GPS positioning correction method and system

Abstract

The invention relates to the technical field of GPS positioning systems, and discloses a GPS positioning correction method and system, and the method comprises the following steps: S1, carrying out the fusion collection of multi-source heterogeneous data, and obtaining the motion state data of a carrier in a variable bandwidth sampling mode, S2, carrying out the deep feature extraction and preprocessing, and carrying out the extraction of a spatial-temporal feature extraction network based on deep learning, the method comprises the following steps: S1, mining potential features of data in time series and spatial distribution, S3, dynamic environment perception and error modeling, and S4, intelligent collaborative error correction. According to the invention, through the established deep Q network, multi-source data characteristics including GPS positioning data, inertial measurement unit data, air pressure and meteorological data and the like are used as input, and environment information can be comprehensively captured. Through deep mining and analysis of the data, the model can more accurately identify the type of the environment where the carrier is located, and meanwhile, by utilizing a reinforcement learning mechanism and setting a reward function, the model is encouraged to accurately identify the type of the environment and evaluate the interference degree, and the adaptability of an error model can be remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of GPS positioning systems, and in particular to a GPS positioning correction method and system. Background Art

[0002] GPS, or the Global Positioning System, is a satellite-based radio navigation and positioning system. It consists of three segments: the space segment, which is composed of multiple satellites orbiting the Earth and continuously transmitting signals containing information such as their location and time to Earth. The ground segment monitors and controls the satellites, ensuring their proper functioning and data accuracy. The user segment consists of various devices equipped with GPS receivers, such as mobile phones and car navigation systems.

[0003] In existing technologies, after a receiver receives signals from at least four satellites, it measures the signal transmission time, combines it with satellite position information, and uses the principle of triangulation to accurately calculate its own three-dimensional position on the earth (longitude, latitude, and altitude), while also measuring its speed and direction of movement. GPS positioning is widely used in many fields such as traffic navigation, logistics tracking, outdoor sports, and surveying and mapping, greatly facilitating people's lives and work.

[0004] However, in the actual application of GPS equipment, the existing technology lacks effective quantitative means to evaluate the degree of environmental interference on GPS positioning. In most cases, it is only qualitatively considered that certain environments will cause interference, but it is impossible to accurately measure the intensity of the interference and the specific impact on positioning error. This makes it difficult for the error model to accurately characterize the error characteristics under different environments. In addition, the geographical environment and climatic conditions in different regions vary greatly. The existing general error model is difficult to take into account various complex situations. In view of this, we propose a correction method and system for GPS positioning. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a GPS positioning correction method and system, which solves the problem that the intensity of interference and its specific impact on positioning error cannot be accurately measured, making it difficult for the error model to accurately characterize the error characteristics under different environments.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a GPS positioning correction method and system, comprising the following steps:

[0007] S1: Multi-source heterogeneous data fusion collection

[0008] An inertial measurement unit with magnetic field interference self-compensation function is used to obtain carrier motion state data using a variable bandwidth sampling method, while a GPS receiver equipped with an adaptive sampling frequency mechanism is used to collect raw GPS positioning data.

[0009] S2: Deep feature extraction and preprocessing

[0010] For GPS positioning data, a deep learning-based spatiotemporal feature extraction network is used to mine the data's potential features in time series and spatial distribution. Wavelet transform is also used for multi-scale noise reduction. Furthermore, for inertial measurement unit data, a rotation-invariant feature extraction algorithm is used to eliminate measurement deviations caused by changes in the carrier's posture. The data is then optimized using a particle filter algorithm.

[0011] S3: Dynamic Environment Perception and Error Modeling

[0012] Build an environmental perception model based on reinforcement learning. This model uses multi-source data features as input, continuously interacts with the environment to learn, and identifies the type of environment the carrier is in in real time. It also assesses the degree of interference the environment has on GPS positioning. At the same time, based on the environmental perception results, it combines a Bayesian network and a deep learning hybrid model to dynamically establish a GPS positioning error model.

[0013] S4: Intelligent Collaborative Error Correction

[0014] A quantum computing acceleration mechanism is introduced to perform parallel processing on multi-source data collected in real time, and quickly input it into the dynamic error model for error prediction. At the same time, a multi-agent collaborative correction strategy is adopted based on the predicted error value.

[0015] Preferably, the adaptive sampling frequency mechanism established in the S1 multi-source heterogeneous data fusion acquisition predicts the motion state of the carrier based on the long short-term memory network, adjusts the sampling frequency of the GPS receiver in advance, balances the data volume and positioning accuracy requirements, samples at a lower frequency when stationary or at low speed, and increases to a higher frequency when moving at high speed.

[0016] Preferably, the inertial measurement unit with magnetic field interference self-compensation function in the S1 multi-source heterogeneous data fusion acquisition detects magnetic field interference in real time through a built-in magnetic anomaly detection algorithm, automatically adjusts the sampling bandwidth in a complex electromagnetic environment, and uses a pre-trained neural network model to generate a compensation coefficient to correct the magnetic field intensity data.

[0017] Preferably, in the S2 deep feature extraction and preprocessing of air pressure and meteorological data, a data fusion algorithm is used to combine air pressure data with meteorological trend information to construct an atmospheric environment feature vector, and dimensionality reduction is performed through principal component analysis to remove redundant information and retain key features.

[0018] Preferably, the environment perception model based on reinforcement learning in the S3 dynamic environment perception and error modeling adopts a deep Q network structure. Through the set reward function, the model is encouraged to accurately identify the environment type and evaluate the degree of interference, thereby improving the adaptability of the error model.

[0019] Preferably, the quantum computing acceleration mechanism in the S4 intelligent collaborative error correction utilizes the parallelism of quantum bits to perform quantum encoding and parallel operations on multi-source data, establishes matrix multiplication and convolution operations through quantum gate operations, and accelerates the calculation process of the error model.

[0020] Preferably, in the multi-agent collaborative correction strategy in the S4 intelligent collaborative error correction, each agent uses a distributed consensus algorithm to interact with information, ensuring consensus between different error correction tasks and avoiding correction conflicts.

[0021] A GPS positioning correction system includes the following modules:

[0022] Multi-source data acquisition module:

[0023] Smart GPS receiver: With adaptive sampling frequency function, it dynamically adjusts the sampling frequency according to the carrier's motion state, and collects data including longitude, latitude, altitude and nanosecond positioning timestamp data;

[0024] Self-compensating inertial measurement unit: Equipped with a magnetic field interference self-compensation device, it can detect and correct the impact of magnetic field interference on acceleration, angular velocity and magnetic field strength data in real time, and use variable bandwidth sampling to ensure data accuracy;

[0025] Weather-linked air pressure sensor: interacts with the meteorological data cloud platform in real time, collecting not only air pressure data but also surrounding weather trend information for analyzing the impact of the atmospheric environment on GPS signals.

[0026] Data preprocessing and feature extraction module:

[0027] GPS data processing unit: uses deep learning spatiotemporal feature extraction network and wavelet transform to mine potential features of GPS data and reduce noise;

[0028] IMU data processing unit: uses rotation invariance feature extraction algorithm and particle filter algorithm to eliminate posture deviation and optimize data;

[0029] Meteorological data processing unit: uses data fusion and principal component analysis to construct and optimize atmospheric environment feature vectors.

[0030] Environmental perception and error modeling module:

[0031] Reinforcement Learning Environment Perception Unit: Based on the deep Q-network structure of reinforcement learning, it can identify the environment type in real time and evaluate the degree of interference to GPS positioning;

[0032] Hybrid Error Modeling Unit: Combines Bayesian networks and deep learning models to dynamically adjust parameters according to environmental changes and establish an accurate error model.

[0033] Error correction and output module:

[0034] Quantum acceleration computing unit: Utilizes quantum computing acceleration mechanism to quickly predict error values ​​through parallel operation of quantum bits and quantum gate operations;

[0035] Multi-agent collaborative correction unit: adopts a multi-agent collaborative strategy, with each agent responsible for error correction in a specific dimension. They work together through a distributed consensus algorithm to output high-precision positioning results.

[0036] Preferably, the intelligent GPS receiver, self-compensating inertial measurement unit and meteorological linkage pressure sensor in the data preprocessing and feature extraction module are connected through a dedicated data bus. Each hardware device has a built-in micro-control chip, and a collaborative work protocol is pre-stored in the chip. When the self-compensating inertial measurement unit detects that the change in the carrier's motion state exceeds a preset threshold, its control chip immediately sends a command signal for adjusting the sampling frequency to the control chip of the intelligent GPS receiver according to the protocol. The command signal includes the type of motion state change and recommended sampling frequency information.

[0037] Preferably, the system adopts a reconfigurable modular algorithm design, and the system is provided with an intelligent algorithm selection module. This module uses a fuzzy logic reasoning algorithm to evaluate the applicability of each algorithm module combination in different application scenarios based on the data characteristics collected in real time and the environmental information fed back by the environmental perception module, and controls the switching combination of different algorithm modules in different environments.

[0038] The present invention provides a GPS positioning correction method and system, which has the following beneficial effects:

[0039] 1. This invention establishes a deep Q-network that uses multi-source data features as input, including GPS positioning data, inertial measurement unit data, air pressure, and meteorological data, to comprehensively capture environmental information. By deeply mining and analyzing this data, the model can more accurately identify the type of environment the carrier is in. At the same time, using reinforcement learning mechanisms and setting a reward function to encourage the model to accurately identify the environment type and assess the degree of interference, the adaptability of the error model can be significantly enhanced.

[0040] 2. The present invention establishes an adaptive sampling frequency mechanism that predicts the carrier's motion state based on a long short-term memory (LSTM) network, enabling real-time adjustment of the sampling frequency to accommodate changes in the carrier's motion state. This mechanism dynamically adjusts the sampling interval based on parameters such as the carrier's velocity and acceleration, thereby ensuring data acquisition accuracy while reducing unnecessary data processing and improving overall system efficiency. Furthermore, the introduction of the LSTM network enables the system to learn and memorize long-term dependencies in carrier motion, further enhancing prediction accuracy and system robustness.

[0041] 3. The present invention establishes an intelligent algorithm selection module, which can use fuzzy logic reasoning algorithm to evaluate the applicability of each algorithm module combination in different application scenarios based on the real-time collected data features and the environmental information feedback from the environmental perception module, and control the switching combination of different algorithm modules in different environments, thereby achieving optimal positioning accuracy and system response speed, ensuring that stable and reliable positioning services can be provided in various complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of the correction method for GPS positioning;

[0043] Figure 2 This is the S3 dynamic environment perception and error modeling flow chart of the present invention;

[0044] Figure 3 This is the module diagram of the GPS positioning correction system;

[0045] Figure 4 Schematic diagram of a multi-source data acquisition module of the present invention;

[0046] Figure 5 Schematic diagram of the data preprocessing and feature extraction module of the present invention;

[0047] Figure 6 This is a schematic diagram of the environment perception and error modeling module of the present invention;

[0048] Figure 7 Schematic diagram of the error correction and output module of the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the specification of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] Example:

[0051] Please see the attached Figure 1 - Attachment Figure 2 The present invention provides a method for correcting GPS positioning, comprising the following steps:

[0052] S1: Multi-source heterogeneous data fusion collection

[0053] An inertial measurement unit with magnetic field interference self-compensation function is used to obtain carrier motion state data using a variable bandwidth sampling method, while a GPS receiver equipped with an adaptive sampling frequency mechanism is used to collect raw GPS positioning data.

[0054] S2: Deep feature extraction and preprocessing

[0055] For GPS positioning data, a deep learning-based spatiotemporal feature extraction network is used to mine the data's potential features in time series and spatial distribution. Wavelet transform is also used for multi-scale noise reduction. Furthermore, for inertial measurement unit data, a rotation-invariant feature extraction algorithm is used to eliminate measurement deviations caused by changes in the carrier's posture. The data is then optimized using a particle filter algorithm.

[0056] S3: Dynamic Environment Perception and Error Modeling

[0057] Build an environmental perception model based on reinforcement learning. This model uses multi-source data features as input, continuously interacts with the environment to learn, and identifies the type of environment the carrier is in in real time. It also assesses the degree of interference the environment has on GPS positioning. At the same time, based on the environmental perception results, it combines a Bayesian network and a deep learning hybrid model to dynamically establish a GPS positioning error model.

[0058] S4: Intelligent Collaborative Error Correction

[0059] A quantum computing acceleration mechanism is introduced to perform parallel processing on multi-source data collected in real time, and quickly input it into the dynamic error model for error prediction. At the same time, a multi-agent collaborative correction strategy is adopted based on the predicted error value.

[0060] The adaptive sampling frequency mechanism established in the S1 multi-source heterogeneous data fusion acquisition predicts the carrier motion state based on the long short-term memory network, adjusts the GPS receiver sampling frequency in advance, and balances the data volume and positioning accuracy requirements. It samples at a lower frequency when stationary or at low speed, and increases to a higher frequency when moving at high speed. The memory network algorithm established here is as follows:

[0061] Input Gate:

[0062] i t =σ(W ii x t +b ii +W hi h t-1 +b hi )

[0063] Forget Gate:

[0064] f t =σ(W if x t +b if +W hf h t-1 +b hf )

[0065] Output Gate:

[0066] o t =σ(W io x t +b io +W ho h t-1 +b ho )

[0067] Memory cell renewal:

[0068]

[0069] Hide status update:

[0070] h t =o t ⊙tanh ( C t

[0071] Among them, x t is the input at the current moment;

[0072] h t-1 is the hidden state at the previous moment;

[0073] C t-1 It is the state of the memory cells at the previous moment;

[0074] σ is the Si gmo id function;

[0075] tanh is the hyperbolic tangent function;

[0076] G represents element-by-element multiplication;

[0077] W and b are the weight matrix and bias vector respectively;

[0078] Example: Suppose you want to predict the change in the carrier's speed over the next period of time to determine the sampling frequency of the GPS receiver.

[0079] Input data x t It can be the acceleration, position and other information of the carrier at the current moment. Through the above calculation steps of the LSTM network, the network learns the change pattern of the carrier's motion state over time, predicts the future motion state, and thus adjusts the sampling frequency of the GPS receiver.

[0080] The inertial measurement unit with magnetic field interference self-compensation function in the S1 multi-source heterogeneous data fusion acquisition uses a built-in magnetic anomaly detection algorithm to detect magnetic field interference in real time. In complex electromagnetic environments, it automatically adjusts the sampling bandwidth and uses a pre-trained neural network model to generate compensation coefficients to correct the magnetic field intensity data.

[0081] In the S2 deep feature extraction and preprocessing of air pressure and meteorological data, a data fusion algorithm is used to combine air pressure data with meteorological trend information to construct an atmospheric environment feature vector, and a dimensionality reduction process is performed through principal component analysis to remove redundant information and retain key features.

[0082] The reinforcement learning-based environment perception model in the S3 dynamic environment perception and error modeling adopts a deep Q network structure. Through the set reward function, the model is encouraged to accurately identify the environment type and evaluate the interference level, thereby improving the adaptability of the error model. The network structure and operation steps established here are as follows:

[0083] initialization:

[0084] Initialize the parameters θ and θ of the Q network (main network) and the target Q network - , initialize both parameters to the same value;

[0085] Initialize the experience replay memory D to store the experience samples s,a,r,s of the interaction between the agent and the environment ′ ) , where s is the current state, a is the action taken, r is the reward obtained, and s ′ It is the next state.

[0086] Environmental interaction and experience storage

[0087] The agent is based on the current state s t , select action a through ∈-greedy strategy t , the ∈-greedy strategy randomly selects an action with a probability of ∈, and selects the action with the maximum Q value estimated by the current Q network with a probability of 1-∈. This can strike a balance between exploring new actions and utilizing existing experience;

[0088] The agent performs action a t , interact with the environment and obtain reward r t and the new state s t +1;

[0089] The experience sample ( s t ,a t ,r t ,s t+1) Stored in the experience replay memory library D.

[0090] Experience replay and training

[0091] Randomly sample a batch of experience samples from the experience replay memory bank D

[0092] For each sample, calculate the target Q value y i :

[0093]

[0094] Where γ is a discount factor, ranging from [0, 1], which is used to balance the importance of current rewards and future rewards. The closer γ is to 1, the more emphasis is placed on future rewards; the closer γ is to 0, the more attention is paid to current rewards.

[0095] Calculate the loss function of the main network:

[0096] L i (θ)=(y i -Q(s i , a i :θ)) 2

[0097] Through optimization algorithms such as stochastic gradient descent, the gradient of the loss function with respect to the main network parameters θ is calculated, and the main network parameters are updated to minimize the loss function.

[0098] Target Network Update

[0099] After a certain number of training steps, the parameters θ of the main network are copied to the target network, i.e. θ - ←θ. The purpose of this is to keep the target Q value relatively stable over a period of time, avoid oscillation and instability during training, and help improve the stability and convergence of training;

[0100] Through this training process, DQN is able to learn the optimal mapping relationship between different states (combinations of multi-source data features) and actions (environmental type judgment), that is, the optimal strategy. In practical applications, when the GPS positioning correction system obtains real-time multi-source data, the DQN model can quickly and accurately identify the type of environment the carrier is in based on the learned strategy and assess the degree of environmental interference with GPS positioning. This information is crucial for the subsequent dynamic establishment of a GPS positioning error model by combining a Bayesian network and a deep learning hybrid model. This allows the error model to better adapt to the GPS positioning error characteristics in different environments, thereby improving the accuracy and reliability of GPS positioning corrections.

[0101] The quantum computing acceleration mechanism in the S4 intelligent collaborative error correction utilizes the parallelism of quantum bits to perform quantum encoding and parallel operations on multi-source data, establishes matrix multiplication and convolution operations through quantum gate operations, and accelerates the calculation process of the error model.

[0102] In the multi-agent collaborative correction strategy of the S4 intelligent collaborative error correction, each agent uses a distributed consensus algorithm to exchange information, ensuring consensus between different error correction tasks and avoiding correction conflicts.

[0103] Please see the attached Figure 3 - Attachment Figure 7 , a GPS positioning correction system, including the following modules:

[0104] Multi-source data acquisition module:

[0105] Smart GPS receiver: With adaptive sampling frequency function, it dynamically adjusts the sampling frequency according to the carrier's motion state, and collects data including longitude, latitude, altitude and nanosecond positioning timestamp data;

[0106] Self-compensating inertial measurement unit: Equipped with a magnetic field interference self-compensation device, it can detect and correct the impact of magnetic field interference on acceleration, angular velocity and magnetic field strength data in real time, and use variable bandwidth sampling to ensure data accuracy;

[0107] Weather-linked air pressure sensor: interacts with the meteorological data cloud platform in real time, collecting not only air pressure data but also surrounding weather trend information for analyzing the impact of the atmospheric environment on GPS signals.

[0108] Data preprocessing and feature extraction module:

[0109] GPS data processing unit: uses deep learning spatiotemporal feature extraction network and wavelet transform to mine potential features of GPS data and reduce noise;

[0110] IMU data processing unit: uses rotation invariance feature extraction algorithm and particle filter algorithm to eliminate posture deviation and optimize data;

[0111] Meteorological data processing unit: uses data fusion and principal component analysis to construct and optimize atmospheric environment feature vectors.

[0112] Environmental perception and error modeling module:

[0113] Reinforcement Learning Environment Perception Unit: Based on the deep Q-network structure of reinforcement learning, it can identify the environment type in real time and evaluate the degree of interference to GPS positioning;

[0114] Hybrid Error Modeling Unit: Combines Bayesian networks and deep learning models to dynamically adjust parameters according to environmental changes and establish an accurate error model.

[0115] Error correction and output module:

[0116] Quantum acceleration computing unit: Utilizes quantum computing acceleration mechanism to quickly predict error values ​​through parallel operation of quantum bits and quantum gate operations;

[0117] Multi-agent collaborative correction unit: adopts a multi-agent collaborative strategy, with each agent responsible for error correction in a specific dimension. They work together through a distributed consensus algorithm to output high-precision positioning results.

[0118] The intelligent GPS receiver, self-compensating inertial measurement unit, and weather-linked pressure sensor in the data preprocessing and feature extraction module are connected via a dedicated data bus. Each hardware device has a built-in microcontroller chip with a pre-stored collaborative working protocol. When the self-compensating inertial measurement unit detects that the change in the carrier's motion state exceeds a preset threshold, its control chip immediately sends a command signal for adjusting the sampling frequency to the control chip of the intelligent GPS receiver according to the protocol. The command signal contains information about the type of motion state change and the recommended sampling frequency.

[0119] The system adopts a reconfigurable modular algorithm design. The system is equipped with an intelligent algorithm selection module. Based on the real-time collected data features and the environmental information fed back by the environmental perception module, this module uses a fuzzy logic reasoning algorithm to evaluate the applicability of each algorithm module combination in different application scenarios and control the switching combination of different algorithm modules in different environments. The following is the reconfigurable modular algorithm design in this technical solution:

[0120] Fuzzification: Convert the precise value of the input (such as data features, environmental information, etc.) into the membership in the fuzzy set. Assume that the input variable is x and its domain is U. For the fuzzy set A, the membership function μ A( x ) Indicates the degree to which x belongs to A;

[0121] function:

[0122]

[0123] Where c is the center of the function and σ is the standard deviation;

[0124] Rule base: consists of a series of fuzzy rules in the form of "If x1 is A1 and x2 is A2 and ... then y is B ” ,For example, the rule "If the GPS signal strength is weak and the surrounding building density is high, then it is appropriate to select an algorithm module combination targeting multipath effects";

[0125] Fuzzy reasoning: Based on the input fuzzy set and rule base, the fuzzy output is obtained through reasoning method. For the rule "If x is A then y is B ″ , given input x = x0, its corresponding fuzzy output B ′ Determined by the following formula:

[0126]

[0127] Where ∧ represents the smaller operation, and sup represents the supremum operation;

[0128] Defuzzification: Convert fuzzy output to exact value:

[0129]

[0130] If it is a discrete case, it is:

[0131]

[0132] Example:

[0133] Assume that the inputs to the intelligent algorithm selection module are GPS signal strength x1 and environment type (represented by building density) x2. For GPS signal strength, define the fuzzy sets "strong," "medium," and "weak," and their membership functions are Gaussian functions, such as the membership function for "weak":

[0134]

[0135] (Assume that the signal strength range is 0-100, c = 20, σ = 10) For the building density x2, define the fuzzy set "low", "medium", and "high", and also use the Gaussian membership function;

[0136] There is a rule in the rule base: "If x1 is "weak" and x2 is "high", then the algorithm module combination C is suitable 1‘’ Assume that the calculated membership of x1 to "weak" is 0.8, and the membership of x2 to "high" is 0.7. According to fuzzy reasoning (taking the smaller operation), the intensity of triggering the rule is 0.7.

[0137] Assume that multiple rules are triggered and the fuzzy output B is obtained ′ , defuzzified by the centroid method. For example, in the discrete case, there are three possible algorithm module combinations y1, y2, y3, and their corresponding membership degrees are μ B′ ( y1 ) =0.6,μ B′ ( y2 ) =0.8,μ B′ ( y3 ) =0.5, then the final selected algorithm module combination is:

[0138]

[0139] Assume that y1 corresponds to the algorithm module combination C1, y2 corresponds to C2, and y3 corresponds to C3. After calculation, the most suitable algorithm module combination is obtained.

[0140] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A GPS positioning correction method, characterized in that: The following steps are involved: S1: Multi-source heterogeneous data fusion collection An inertial measurement unit with magnetic field interference self-compensation function is used to obtain carrier motion state data using a variable bandwidth sampling method, while a GPS receiver equipped with an adaptive sampling frequency mechanism is used to collect raw GPS positioning data. S2: Deep feature extraction and preprocessing For GPS positioning data, a deep learning-based spatiotemporal feature extraction network is used to mine the data's potential features in time series and spatial distribution. Wavelet transform is also used for multi-scale noise reduction. Furthermore, for inertial measurement unit data, a rotation-invariant feature extraction algorithm is used to eliminate measurement deviations caused by changes in the carrier's posture. The data is then optimized using a particle filter algorithm. S3: Dynamic Environment Perception and Error Modeling Build an environmental perception model based on reinforcement learning. This model uses multi-source data features as input, continuously interacts with the environment to learn, and identifies the type of environment the carrier is in in real time. It also assesses the degree of interference the environment has on GPS positioning. At the same time, based on the environmental perception results, it combines a Bayesian network and a deep learning hybrid model to dynamically establish a GPS positioning error model. S4: Intelligent Collaborative Error Correction A quantum computing acceleration mechanism is introduced to perform parallel processing on multi-source data collected in real time, and quickly input it into the dynamic error model for error prediction. At the same time, a multi-agent collaborative correction strategy is adopted based on the predicted error value.

2. A GPS positioning correction method according to claim 1, characterized in that: The adaptive sampling frequency mechanism established in the S1 multi-source heterogeneous data fusion acquisition predicts the carrier motion state based on the long short-term memory network, adjusts the GPS receiver sampling frequency in advance, balances the data volume and positioning accuracy requirements, and samples at a lower frequency when stationary or at low speed, and increases to a higher frequency when moving at high speed.

3. The GPS positioning correction method according to claim 1, characterized in that: The inertial measurement unit with magnetic field interference self-compensation function in the S1 multi-source heterogeneous data fusion acquisition uses a built-in magnetic anomaly detection algorithm to detect magnetic field interference in real time. In complex electromagnetic environments, it automatically adjusts the sampling bandwidth and uses a pre-trained neural network model to generate compensation coefficients to correct the magnetic field intensity data.

4. A GPS positioning correction method according to claim 1, characterized in that: In the S2 deep feature extraction and preprocessing of air pressure and meteorological data, a data fusion algorithm is used to combine air pressure data with meteorological trend information to construct an atmospheric environment feature vector, and a dimensionality reduction process is performed through principal component analysis to remove redundant information and retain key features.

5. The GPS positioning correction method according to claim 1, characterized in that: The reinforcement learning-based environment perception model in the S3 dynamic environment perception and error modeling adopts a deep Q-network structure. Through the set reward function, the model is encouraged to accurately identify the environment type and evaluate the interference degree, thereby improving the adaptability of the error model.

6. A GPS positioning correction method according to claim 1, characterized in that: The quantum computing acceleration mechanism in the S4 intelligent collaborative error correction utilizes the parallelism of quantum bits to perform quantum encoding and parallel operations on multi-source data, establishes matrix multiplication and convolution operations through quantum gate operations, and accelerates the calculation process of the error model.

7. A GPS positioning correction method according to claim 1, characterized in that: In the multi-agent collaborative correction strategy of the S4 intelligent collaborative error correction, each agent uses a distributed consensus algorithm to exchange information, ensuring consensus between different error correction tasks and avoiding correction conflicts.

8. A GPS positioning correction system according to claim 1, characterized in that: Includes the following modules: Multi-source data acquisition module: Smart GPS receiver: With adaptive sampling frequency function, it dynamically adjusts the sampling frequency according to the carrier's motion state, and collects data including longitude, latitude, altitude and nanosecond positioning timestamp data; Self-compensating inertial measurement unit: Equipped with a magnetic field interference self-compensation device, it can detect and correct the impact of magnetic field interference on acceleration, angular velocity and magnetic field strength data in real time, and use variable bandwidth sampling to ensure data accuracy; Weather-linked air pressure sensor: interacts with the meteorological data cloud platform in real time, collecting not only air pressure data but also surrounding weather trend information for analyzing the impact of the atmospheric environment on GPS signals. Data preprocessing and feature extraction module: GPS data processing unit: uses deep learning spatiotemporal feature extraction network and wavelet transform to mine potential features of GPS data and reduce noise; IMU data processing unit: uses rotation invariance feature extraction algorithm and particle filter algorithm to eliminate posture deviation and optimize data; Meteorological data processing unit: uses data fusion and principal component analysis to construct and optimize atmospheric environment feature vectors. Environmental perception and error modeling module: Reinforcement Learning Environment Perception Unit: Based on the deep Q-network structure of reinforcement learning, it can identify the environment type in real time and evaluate the degree of interference to GPS positioning; Hybrid Error Modeling Unit: Combines Bayesian networks and deep learning models to dynamically adjust parameters according to environmental changes and establish an accurate error model. Error correction and output module: Quantum acceleration computing unit: Utilizes quantum computing acceleration mechanism to quickly predict error values ​​through parallel operation of quantum bits and quantum gate operations; Multi-agent collaborative correction unit: adopts a multi-agent collaborative strategy, with each agent responsible for error correction in a specific dimension. They work together through a distributed consensus algorithm to output high-precision positioning results.

9. A GPS positioning correction system according to claim 8, characterized in that: The intelligent GPS receiver, self-compensating inertial measurement unit, and weather-linked pressure sensor in the data preprocessing and feature extraction module are connected via a dedicated data bus. Each hardware device has a built-in microcontroller chip with a pre-stored collaborative working protocol. When the self-compensating inertial measurement unit detects that the change in the carrier's motion state exceeds a preset threshold, its control chip immediately sends a command signal for adjusting the sampling frequency to the control chip of the intelligent GPS receiver according to the protocol. The command signal contains information about the type of motion state change and the recommended sampling frequency.

10. A GPS positioning correction system according to claim 8, characterized in that: The system adopts a reconfigurable modular algorithm design. The system is equipped with an intelligent algorithm selection module. This module uses fuzzy logic reasoning algorithm to evaluate the applicability of various algorithm module combinations in different application scenarios based on the real-time collected data features and environmental information feedback from the environmental perception module, and controls the switching combination of different algorithm modules in different environments.

Images