A communication detection system and method based on BeiDou positioning

Abstract

This invention discloses a communication detection system and method based on BeiDou positioning, belonging to the field of satellite communication technology. Specifically, it includes the following steps: S1, Multi-source data acquisition and preprocessing: Simultaneous acquisition of multi-dimensional data, followed by multi-level preprocessing of the acquired multi-source data; S2, Signal feature identification and interference identification: Extracting time-frequency domain features of the signal, constructing a hybrid model for deep learning interference identification, and adaptively suppressing interference. By simultaneously acquiring BeiDou positioning data, sensor data, and communication status parameters from multiple dimensions, a comprehensive perception of the communication system's operating status is achieved, breaking through the limitations of traditional methods that rely solely on single data points. This provides rich foundational data support for subsequent analysis, and the multi-level preprocessing process further improves data quality, providing a high-quality data foundation for subsequent interference identification and anomaly detection.

Description

Technical Field

[0001] This invention relates to the field of satellite communication technology, and in particular to a communication detection system and method based on BeiDou positioning. Background Technology

[0002] The BeiDou Navigation Satellite System has a wide range of applications, including transportation, agriculture, forest fire prevention, disaster monitoring and rescue, construction, public safety, time services, mobile communications, marine fisheries, hydrological monitoring, weather forecasting, and surveying and mapping. With continuous technological advancements and enhanced service capabilities, the BeiDou Navigation Satellite System has become one of the world's most important infrastructures. Therefore, monitoring the communication status of the BeiDou Navigation Satellite System is essential and crucial. However, traditional communication monitoring techniques simply splice positioning data with communication parameters without fully exploring data correlations, resulting in insufficient fault location accuracy. This makes it difficult to provide a reliable basis for subsequent maintenance of potential communication faults. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide a communication detection system and method based on BeiDou positioning; it can solve the problem of low multi-source data fusion rate in traditional communication detection technology.

[0004] Technical Solution: To solve the above-mentioned technical problems, according to one aspect of the present invention, more specifically, a communication detection method based on BeiDou positioning, specifically including the following steps:

[0005] S1. Multi-source data acquisition and preprocessing: Simultaneous acquisition of multi-dimensional data, and multi-level preprocessing of the acquired multi-source data;

[0006] S2. Signal Feature Recognition and Interference Recognition: Extract time-frequency domain features of the signal, construct a hybrid model for deep learning interference recognition, and adaptively suppress interference;

[0007] S3. Optimization of dynamic transfer strategy: Define the state space of reinforcement learning, design the action space and reward function, formulate the corresponding reward mechanism, and update the online decision and model in real time;

[0008] S4. Anomaly Detection and Early Warning: Clean all parameter data and extract data features, score anomalies using an isolated forest, and generate multi-level early warnings based on the scoring structure;

[0009] S5. Multi-mode communication switching: Real-time calculation of core signal indicators to dynamically evaluate signal quality, adaptive switching response strategies based on signal evaluation results, verification of switching strategies, and retransmission of data lost during the switching process.

[0010] Step S2 specifically includes the following steps:

[0011] S21. Perform Fourier transform on each frame of communication signal to generate a 256×256 time-frequency matrix, then perform 4-level wavelet packet decomposition on the signal to obtain the decomposition coefficients of 16 frequency bands, calculate the sum of squares of the coefficients and the energy proportion of each frequency band, and finally calculate the kurtosis and skewness of the signal. Determine whether the signal has distortion based on the kurtosis and skewness of the signal.

[0012] S22. The 256×256 time-frequency matrix is ​​concatenated with the 16-dimensional frequency band energy features to form a three-dimensional tensor input layer. The convolutional layer consists of 3 convolutional blocks, with 64 3×3 convolutional kernels outputting 254×254×64, 64 3×3 convolutional kernels outputting 254×254×64, and 2×2 max pooling outputting 126×12 and 6×64 respectively. The pooled feature sequence is processed by bidirectional LSTM to output 32×128 temporal features. The probability of 12 types of interference is output through a 2-layer fully connected network, including Gaussian white noise, narrowband sweep frequency, and wideband blocking.

[0013] S23. If the model determines that the probability of a certain type of narrowband interference is greater than the probability threshold and the center frequency fluctuation of three consecutive frames is less than the fluctuation threshold, the digital notch filter is activated. If the bandwidth is greater than the bandwidth threshold, it is determined to be broadband blocking interference. Fractional Fourier transform is used to process the transform order in the range of 0.1 to 0.9. After finding the order in which the interference energy is concentrated, the interference region is set to zero in the time-frequency domain.

[0014] Step S4 specifically includes the following steps:

[0015] S41. Use the interquartile range method to filter out outliers for all parameters, set a normal value range, mark data that exceeds the range as outliers and exclude them in subsequent processing, and use an additional three-standard-deviation filter for the positioning data.

[0016] S42. Randomly select 256 samples from normal data and construct 100 isolated trees. Starting from the root node of each tree, randomly select a feature and randomly select a split value between the minimum and maximum values ​​of the feature to divide the sample into two parts. Repeat this process until each leaf node contains only 1 sample.

[0017] S43. For each sample to be detected, calculate the average path length from the root node to the leaf node in the isolated tree. The shorter the path length, the higher the anomaly score. Determine the anomaly risk level based on the anomaly score.

[0018] S44. Based on the number of abnormality limits, set the conditions for determining medium-risk and high-risk warnings, execute different warning mechanisms, and automatically store complete data for a certain period of time before and after the abnormality when a warning is triggered.

[0019] Furthermore, in step S1, the multi-source data acquisition methods include: BeiDou positioning data acquisition including latitude and longitude, altitude, velocity vector, and time reference data; sensor data acquisition including accelerometer data, gyroscope data, magnetometer data, temperature, and humidity; and communication status parameter acquisition including signal strength, bit error rate, and carrier-to-noise ratio.

[0020] Furthermore, in step S1, the multi-level preprocessing of the collected multi-source data includes the following steps:

[0021] S11. Wavelet denoising preprocessing: The acquired communication signal is decomposed into 5-level db4 wavelet decomposition, which decomposes the signal into 1 low-frequency approximation component and 5 high-frequency detail components.

[0022] S12. Spatiotemporal alignment preprocessing: Using the BeiDou PPS signal as a reference, all sensor data are calibrated.

[0023] S13. Feature normalization processing: Convert latitude and longitude coordinates into UTM plane coordinates, perform MIN-MAX normalization on the parameters, and map them to [-1, 1].

[0024] S14. Missing value repair: For missing values, the K-nearest neighbor interpolation method is first used to take the average of the three nearest valid data points to fill the initial value, and then Kalman filtering is used for secondary correction.

[0025] Furthermore, step S3 specifically includes the following steps:

[0026] S31. The signal-to-noise ratio, bit error rate, remaining battery power, positioning accuracy factor, and interference type are used as observation state parameters.

[0027] S32. Define the transmission frequency, reflected power, and communication mode as executable actions, and assign corresponding reward and penalty scores to successful transmission, communication interruption, reduction of bit error rate, reduction of reflected power, and remaining power.

[0028] S33. Select the current optimal action with a 90% probability and randomly select an action with a 10% probability. If the remaining power is less than the power threshold, force the low-power mode to be triggered. Store historical data in real time, including the current state, the action executed, the reward obtained, and the next state. Sort the data by combining time decay and reward weights. Update the model parameters and the target network according to the set number of data entries.

[0029] Furthermore, in step S5, the reliability of the current communication mode is determined based on the relationship between the core indicators and the corresponding thresholds. If the current communication mode is determined to be unreliable, a switching process is triggered. The switching priority is Beidou short message > LoRa > 4G. After the switch, 10 sets of test data packets are sent continuously. The effectiveness of the switch is confirmed based on the relationship between the transmission indicators and the corresponding thresholds. For data lost during the switch, a sliding window protocol is used: the sending end records the number of the sent data, the receiving end reports the missing number, and the sending end retransmits the corresponding data to ensure the success rate of retransmission.

[0030] According to another aspect of the present invention, a communication detection system based on BeiDou positioning is provided. The system is used to implement the above-described communication detection method based on BeiDou positioning, and includes: a multi-source data acquisition and preprocessing module, a signal feature extraction and interference identification module, a dynamic transmission strategy optimization module, an anomaly detection and early warning module, and a multi-mode communication switching module.

[0031] Multi-source data acquisition and preprocessing module: used to simultaneously acquire multi-dimensional data and perform multi-level preprocessing on the acquired multi-source data;

[0032] Signal feature extraction and interference identification module: used to extract the time-frequency domain features of the signal, construct a hybrid model for deep learning interference identification, and adaptively suppress interference;

[0033] Dynamic transfer strategy optimization module: used for defining the state space of reinforcement learning, designing the action space and reward function, formulating the corresponding reward mechanism, and updating online decisions and models in real time;

[0034] Anomaly detection and early warning module: This module cleans all parameter data and extracts data features, performs anomaly scoring using an isolated forest algorithm, and generates multi-level early warnings based on the scoring structure.

[0035] Multi-mode communication switching module: used to calculate the core indicators of the signal in real time, dynamically evaluate the signal quality, adaptively switch the response strategy according to the signal evaluation results, verify the switching strategy, and resend the data lost during the switching process. Beneficial effects

[0036] 1. By synchronously collecting BeiDou positioning data, sensor data, and communication status parameters from multiple dimensions, a comprehensive perception of the communication system's operating status was achieved. This breaks through the limitations of traditional methods that rely solely on single data points, providing rich basic data support for subsequent analysis. The multi-level preprocessing process further improves data quality, providing a high-quality data foundation for subsequent interference identification, anomaly detection, and other tasks.

[0037] 2. By generating a time-frequency matrix through Fourier transform and extracting 16 frequency band features through four-layer wavelet packet decomposition, combined with statistical features such as kurtosis and skewness, a comprehensive characterization of the signal's time-frequency domain features is achieved. Compared with traditional single feature extraction methods, this approach can more accurately capture the signal's distortion features. The hybrid deep learning model fully utilizes the spatial and temporal features of the signal, improving the recognition accuracy of 12 common types of interference. Adaptive suppression strategies for different interference types can specifically eliminate the impact of interference, reduce communication quality degradation caused by interference, and significantly improve communication stability in complex electromagnetic environments.

[0038] 3. Using key parameters such as signal-to-noise ratio and bit error rate as state inputs for reinforcement learning, combined with executable actions such as transmission frequency and reflection power, and reward / penalty mechanisms, the system can autonomously learn and dynamically adjust its transmission strategy, achieving closed-loop optimization of "environment-policy-effect". The combination of 90% probability selection of the optimal action and 10% random exploration ensures both the effectiveness of the strategy and adaptability to unknown environments. Real-time storage of historical data and updating the model according to time decay and reward weights ensure the timeliness of the strategy and rapid response to changes in the communication environment. When the remaining battery power is below a threshold, a low-power mode is forcibly triggered, effectively reducing energy consumption while ensuring communication needs, extending device battery life, and improving the system's practicality in power-free scenarios such as in the wild.

[0039] 4. A dual-filtering approach using the interquartile range (ICM) and three-standard-deviation method effectively eliminates interference from extreme noise, improving data cleanliness. A model of 100 isolated trees constructed from 256 samples calculates sample path lengths to achieve anomaly scoring, accurately identifying abnormal fluctuations in communication parameters and demonstrating greater generalization ability compared to traditional threshold methods. A multi-level early warning mechanism based on anomaly risk levels can trigger different response strategies, achieving tiered anomaly response. Automatic storage of complete data before and after anomalies provides detailed evidence for subsequent fault tracing and system optimization, solving the problems of vague anomaly localization and lack of data support in traditional methods.

[0040] 5. By dynamically evaluating the quality of core signal indicators in real time, the accuracy of communication mode switching timing is ensured. A priority switching strategy of "BeiDou short message > LoRa > 4G" prioritizes high-reliability communication methods in complex environments, avoiding communication failures caused by single-mode interruptions. A verification mechanism involving sending 10 sets of test data packets after switching quickly confirms the effectiveness of the switch, reducing the risk of switching failure. The sliding window protocol, by recording the transmission number, reporting missing numbers, and retransmitting data, ensures the complete retransmission of lost data during the switching process, solving the data loss problem in traditional switching methods and guaranteeing the continuity and integrity of communication data. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating the method. Detailed Implementation

[0042] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Example

[0043] The first step involves the multi-source data acquisition and preprocessing module simultaneously acquiring multi-dimensional data and performing multi-level preprocessing on the acquired multi-source data. The acquisition methods for multi-source data are as follows: BeiDou positioning data acquisition includes latitude and longitude, altitude, velocity vector, and time reference data; sensor data acquisition includes accelerometer data, gyroscope data, magnetometer data, temperature, and humidity; and communication status parameter acquisition includes signal strength, bit error rate, and carrier-to-noise ratio.

[0044] Multi-level preprocessing of collected multi-source data includes the following steps:

[0045] 1. Wavelet denoising preprocessing: The acquired communication signal is decomposed into 5-level db4 wavelet decomposition, which decomposes the signal into 1 low-frequency approximation component and 5 high-frequency detail components.

[0046] 2. Spatiotemporal alignment preprocessing: Using the BeiDou PPS signal as a reference, all sensor data are calibrated;

[0047] 3. Feature normalization processing: Convert latitude and longitude coordinates into UTM plane coordinates, perform MIN-MAX normalization on the parameters, and map them to [-1, 1];

[0048] 4. Missing value repair: For missing values, the K-nearest neighbor interpolation method is first used to obtain the average value of the three nearest valid data points to fill the initial value, and then Kalman filtering is used for secondary correction.

[0049] By simultaneously collecting BeiDou positioning data, sensor data, and communication status parameters from multiple dimensions, a comprehensive perception of the communication system's operating status was achieved. This breaks through the limitations of traditional methods that rely solely on single data points, providing rich basic data support for subsequent analysis. The multi-level preprocessing process further improves data quality, providing a high-quality data foundation for subsequent interference identification, anomaly detection, and other tasks.

[0050] The second step involves the signal feature recognition and interference recognition module extracting the time-frequency domain features of the signal, constructing a hybrid deep learning model for interference recognition, and adaptively suppressing interference. Specifically, this includes the following steps:

[0051] 1. Perform Fourier transform on each frame of communication signal to generate a 256×256 time-frequency matrix, then perform 4-level wavelet packet decomposition on the signal to obtain the decomposition coefficients of 16 frequency bands, calculate the sum of squares of the coefficients and the energy proportion of each frequency band, and finally calculate the kurtosis and skewness of the signal. Determine whether the signal has distortion based on the kurtosis and skewness of the signal.

[0052] 2. The 256×256 time-frequency matrix is ​​concatenated with the 16-dimensional frequency band energy features to form a three-dimensional tensor input layer. The convolutional layer consists of three convolutional blocks, with 64 3×3 convolutional kernels outputting 254×254×64, 64 3×3 convolutional kernels outputting 254×254×64, and 2×2 max pooling outputting 126×12 and 6×64 respectively. The pooled feature sequence is processed by bidirectional LSTM to output 32×128 temporal features. The probability of 12 types of interference is output through a 2-layer fully connected network, including Gaussian white noise, narrowband frequency sweep, and wideband blocking.

[0053] 3. If the model determines that the probability of a certain type of narrowband interference is greater than the probability threshold and the center frequency fluctuation is less than the fluctuation threshold for three consecutive frames, the digital notch filter is activated. If the bandwidth is greater than the bandwidth threshold, it is determined to be broadband blocking interference. Fractional Fourier transform is used to process the interference by traversing the transform order in the range of 0.1 to 0.9. After finding the order in which the interference energy is concentrated, the interference region is set to zero in the time-frequency domain.

[0054] By generating a time-frequency matrix through Fourier transform and extracting 16 frequency band features through four-layer wavelet packet decomposition, combined with statistical features such as kurtosis and skewness, a comprehensive characterization of the signal's time-frequency domain features is achieved. Compared with traditional single feature extraction methods, this approach can more accurately capture the signal's distortion features. The hybrid deep learning model fully utilizes the spatial and temporal features of the signal, improving the recognition accuracy of 12 common types of interference. Adaptive suppression strategies for different types of interference can specifically eliminate the impact of interference, reduce the degradation of communication quality caused by interference, and significantly improve communication stability in complex electromagnetic environments.

[0055] The third step involves the dynamic transfer strategy optimization module, which defines the reinforcement learning state space, designs the action space and reward function, formulates a corresponding reward mechanism, and updates the online decision-making and model in real time. Specifically, this includes the following steps:

[0056] 1. Use signal-to-noise ratio, bit error rate, remaining battery power, positioning accuracy factor, and interference type as observation status parameters;

[0057] 2. Define transmission frequency, reflected power, and communication mode as executable actions, and assign corresponding reward and penalty scores to successful transmission, communication interruption, reduction of bit error rate, reduction of reflected power, and remaining power.

[0058] 3. The system selects the current optimal action with a 90% probability and randomly selects an action with a 10% probability. If the remaining battery power is less than the battery threshold, a low-power mode is forcibly triggered. Historical data, including the current state, executed actions, obtained rewards, and the next state, are stored in real time and sorted according to a combination of time decay and reward weights. The model parameters and target network are updated based on the set number of data entries.

[0059] Using key parameters such as signal-to-noise ratio and bit error rate as state inputs for reinforcement learning, combined with executable actions such as transmission frequency and reflection power, and a reward and punishment mechanism, the system can autonomously learn and dynamically adjust its transmission strategy, achieving closed-loop optimization of "environment-policy-effect". The combination of 90% probability selection of the optimal action and 10% random exploration ensures both the effectiveness of the strategy and adaptability to unknown environments. Real-time storage of historical data and updating the model according to time decay and reward weights ensures the timeliness of the strategy and allows for rapid response to changes in the communication environment. When the remaining battery power is below a threshold, a low-power mode is forcibly triggered, effectively reducing energy consumption while ensuring communication needs, extending device battery life, and improving the system's practicality in power-free scenarios such as in the wild.

[0060] The fourth step involves the anomaly detection and early warning module cleaning all parameter data and extracting data features. Anomaly scoring is performed using an isolated forest algorithm, and multi-level early warnings are generated based on the scoring structure. This includes the following steps:

[0061] 1. Use the interquartile range method to filter out outliers for all parameters, set a normal value range, and mark data that exceeds the range as outliers and exclude them in subsequent processing. Use an additional three-standard-deviation filter for the positioning data.

[0062] 2. Randomly select 256 samples from normal data to construct 100 isolated trees. Starting from the root node of each tree, randomly select a feature and randomly select a split value between the minimum and maximum values ​​of the feature to divide the sample into two parts. Repeat this process until each leaf node contains only 1 sample.

[0063] 3. For each sample to be detected, calculate the average path length from the root node to the leaf node in the isolated tree. The shorter the path length, the higher the anomaly score. Determine the anomaly risk level based on the anomaly score.

[0064] 4. Based on the number of abnormality limits, set the conditions for determining medium-risk and high-risk warnings, execute different warning mechanisms, and automatically store complete data for a certain period of time before and after the abnormality when a warning is triggered.

[0065] A dual-filtering approach using the interquartile range (ICM) and three-standard-deviation method effectively eliminates interference from extreme noise, improving data cleanliness. A model of 100 isolated trees constructed from 256 samples calculates sample path lengths to achieve anomaly scoring, accurately identifying abnormal fluctuations in communication parameters and demonstrating greater generalization ability compared to traditional threshold methods. A multi-level early warning mechanism based on anomaly risk levels can trigger different response strategies, achieving tiered anomaly response. Automatic storage of complete data before and after anomalies provides detailed evidence for subsequent fault tracing and system optimization, resolving the problems of vague anomaly localization and lack of data support in traditional methods.

[0066] The fifth step involves the multi-mode communication switching module dynamically evaluating signal quality by calculating core signal indicators in real time. Based on the evaluation results, it adaptively switches to a switching strategy, verifies the switching strategy, and retransmits data lost during the switching process. The reliability of the current communication mode is determined based on the relationship between core indicators and corresponding thresholds. If the current communication mode is deemed unreliable, a switching process is triggered, with the switching priority being BeiDou short message > LoRa > 4G. After switching, 10 sets of test data packets are continuously sent. The effectiveness of the switching is confirmed based on the relationship between transmission indicators and corresponding thresholds. For data lost during the switching process, a sliding window protocol is used: the sending end records the number of transmitted data, the receiving end reports the missing number, and the sending end retransmits the corresponding data to ensure a high retransmission success rate.

[0067] By dynamically evaluating signal quality in real-time using core signal indicators, the accuracy of communication mode switching timing is ensured. A priority switching strategy of "BeiDou short message > LoRa > 4G" prioritizes high-reliability communication methods in complex environments, avoiding communication failures caused by single-mode interruptions. A verification mechanism involving sending 10 sets of test data packets after switching quickly confirms the switching validity and reduces the risk of switching failure. The sliding window protocol, by recording the transmission number, reporting missing numbers, and retransmitting data, ensures the complete retransmission of lost data during switching, solving the data loss problem in traditional switching methods and guaranteeing the continuity and integrity of communication data.

[0068] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for detecting communication based on Beidou positioning, characterized in that, Specifically comprising the following steps: S1, multi-source data acquisition and preprocessing: multi-dimensional data is synchronously collected, and multi-level preprocessing is performed on the collected multi-source data; S2, signal feature recognition and interference identification: time-frequency domain features of signals are extracted, a mixed model deep learning interference identification is constructed, and interference is adaptively suppressed; S3, dynamic transmission strategy optimization: a state space definition is strengthened, an action space and a reward function are designed, a corresponding reward mechanism is formulated, and online decision and model are updated in real time; S4, abnormality detection and early warning: all parameter data are cleaned and data features are extracted, abnormal scores are obtained through isolation forest, and multi-level early warnings are generated according to the score structure; S5, multi-mode communication switching: signal core index is calculated in real time, signal quality is dynamically evaluated, a coping strategy is adaptively switched according to the signal evaluation result, the switching strategy is verified, and lost data in the switching process is retransmitted; The step S2 specifically comprises the following steps: S21, Fourier transform is performed on each frame of communication signal to generate a time-frequency matrix of 256x256, then the signal is subjected to 4-layer wavelet packet decomposition to obtain 16 decomposition coefficients of frequency bands, the coefficient square sum and energy proportion of each frequency band are calculated, and finally the kurtosis and skewness of the signal are calculated, and whether the signal is distorted is determined according to the kurtosis and skewness of the signal; S22, the 256x256 time-frequency matrix and the 16-dimensional frequency band energy feature are spliced to form an input layer of a three-dimensional tensor, the convolution layer is three convolution blocks, which are 64 3x3 convolution kernels outputting 254x254x64, 64 3x3 convolution kernels outputting 254x254x64, and 2x2 maximum pooling outputting 126x12, 6x64, the feature sequence after pooling is processed by bidirectional LSTM, time sequence features of 32x128 are outputted, and 12 types of interference probabilities are outputted through 2-layer full connection network, wherein the 12 types of interference include Gaussian white noise, narrowband sweep, and wideband blocking; S23, if the model determines that the probability of a certain type of narrowband interference is greater than a probability threshold value and the center frequency fluctuation of three consecutive frames is less than a fluctuation threshold value, a digital wave trap filter is started, if the bandwidth is greater than a bandwidth threshold value, the wideband blocking interference is determined, the fractional Fourier transform is processed in the range of 0.1-0.9 to traverse the transform order, the order with concentrated interference energy is found, and the interference area is set to zero in the time-frequency domain; The step S4 specifically comprises the following steps: S41, all parameters are filtered for abnormal values by using the quartile range method, a normal value range is set, data beyond the range is marked as an abnormal value, and the abnormal value is excluded in subsequent processing, and the positioning data is additionally filtered by using three times the standard deviation; S42, 256 samples are randomly selected from normal data, 100 isolated trees are constructed, each tree starts from a root node, a feature is randomly selected, a split value is randomly taken between the minimum value and the maximum value of the feature, the samples are divided into two parts, and the process is repeated until each leaf node contains only one sample; S43, the average path length of each to-be-detected sample from the root node to the leaf node in the isolated tree is calculated, the shorter the path length, the higher the abnormal score, and the abnormal risk degree is determined according to the abnormal score; S44, according to the number of abnormal limit, set the condition of risk warning and high risk warning in the judgment, execute different warning mechanism, when triggering the warning, automatically store the complete data of a certain time before and after the abnormality.

2. The communication detection method based on Beidou positioning according to claim 1, characterized in that: In the step S1, the multi-source data collection mode is Beidou positioning data collection including longitude and latitude, altitude, speed vector, time reference data; sensor data collection including accelerometer data, gyroscope data, magnetometer data, temperature, humidity; The communication state parameter collection includes signal strength, bit error rate, carrier-to-noise ratio.

3. The communication detection method based on Beidou positioning according to claim 1, characterized in that: In the step S1, the multi-level preprocessing of the collected multi-source data includes the following steps: S11, wavelet denoising preprocessing: 5-layer db4 wavelet decomposition is performed on the collected communication signal, and the signal is decomposed into one low-frequency approximate component and five high-frequency detail components; S12, time-space alignment preprocessing: taking Beidou PPS signal as the reference, all sensor data are calibrated; S13, feature normalization processing: converting the longitude and latitude coordinates into UTM plane coordinates, and performing MIN-MAX normalization on the parameters to map to [-1, 1]; S14, missing value repair: the missing value is firstly filled with the average value of the nearest three valid data points by K-nearest neighbor interpolation method, and then is secondly modified by Kalman filter.

4. The communication detection method based on Beidou positioning according to claim 1, characterized in that: The step S3 specifically includes the following steps: S31, taking the signal-to-noise ratio, bit error rate, residual power, positioning accuracy factor and interference type as the observation state parameters; S32, defining the transmission frequency, reflected power and communication mode as executable actions, and assigning the successful transmission, communication interruption, bit error rate reduction, reflected power reduction and residual power to the corresponding reward and punishment shares; S33, selecting the current optimal action by 90% probability and randomly selecting the action by 10% probability, if the residual power is less than the power threshold, triggering the low-power mode, storing the historical data including the current state, executed action, obtained reward and next state in real time, sorting according to the time decay and reward system weight combination, updating the model parameters and target network according to the set data number.

5. The communication detection method based on Beidou positioning according to claim 1, characterized in that: In the step S5, it is determined whether the current communication mode is reliable according to the relationship between the core index and the corresponding threshold value, if it is determined that the current communication mode is unreliable, triggering the switching process, the switching priority is Beidou short message> LoRa> 4G, after switching, 10 groups of test data packets are continuously sent, whether the switching is effective is confirmed according to the relationship between the transmission index and the corresponding threshold value, the sliding window protocol is adopted for the lost data in the switching process: the sending end records the sent data number, the receiving end feeds back the missing number, and the sending end retransmits the corresponding data to ensure the success rate of retransmission.

6. A communication detection system based on Beidou positioning, characterized in that: The system is used to realize the communication detection method based on Beidou positioning in any one of claims 1-5, comprising: a multi-source data acquisition and preprocessing module, a signal feature extraction and interference identification module, a dynamic transmission strategy optimization module, an abnormality detection and warning module, and a multi-mode communication switching module; The multi-source data acquisition and preprocessing module is used for synchronously collecting multi-dimensional data and performing multi-level preprocessing on the collected multi-source data. Signal feature extraction and interference identification module: used for extracting signal time-frequency domain features, constructing hybrid model deep learning interference identification, and adaptively suppressing interference; Dynamic transmission strategy optimization module: used for defining state space, designing action space and reward function, formulating corresponding reward mechanism, and updating online decision and model in real time; Abnormality detection and early warning module: used for cleaning and extracting data features of all parameter data, performing abnormality scoring through isolation forest, and generating multi-level early warning according to scoring structure Multi-mode communication switching module: used for real-time calculation of signal core index to evaluate signal quality, adaptive switching of response strategies according to signal evaluation results, verification of switching strategies, and retransmission of lost data in the switching process.

Images