strength of the terminal device. During the communication peak period, the communication delay of the traditional Beidou system can reach 500 milliseconds, and the packet loss rate is as high as 10%. However, through dynamic optimization, our system reduces the communication delay to within 200 milliseconds and the packet loss rate to less than 1%. This optimization greatly improves communication stability, ensuring that the system can provide fast and reliable communication services in a high-load environment.

[0204] To further verify the system performance, we conducted a one-week continuous test. The test vehicle traveled in this complex environment daily, covering different time periods and road types. The data shows that the average positioning error of the traditional Beidou system is 18.5 meters, while the positioning error of the system of the present invention is only 2.8 meters on average, with an accuracy improvement of 85%. In terms of communication latency, the average latency of the system of the present invention during the communication peak period is 180 milliseconds, which is much lower than 450 milliseconds of the traditional system, and the latency is reduced by more than 60%. At the same time, the packet loss rate of the system of the present invention during the communication peak period always remains below 2%, while the packet loss rate of the traditional system is as high as over 7%.

[0205] Table 1 Data comparison table between the traditional system and the system of the present invention under different test scenarios

[0206]

[0207] From Table 1, we can clearly see the performance comparison between the traditional system and the Beidou satellite intelligent positioning and communication system of the present invention under different test scenarios. First of all, in the area with dense high-rise buildings, the positioning error of the traditional system is as high as 30 meters, while the positioning error of the system of the present invention is only 3 meters, and the error is reduced by 90%. This shows that through the AI-driven multi-modal data fusion and dynamic environment perception model, the present invention can effectively cope with the multipath effect and signal occlusion problems in complex urban environments and provide higher-precision positioning services.

[0208] In the signal occlusion area, the positioning error of the traditional system is 25 meters, and the system of the present invention reduces the error to 2.5 meters, showing that in an environment with weak or lost signals, the system of the present invention significantly improves the continuity and accuracy of positioning through adaptive weight adjustment and the fusion of inertial navigation and external data.

[0209] In the underground passage scenario, it is difficult for the traditional system to achieve positioning and communication, while the system of the present invention can still provide a positioning error within 5 meters relying on inertial navigation data and geographic information system. At the same time, in terms of communication, the traditional system cannot communicate in this scenario, while the communication latency of the system of the present invention remains at 220 milliseconds, demonstrating good communication adaptability.

[0210] During the communication peak period, the communication latency of the traditional system is as high as 500 milliseconds, and the packet loss rate is 10%, while the system of the present invention reduces the communication latency to 180 milliseconds and the packet loss rate to 1%, indicating that in the face of a relatively high network load, the present invention can dynamically optimize the communication frequency and path through the Advantage Actor-Critic algorithm, greatly improving the stability and efficiency of communication.

[0211] Generally speaking, the system of the present invention has been greatly improved compared with the traditional system in terms of positioning accuracy, communication delay and packet loss rate. Especially in complex environments and high-load situations, it has demonstrated excellent performance. These data fully prove the practicality and technical advantages of the present invention.

[0212] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, should be covered by the protection scope of the present invention.

Claims

1. Beidou satellite intelligent positioning and communication method based on artificial intelligence, characterized in that: The steps include: S1, receive satellite signals of Beidou satellite navigation system, and simultaneously receive coordinated signals of other satellite systems to perform coordinated positioning of multiple satellite systems; S2, using pseudo-range measurement and carrier phase differential technology to process the received satellite signals, and combining the collaborative positioning data to perform preliminary positioning calculations; S3, through the AI-driven multimodal data fusion module, the received satellite signals, inertial navigation data, geographic information system data and external auxiliary data are integrated to build a multimodal data set; S4. Use a dynamic environment perception model based on graph neural networks to dynamically adjust the weights of different data sources in the multimodal data set according to real-time environmental changes. When satellite signals are blocked, weak or lost, increase the weights of inertial navigation data and external auxiliary data. S5. Adopt the adaptive Monte Carlo variational inference algorithm to process the multimodal data set, correct the positioning error caused by multipath effect and signal interference, and optimize the positioning accuracy; S6, through the adaptive signal processing module, monitor the interference in the communication and positioning signals, identify the electromagnetic interference and noise characteristics, and adjust the filtering parameters and signal processing strategies in real time; S7. Use the dominant actor-critic algorithm to dynamically optimize the communication frequency and path according to the real-time location of the terminal device, network load and signal strength.

2. The Beidou satellite intelligent positioning and communication method based on artificial intelligence according to claim 1 is characterized in that: The other satellite systems include the Global Positioning System, the GLONASS system and the Galileo positioning system.

3. The Beidou satellite intelligent positioning and communication method based on artificial intelligence according to claim 1 is characterized in that: The S2 specifically includes: S21. Receive satellite signals, calculate and record the pseudorange of each satellite: Among them, P i represents the pseudorange between the i-th satellite and the receiver, x i ,y i and z i represents the known coordinates of the i-th satellite at the time of reception, x r ,y r and z r represents the unknown coordinates of the receiver, c represents the speed of light, Δt r represents the receiver clock error, I i represents the ionospheric delay of the i-th satellite, ε i represents the measurement noise of the i-th satellite; S22. Carrier phase observation for each satellite: Among them, Φ i represents the carrier phase observation value between the i-th satellite and the receiver, λ represents the carrier wavelength, η i represents the phase measurement noise of the i-th satellite, N i represents the integer ambiguity of the i-th satellite; S23. Construct a single-difference carrier phase observation equation by receiving multiple satellite signals: Among them, ΔΦ ij represents the single difference carrier phase observation value between the i-th satellite and the j-th satellite, N j represents the integer ambiguity of the jth satellite; S24. Construct a double-difference observation model through single-difference observations: D 2 F ijk =DF ij -DF ik ; Among them, ΔΦ ik represents the single difference carrier phase observation between the i-th satellite and the k-th satellite, Δ 2 Φ ijk represents the double difference observation value between the i-th satellite, the j-th satellite and the k-th satellite; S25, combining the pseudorange and carrier phase double difference observations, constructing a nonlinear equation set, using the extended Kalman filter or adaptive Kalman filter algorithm to solve the receiver's preliminary position coordinates (x r ,y r ,z r ) and receiver clock deviation Δt r .

4. The Beidou satellite intelligent positioning and communication method based on artificial intelligence according to claim 1 is characterized in that: The S4 specifically includes: S41, obtaining a multimodal data set D = {S, I, G, A}, where S represents satellite signal data, I represents inertial navigation data, G represents geographic information system data, and A represents external auxiliary data; S42. Construct a graph structure G = (V, E) for the acquired multimodal data set, where V represents the node set of the multimodal data source, E represents the spatiotemporal dependency relationship between the data sources, and the edge weight of the graph is: W uv =f(d(u,v),t(u,v))·g(σ u ,s v ,i); Among them, W uv represents the edge weight from node u to node v, d(u,v) represents the spatial distance between node u and node v, t(u,v) represents the temporal correlation between node u and node v, σ u represents the environmental state of node u, σ v represents the environmental state of node v, θ represents the environmental adaptive adjustment coefficient, function f represents the spatial-temporal coupling relationship between nodes, and function g represents the correlation between different data sources under the environmental state; S43. On the constructed graph structure, a graph neural network is used to update the node features. The feature of each node is represented as a vector Through the message passing mechanism, the node feature update formula is: in, represents the feature vector of node v in the l+1th layer, N(v) represents the set of neighbor nodes of node v, represents the edge weight from node u to node v in the lth layer, represents the bias vector, σ represents the activation function; S44. Model the environmental perception process and use the graph neural network model to dynamically adjust the weights of different data sources according to the real-time environmental status: Among them, w i represents the i-th data source, β i represents the adaptive weight factor of the i-th data source, β j represents the adaptive weight factor of the jth data source, τ represents the environment state vector, and α i represents the weight adjustment parameter of the i-th data source, α j represents the weight adjustment parameter of the j-th data source; S45. Through multi-layer iterative optimization of the graph neural network, the final data source weighted fusion model is generated, and the information of multiple data sources is integrated to generate the fused feature vector F: Among them, D i represents the original data of the i-th data source, γ uv represents the coupling strength of the edge connecting node u and node v, h u represents the feature vector of node u, h v Represents the feature vector of node v.

5. The Beidou satellite intelligent positioning and communication method based on artificial intelligence according to claim 1 is characterized in that: The S5 specifically includes: S51. Define a latent variable K, which follows a Gaussian distribution: Among them, μ i represents the mean of the ith latent variable, Σ i represents the covariance matrix of the ith latent variable, and M represents the number of latent variables; S52, using adaptive Monte Carlo sampling to generate an initial sample set, and update the initial sample set: in, represents the sample value at the k+1th iteration, represents the sample value at the kth iteration, δ represents the adaptive step size, represents the gradient of the latent variable, ξ i represents standard normal distribution noise, D represents a multimodal dataset; S53. Optimize the distribution of latent variables through variational inference and minimize KL divergence: Where L(q) represents the objective function, q(K) represents the approximate distribution of variational inference, and p(D,K) represents the joint distribution of the multimodal dataset D and the latent variable K; S54. Combining importance sampling and variational inference, the weight of each latent variable sample is corrected through the weight resampling formula: Among them, ω i represents the weight of the i-th sample, p(D|z i ) represents the likelihood function, p(z i ) represents the prior distribution, q(z i |D) represents the variational posterior distribution, χ represents the adaptive correction factor, and f(σ,D) represents the weight correction function caused by environmental noise and system uncertainty; S55. Based on the corrected weights, generate the final calibration data set: Among them, S represents the corrected positioning data, k i represents the i-th latent variable, α i represents the correction factor of external auxiliary data, A i represents the i-th external data, and N represents the total number of latent variables; S56. Through multiple iterations, the adaptive Monte Carlo variational inference algorithm continuously optimizes the distribution of latent variables, and combines with the weighted correction mechanism to ultimately generate a high-precision correction data set to optimize positioning accuracy.

6. The Beidou satellite intelligent positioning and communication method based on artificial intelligence according to claim 1 is characterized in that: The S6 specifically includes: S61. Obtaining original communication and positioning signal data, including target signals, environmental noise, and electromagnetic interference: S raw =S target +N EMI +N thermal ; Among them, S raw represents the original signal, S target represents the target signal, N EMI Indicates electromagnetic interference, N thermal represents thermal noise; S62, preliminarily processing the original signal through an adaptive filter, dynamically adjusting the filter parameters according to the current environment, and generating preprocessed signal data; S63, using a convolutional neural network to extract features from the filtered signal, identify interference patterns and noise features in the signal, and generate an interference feature matrix; S64. Based on the extracted interference feature matrix, the power spectrum density of the noise is calculated to identify the impact of different noise sources on the signal: Among them, P noise represents the power spectral density of the noise, T represents the signal observation time, S filtered (t) represents the value of the filtered signal at time t; S65. Dynamically adjust the frequency response of the adaptive filter according to the calculation result of the noise power spectrum density: Among them, W (k+1) represents the parameters of the filter at the k+1th iteration, W (k) represents the parameters of the filter at the kth iteration, ξ represents the adaptive learning rate, represents the gradient of the noise power spectral density with respect to the filter parameters; S66. Optimize the signal processing results online, gradually reduce noise and electromagnetic interference through an adaptive mechanism, and finally output the processed target signal.

7. The Beidou satellite intelligent positioning and communication method based on artificial intelligence according to claim 1 is characterized in that: The S7 specifically includes: S71. Initialize the status information of the terminal device, including the real-time location of the device. t 、Network load condition N t and signal strength S t , construct the state vector s t : s t ={L t ,N t ,S t }; S72. Generate the optimal action strategy under the current state through the actor network of the dominant actor-critic algorithm. The action strategy is expressed as: π(a t |s t ,i actor ); Among them, π(a t |s t ) means in state s t Take action t The probability of t represents the action vector, θ actor Parameters representing the actor network, actions include the adjustment of communication frequency and the choice of path; S73. Estimate the value function in the current state through the critic network and update the action strategy. The value function is defined as: Among them, V(s t ) indicates state s t The value function, r t+k+1 represents the immediate return at time t+k+1, represents the discount factor, θ critic represents the parameters of the critic network; S74. Based on the feedback from the critic network, the strategy of the actor network is improved, with the goal of maximizing the advantage function A(s t ,a t ): S75. Instant rewards based on feedback t , update the parameters of the actor network by gradient descent method, and the update formula is: Among them, α r represents the learning rate, represents the policy gradient; S76. Update the state vector s after each communication event t+1 , and dynamically adjusts communication parameters through the dominant actor-critic algorithm to optimize communication frequency and path in real time.

8. The Beidou satellite intelligent positioning and communication system based on artificial intelligence executes the Beidou satellite intelligent positioning and communication method based on artificial intelligence according to any one of claims 1 to 7, characterized in that: Includes the following modules: Satellite signal receiving module, used to receive satellite signals of Beidou satellite navigation system and receive coordinated signals at the same time to realize coordinated positioning of multiple satellite systems; The pseudo-range measurement and carrier phase processing module is used to process the received satellite signals through pseudo-range measurement and carrier phase difference technology, and perform preliminary positioning calculation in combination with the collaborative positioning data; AI-driven multimodal data fusion module, which receives and fuses satellite signals, inertial navigation data, geographic information system data, and external auxiliary data to construct a multimodal data set; A dynamic environment perception module based on graph neural networks, which is used to dynamically adjust the weights of different data sources in a multimodal dataset according to real-time environmental changes; The adaptive Monte Carlo variational inference module is used to process multimodal data sets and correct the positioning errors caused by multipath effects and signal interference through the adaptive Monte Carlo variational inference algorithm to optimize the positioning accuracy. Adaptive signal processing module, used to monitor electromagnetic interference and noise characteristics in communication and positioning signals, and adjust filtering parameters and signal processing strategies in real time; The communication optimization module based on the dominant actor-critic algorithm dynamically optimizes the communication frequency and path according to the real-time location of the terminal device, network load and signal strength.

Images