Multi-source data fusion water radio interference identification and positioning method and system

Abstract

The invention relates to the field of radio interference intelligent identification and positioning, in particular to a multi-source data fusion water radio interference identification and positioning method and system, and the method comprises the following steps: collecting multi-source data; performing multi-source data fusion processing; constructing a rule base, and screening out suspected abnormal signals from the frequency spectrum situation map through threshold judgment and rule matching; a support vector machine model is called, multi-dimensional static signal features are used as input, a classification hyperplane is constructed through a radial basis kernel function, and normal signals and abnormal signals are distinguished from suspected abnormal signals; calling a decision tree model, and performing scene judgment on the suspected abnormal signal to identify the abnormal signal; calling a convolutional neural network model, converting the time domain signal into a spectrogram through short-time Fourier transform, and extracting texture features through multilayer convolution to identify an abnormal signal; through the method and the system, real-time identification and high-precision positioning of interference signals can be realized.

Description

Technical Field

[0001] This invention relates to the field of radio interference identification and location technology, and in particular to a method and system for identifying and locating radio interference on water using multi-source data fusion. Background Technology

[0002] With the surge in water traffic and wireless equipment, the conflict over spectrum resources has intensified, and interference problems have become more prominent and diverse. Traditional manual patrol modes have shortcomings such as coverage blind spots, data isolation, and delayed response, making it difficult to meet the needs of interference control.

[0003] In the prior art, Chinese patent application number CN109348536A proposes an automatic positioning system and method for radio interference signals on water. This automatic positioning system uses more than three fixed monitoring stations to collect radio signal information in the regulated waters and uses TDOA positioning technology to locate the radio interference source. Although it can achieve automatic identification and positioning of radio interference signals to a certain extent, it only uses a few fixed monitoring stations with fixed relative positions to collect interference source data in the target area, resulting in limited coverage. It also has problems such as coverage blind spots, data isolation, insufficient identification accuracy due to the single data source and type, and response lag caused by the inability to predict interference changes. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a method and system for identifying and locating maritime radio interference through multi-source data fusion. This method and system can achieve real-time identification and high-precision location of maritime radio interference signals.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for identifying and locating maritime radio interference through multi-source data fusion includes the following steps: S1. Use mobile monitoring stations plus 24-hour fixed monitoring stations to collect radio signal information in the monitored waters; S2. Extract the spectral characteristics of the radio signal information collected by each monitoring station, predict the trend changes of the spectral characteristics using the Kalman filter algorithm, and output the spectral trend map. S3. Construct a rule base and filter out suspected abnormal signals from the spectrum situation map through threshold judgment and rule matching; S4. Call the support vector machine model, take multidimensional static signal features as input, construct a classification hyperplane through radial basis kernel function, and distinguish normal signals from abnormal signals from suspected abnormal signals. S5. Call the decision tree model, take multi-dimensional scene features as input, and perform scene-based judgment on suspected abnormal signals to identify abnormal signals. S6. Call the convolutional neural network model to convert the time-domain signal into a spectrum through short-time Fourier transform, extract texture features through multi-layer convolution, and identify abnormal signals from suspected abnormal signals. S7. Cross-validate the abnormal signals identified by the model with multi-source data from various monitoring stations, screen out interference signals, and perform TDOA positioning to determine the location of the interference source.

[0006] To optimize the above plan, the following measures will be further taken: As one implementation method, in S4 to S6, the probability score of the support vector machine model outputting an abnormal signal is: The decision tree model outputs a probability score for an abnormal signal. The probability score of the convolutional neural network model outputting an abnormal signal is: ; The steps following S7 include dynamically weighting and summing the probability scores of the three models' outputs being abnormal signals according to the following formula, and finally calculating a comprehensive abnormal probability score P. P= Implement tiered early warning based on the comprehensive anomaly probability score P, including the following steps: When 0.30 ≤ P < 0.50, it is judged as a mild abnormal situation, a level 1 warning is implemented, a warning message is sent and a level 1 handling strategy is executed; When 0.50 ≤ P < 0.80, it is judged as a moderate abnormal situation, a level-two warning is implemented, a warning message is sent and a level-two handling strategy is executed; When P ≥ 0.80, it is determined to be a serious abnormal situation, a level 3 early warning is implemented, an early warning message is sent and a level 3 handling strategy is executed.

[0007] In one implementation, the support vector machine model in S4 takes 8-dimensional signal features as input, including frequency offset, peak power spectral density, modulation depth, modulation frequency stability, signal bandwidth occupancy, duty cycle, signal-to-noise ratio, and harmonic distortion coefficient. The abnormal signals output by the support vector machine model include co-frequency interference signals and intermodulation interference signals. In S5, the decision tree model takes 4-dimensional scene features as input, including signal occurrence time, geographical location type, associated ship type, and surrounding equipment density. In S6, the abnormal signals output by the convolutional neural network model include frequency hopping interference signals and narrowband burst interference signals.

[0008] In one implementation, the fixed monitoring station is equipped with a spectrum analyzer, a 24-hour monitoring antenna, and AIS equipment to continuously collect radio signal information in the monitored waters. The mobile monitoring station includes patrol boats, mobile monitoring vehicles, and drones. The patrol boat is equipped with a portable spectrum analyzer, a directional monitoring antenna, satellite communication equipment, AIS equipment, and communication facilities to monitor the use of shipboard radio equipment during patrols. The mobile monitoring vehicle is equipped with spectrum monitoring equipment and direction-finding equipment to conduct spectrum monitoring and direction finding in response to regulatory needs in areas around coastal ports, wind farms, and shore-based radio facilities. The drone is equipped with a miniaturized spectrum monitoring module and a high-definition camera to collect spectrum data and capture images in monitoring blind spots.

[0009] In one embodiment, S2 includes the following steps: S2-1. Perform time synchronization and spatial consistency checks on the radio signal information data collected by fixed monitoring stations and mobile monitoring stations. The radio signal information data includes spectrum data, AIS data, and environmental data. S2-2, Utilization The criteria and Mahalanobis distance will remove outlier data after the test; S2-3. Dynamically assign data weights to each monitoring station according to the following formula: in, It is the calculated number of The data weight of each monitoring station Is this the first The equipment reliability coefficient of the monitoring station, when the... When each monitoring station is a fixed monitoring station =0.9, for mobile monitoring vehicles =0.85, for cruising ships =0.7, for drones =0.6; Is this the first The distance attenuation factor of each monitoring station is determined by the distance between the monitoring station and the monitored object, where n is the total number of monitoring stations; S2-4. Construct state equations based on the system evolution model of each monitoring station and observation equations based on the measurement model of each monitoring station. Then, fuse the predicted values ​​of the information data derived from the state equations with the actual measured values ​​of the monitoring stations using Kalman filtering to obtain the optimal estimated values, including power and frequency. S2-5. The data from multiple monitoring stations that have collected the same radio signal information are weighted and averaged according to the following formula, and a spectrum situation diagram is output based on the fusion result. in, It is the fusion power value. It is the fusion frequency value. The data weights calculated for S2-3, For the first The optimal power estimate for each monitoring station. For the first The optimal frequency estimate for each monitoring station.

[0010] As one implementation method, the following steps are also included: S2-6. The reliability of the fusion result is evaluated and classified according to the following formula; in, Represents the reliability assessment results. This indicates the percentage of valid sites participating in the integration out of the total number of sites. The complement representing the relative standard deviation of the fusion result. This indicates the spatial uniformity of the monitoring stations, ranging from 0.6 to 1.0. The more uniform the station distribution, the better. The larger; when and It is determined to be highly reliable and can be used to generate spectrum situation maps; when and It is determined to be of medium reliability, and other information is needed to help determine whether it can be used to generate a spectrum situation map; when or It was determined to be of low reliability and could not be used to generate a spectrum situation map, but additional monitoring resources were needed for the same radio signal information.

[0011] As one implementation method, in S2-1, the information data is spatially consistent using the following formula. If the condition is met, the data passes the consistency test; otherwise, it is marked as abnormal and the abnormal data is removed. in Representing the first Signal power and frequency of a fixed monitoring station Representing the first The signal power and frequency of each mobile monitoring station These are the power difference threshold and the frequency difference threshold, respectively. For the first The number of fixed monitoring stations to the first The distance between mobile monitoring stations.

[0012] As one implementation method, it also includes a step of adaptively optimizing and adjusting the fusion result according to the following formula; in, This is the current adaptive adjustment value of the output. This is the initially set baseline adjustment threshold. This is the ratio of recent fluctuations to baseline fluctuations. This ratio increases when current environmental disturbances intensify, and approaches 1 when the environment is stable. It is an adjustment coefficient that controls the adaptation speed.

[0013] As one implementation method, the abnormal signals identified by models S4 to S6 are verified by AIS data association to check whether there is any ship radio equipment with parameters matching the abnormal signal, while historical databases are retrieved and manual verification is performed.

[0014] The embodiments of this specification also propose a system for implementing the above-described multi-source data fusion method for identifying and locating maritime radio interference, comprising: The data acquisition layer is responsible for receiving multi-source heterogeneous data of radio signal information collected by fixed monitoring stations and mobile monitoring stations; The data transmission layer uses communication network technology to transmit the data acquired by the data acquisition layer to the data processing layer; The data processing layer standardizes and normalizes the multi-source heterogeneous data transmitted from the data transmission layer, eliminating redundancy and contradictions between data, and performs Kalman filtering fusion; it constructs a rule base and matches real-time acquired data against the rule base one by one, identifying abnormal signals through a multi-model collaborative architecture; it uses positioning technology to determine the direction and location of interference signals, pinpointing the location of the interference source; and... The data storage layer is responsible for storing monitoring data and processing results; classifying and querying data, distinguishing between real-time data and historical data, to facilitate later querying and analysis.

[0015] Because of the above-described solutions, one or more technical solutions provided in this application embodiment have at least the following technical effects or advantages: On one hand, this application employs multi-source data fusion technology, the core of which lies in fusing data from different sources such as fixed monitoring stations and mobile monitoring stations through weighted Kalman filtering. Among them, the 24-hour fixed monitoring station continuously collects radio signal information from the regulated waters, which has the advantages of strong temporal continuity and high data stability. The mobile monitoring station can collect real-time operating parameters of ship radio equipment at close range, and the data has strong spatial specificity and target correlation. In addition, the mobile monitoring station can cover monitoring blind spots and respond to sudden regulatory needs to achieve intensive monitoring of key areas. Based on the fusion of data sources from different monitoring stations and monitoring equipment, a natural complementarity is formed in terms of spatiotemporal dimension, accuracy characteristics and coverage. Therefore, the system can obtain more comprehensive and accurate information, providing a data foundation for high-precision identification of radio interference signals.

[0016] On the other hand, this application adopts a multi-model collaborative architecture of "traditional machine learning + deep learning". Addressing the diverse types and significant differences in features of interference signals, it constructs a system of "static parameter classification - scene dynamic discrimination - complex morphology capture", as detailed below: Through efficient threshold judgment and rule matching, 70% of compliance signals can be filtered out in a very short time, and only the remaining 30% of suspected abnormal signals are passed to the subsequent multi-model architecture. By invoking the Support Vector Machine (SVM) model and using multidimensional signal static features as input, a high-dimensional classification hyperplane is constructed through the Radial Basis Function (RBF) kernel function. This model focuses on identifying co-frequency and intermodulation interference. Statistically, the accuracy rate for identifying co-frequency interference using this model is 97.2%, and the accuracy rate for identifying intermodulation interference is 94.8%. By incorporating multi-dimensional scene features through a decision tree model and establishing a scene-interference probability mapping based on multi-node splitting rules, the limitations of SVM are corrected. Statistical analysis shows that the model achieves an accuracy rate of 91.5% in identifying illegal vessels operating at night and 89.3% in identifying illegal equipment around wind farms. A convolutional neural network (CNN) was used to transform the "time-frequency-power" information data into a spectrogram. Texture features were extracted through multi-layer convolution, focusing on identifying frequency hopping and narrowband burst interference. Statistically, the accuracy rate of frequency hopping interference identification using this model was 96.7%, and the accuracy rate of narrowband burst interference identification was 93.4%. Compared to traditional manual monitoring, the AI ​​recognition system, which integrates static parameter classification, dynamic scene discrimination, and complex morphology capture, significantly improves the efficiency and accuracy of radio interference signal identification while drastically reducing labor costs. A pilot application in a certain water area demonstrated a 96.5% accuracy rate, representing a 5400-fold increase in interference detection speed and a 27.5 percentage point improvement in accuracy compared to traditional manual monitoring methods.

[0017] On the other hand, this application dynamically weights and sums the probability scores of the outputs of the three models, SVM, decision tree, and CNN, which are abnormal signals, to calculate the comprehensive abnormal probability score P. Based on this, a graded early warning is implemented. According to the severity and urgency of the warning, different levels of warning information are pushed to different levels of personnel, which shortens the accident handling time, reduces the accident handling cost, and ensures that the emergency response process is triggered, so as to resolve the interference risk in a timely manner and reduce false alarms and missed alarms. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings of the embodiments will be briefly described below. Obviously, the drawings described below only involve some embodiments of this application and should not be construed as limiting this application.

[0019] Figure 1 This is a schematic diagram of the interference identification logic of the multi-source data fusion method for identifying and locating radio interference on water in this embodiment; Figure 2 This is a schematic diagram of the monitoring path after the monitoring station is installed in this embodiment; Figure 3 This is a schematic diagram of the overall logic of the multi-source data fusion method for identifying and locating radio interference on water in this embodiment; Figure 4 This is a schematic diagram of the system data fusion process in this embodiment. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings, so as to more clearly understand the purpose, features and advantages of this invention. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of this invention, but are only for illustrating the essential spirit of the technical solutions of this invention. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0021] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0022] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0023] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to mean “and / or” unless otherwise expressly stated herein.

[0024] In the following description, in order to clearly demonstrate the structure and working method of the present invention, a number of directional terms will be used. However, terms such as "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", and "down" should be understood as convenient terms and not as limiting terms.

[0025] The implementation details of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The following content is only for the convenience of understanding the implementation details and is not necessary for implementing this solution.

[0026] This embodiment provides a method and system for identifying and locating radio interference on water using multi-source data fusion. It aims to solve the problems in the existing technology, such as coverage blind spots, isolated data, and delayed response in the collection of radio interference signals in the monitored area. It also addresses the problems of relying on manual analysis of the collected radio signals, which results in time-consuming data processing, high false judgment rate, and low positioning accuracy. The system can not only achieve full-area monitoring, multi-source collaboration, and intelligent applications, but also significantly improve the accuracy and efficiency of interference identification.

[0027] This embodiment first proposes a method for identifying and locating radio interference on water using multi-source data fusion, including the following steps: S1. Use mobile monitoring stations plus 24-hour fixed monitoring stations to collect radio signal information in the monitored waters; S2. Extract the spectral characteristics of the radio signal information collected by each monitoring station, predict the trend changes of the spectral characteristics using the Kalman filter algorithm, and output the spectral trend map. S3. Construct a rule base and filter out suspected abnormal signals from the spectrum situation map through threshold judgment and rule matching; like Figure 1 As shown, specifically, after data preprocessing, radio signal information enters a rule base for matching and filtering. If the signal matches the rule base, it is marked as a normal signal and continuously monitored. If it does not match, it enters a multi-model collaborative recognition step, as follows: S4. Call the Support Vector Machine (SVM) model, take multidimensional static signal features as input, and construct a high-dimensional classification hyperplane through the radial basis function (RBF) to distinguish normal signals from abnormal signals from suspected abnormal signals; S5. Call the decision tree model, take multi-dimensional scene features as input, and perform scene-based judgment on suspected abnormal signals to identify abnormal signals. S6. Call the Convolutional Neural Network (CNN) model to convert the time-domain signal into a spectrum through short-time Fourier transform, extract texture features through multi-layer convolution, and identify abnormal signals from suspected abnormal signals. S7. Cross-validate the abnormal signals identified by the model with multi-source data from various monitoring stations, screen out interference signals, and perform TDOA positioning to determine the location of the interference source.

[0028] Specifically, anomalous signals identified by models S4 to S6 are verified through AIS data correlation to check for the existence of shipboard radio equipment with parameters matching the anomalous signal. Simultaneously, historical databases are retrieved and manual verification is performed. Through multi-source data cross-validation, if a signal is confirmed as anomalous, it is marked as an illegal interference signal and the identification result is output. If it cannot be confirmed as anomalous, the decision logic is updated in the rule base, completing dynamic optimization, and the model continuously monitors in real time. Multi-source data cross-validation improves the reliability of the model's prediction results by cross-validating with external data such as AIS data and historical databases.

[0029] In this embodiment, 24-hour fixed monitoring stations are deployed in key areas such as major coastal ports, inland waterway hubs, international waterways, wind farm areas, and core areas for ship navigation. Based on Geographic Information System (GIS) data and electromagnetic environment simulation, electromagnetic interference sources such as high-voltage power lines and industrial equipment are eliminated. Data storage servers are configured to ensure the storage of historical data for a period of no less than one year. Remote control modules are installed to support remote operation and troubleshooting of the monitoring stations.

[0030] Specifically, the fixed monitoring station includes a general monitoring unit, a dedicated maritime frequency band monitoring unit, and an AIS monitoring unit. All of these monitoring units establish communication with the control center via a network. The general monitoring unit includes a general monitoring receiver connected to a GPS antenna and a general monitoring antenna. The dedicated maritime frequency band monitoring unit includes a dedicated maritime frequency band monitoring receiver connected to a maritime frequency band monitoring antenna. The AIS monitoring unit includes an AIS monitoring receiver connected to an AIS antenna. In one configuration, both the general monitoring antenna and the maritime frequency band monitoring antenna are omnidirectional antennas, and the specific frequency range can be selected by the user.

[0031] Specifically, the 24-hour fixed monitoring station is equipped with a high-performance spectrum analyzer with a dynamic range of ≥100dB and a frequency resolution of ≤1Hz, covering the 300KHz-3GHz frequency band. After the equipment is confirmed to be operating normally, it starts to collect radio signal information of the monitored water area 24 hours a day. After setting an appropriate measurement frequency range and resolution, it begins to collect data such as frequency, power, modulation method, and signal bandwidth. The collected data is analyzed in real time to generate a spectrum occupancy map and display the signal distribution of each frequency band.

[0032] In this embodiment, the mobile monitoring station includes a patrol boat, a mobile monitoring vehicle, and a drone. The patrol boat is equipped with a portable spectrum analyzer, a directional monitoring antenna, satellite communication equipment, AIS equipment, and communication facilities to monitor the usage of ship radio equipment during patrols. The patrol boat's monitoring data is linked with the shipborne portable spectrum analyzer and AIS codec module to collect real-time operating parameters of ship radio equipment at close range. The data has strong spatial specificity and target correlation. The mobile monitoring vehicle is equipped with spectrum monitoring equipment and direction finding equipment with the same performance as the fixed monitoring station. It is used to respond to sudden regulatory needs or interference complaints to conduct spectrum monitoring and direction finding in areas around coastal ports, wind farms, and shore-based radio facilities, outputting high signal-to-noise ratio spectrum detail data to support the accurate extraction of interference signal characteristics. The drone is equipped with a miniaturized spectrum monitoring module and a high-definition camera. With its flexible mobility, it can collect spectrum data and capture images in monitoring blind spots. However, its data is fragmented due to limited battery life.

[0033] This embodiment also includes the steps of establishing a control center and a data center. A secure and efficient multi-link transmission system is established at the data transmission end. Fixed monitoring stations use fiber optic dedicated lines to transmit data to the data center. Patrol boats and mobile monitoring vehicles transmit data to the data center through a 5G private network + satellite backup. UAVs use microwave backhaul to the ground control station, which then sends the data to the data center for backup, ensuring data real-time performance and security.

[0034] In this embodiment, a radio spectrum big data platform is deployed at the data transmission end. Real-time collected data is uploaded to the platform via the network. During the upload process, data integrity and security are ensured to prevent loss or leakage. The platform ensures real-time data storage for at least one year and distinguishes between real-time and historical data for easy later querying and analysis.

[0035] Mobile monitoring stations establish patrol routes based on the monitored area, such as covering major waterways daily or weekly to ensure regular monitoring of all important waterways. Key monitoring points along the routes are identified and tracked in real time. Following pre-set routes, patrol vessels conduct patrols in ports, inland waterways, and other areas, monitoring the use of ship radio equipment, particularly VHF communication and AIS equipment. Temporary radio installations, such as communication equipment on construction vessels, are inspected to ensure compliance with parameters. Special attention is paid to merchant ships and fishing vessels, using VHF and other radio devices to communicate navigation safety information, guide them to compliant routes, remind them to maintain designated channel watch, and monitor the compliance of their radio equipment, licenses, and frequency usage.

[0036] During the patrol, the spectrum analyzer on the patrol boat collects data in real time, and all monitoring data is uploaded to the wireless spectrum big data platform via satellite communication. If any illegal signals are detected, such as a vessel not registering a frequency band or exceeding the signal power limit, the patrol personnel will immediately communicate with the relevant vessel through the shipboard communication equipment and request rectification. If any illegal equipment or behavior is found, the patrol personnel must require the vessel to rectify it immediately and record the rectification status to ensure follow-up tracking.

[0037] The drones used for patrols are equipped with miniaturized spectrum monitoring modules and high-definition cameras to ensure effective monitoring of spectrum data and image capture. The drones need to have long-duration flight and long-distance transmission capabilities, and be equipped with a stable ground control station for remote operation. Flight routes for drones should be planned for hard-to-reach areas, such as shallow waters, narrow channels, and under bridges. These flight routes should include areas for monitoring potentially illegal equipment, such as privately installed communication base stations or improperly installed antennas. Once an anomaly is detected, the ground control station will mark the abnormal location and notify patrol boats or mobile monitoring vehicles for further action. If illegally installed equipment is found, such as antennas exceeding height limits, it will be immediately recorded and the relevant departments notified for handling.

[0038] For mobile monitoring vehicle patrols, equip them with spectrum monitoring equipment and direction-finding equipment of equivalent performance to those used at fixed monitoring stations, and ensure the stable operation of the onboard equipment. Configure the mobile monitoring vehicle's power system and communication system to ensure continuous operation of the equipment during patrols.

[0039] Patrol missions are arranged according to regulatory needs (such as receiving interference complaints or warnings from fixed monitoring stations). The focus is on monitoring coastal ports, areas surrounding wind farms, and shore-based radio facilities to ensure a stable electromagnetic environment. During patrols, if unapproved port communication equipment or other illegal facilities are discovered, on-site verification and handling are immediately conducted. Data comparison is performed in conjunction with patrol boats and drones to ensure comprehensive monitoring.

[0040] For mobile monitoring station data integration, data collected by patrol boats, drones, and mobile monitoring vehicles will be coupled through a big data platform to form a real-time monitoring system. Data analysis tools automatically screen for abnormal signals and issue alarms to ensure timely handling. Through the big data platform, supervisory personnel can view the monitoring status of each patrol device in real time. If any anomalies are detected, the platform automatically directs patrol boats, drones, and mobile monitoring vehicles to coordinate and handle the situation. The monitoring system will be regularly evaluated to summarize lessons learned during implementation, optimize patrol plans and equipment configurations, and improve patrol efficiency and coverage.

[0041] In this embodiment, multi-source data fusion technology is adopted to fuse data from different sources such as fixed monitoring stations and mobile monitoring stations through weighted Kalman filtering, making the monitoring data more forward-looking and predictive. Based on the data source fusion of different monitoring stations and monitoring equipment, they form a natural complement in terms of spatiotemporal dimension, accuracy characteristics and coverage. Therefore, the system can obtain more comprehensive and accurate information, providing a high-quality data foundation for high-precision identification of radio interference signals.

[0042] This embodiment also includes the steps of establishing a control center and a data center. A secure and efficient multi-link transmission system is established at the data transmission end. Fixed monitoring stations use fiber optic dedicated lines to transmit data to the data center. Patrol boats and mobile monitoring vehicles transmit data to the data center through a 5G private network + satellite backup. UAVs use microwave backhaul to the ground control station, which then sends the data to the data center for backup, ensuring data real-time performance and security.

[0043] In this embodiment, a radio spectrum big data platform is deployed at the data transmission end. Real-time collected data is uploaded to the platform via the network. During the upload process, data integrity and security are ensured to prevent loss or leakage. The platform ensures real-time data storage for at least one year and distinguishes between real-time and historical data for easy later querying and analysis.

[0044] In one approach, the platform incorporates a trend analysis model to periodically analyze historical data and identify spectrum usage patterns. For example, it analyzes changes in the strength of ship communication signals in a waterway to identify signal fluctuations during peak hours. It also monitors the operational stability of equipment, assesses power fluctuations in shore-based equipment, and predicts whether maintenance is required. The platform generates data analysis reports, including spectrum usage patterns, equipment operational status analysis, and interference source analysis, for management reference.

[0045] In one approach, the monitoring station performs real-time direction finding of surrounding radio signals to identify unidentified signal sources, especially potential illegal interference signals. The signal direction finding function is activated to initially locate the interference signal. Based on the measurement data, the direction and intensity information of the interference signal are uploaded to the platform for subsequent interference investigation, team confirmation, and handling. The approximate location of the interference source is determined, and communication with relevant on-site departments is established to arrange subsequent investigation and handling.

[0046] In one approach, spectrum analysis data is used to identify idle frequency bands in spectrum usage, and recommendations are made to management to reconfigure spectrum resources. Based on the analysis results, the allocation of spectrum resources in areas such as waterways and ports is optimized to improve spectrum utilization.

[0047] In one approach, the control center performs TDOA positioning on the radio interference signal to obtain latitude and longitude. , and time Location result information; Location within a given time period is in latitude and longitude. , A collection of ships within a radius of k kilometers. Enter suspected radio interference vessels into the database; similarly, in After TDOA positioning is performed continuously, the ships are assembled. It will be entered into the suspected radio interference ship database; retrieve , ,... Intersection of sets This refers to the suspected vessel where an illegal radio station emitting radio interference signals is located.

[0048] In this embodiment, a multi-model collaborative architecture of "traditional machine learning + deep learning" is adopted. To address the diverse types and significant differences in features of interference signals, a system of "static parameter classification - dynamic scene discrimination - complex morphology capture" is constructed. The specific hierarchical logic of the architecture is as follows: like Figure 1 As shown, after data preprocessing, rule-based matching and filtering are first performed. Using radio regulations and industry standards, suspected anomalous signals are quickly filtered based on predefined rules. Through efficient threshold judgment and rule matching, 70% of compliant signals can be filtered out in a very short time, with only the remaining 30% of suspected anomalous signals being passed to the subsequent deep recognition layer.

[0049] Specifically, the rule base is built based on the "Radio Regulations" and the "Construction Specifications and Technical Requirements for VHF and UHF Radio Monitoring Facilities (Trial)" as the core basis, and combined with the regulatory characteristics of different water areas, to achieve differentiated judgment of the same parameter in different scenarios.

[0050] In the deep recognition layer, a multi-model collaborative approach is employed for accurate identification. Support Vector Machines (SVMs) excel at handling high-dimensional, small-sample data, making them particularly suitable for identifying interference types with clear boundaries, such as co-frequency interference and intermodulation interference. Therefore, by calling the SVM model and using multi-dimensional signal static features as input, an optimal classification hyperplane is constructed through the Radial Basis Function (RBF) kernel function, focusing on identifying co-frequency and intermodulation interference. The specific workflow involves standardizing the input data, building the model using the training set, calculating the distance to the classification hyperplane, and outputting anomaly probability scores. The accuracy rate for identifying co-frequency interference signals is 97.2%, and the accuracy rate for identifying intermodulation interference signals is 94.8%.

[0051] Decision trees can integrate information from multiple scenarios and simulate expert reasoning logic, making them suitable for handling complex spatiotemporal information and scenario-based judgments of device behavior. They are particularly advantageous in identifying anomalies related to "when, where, and which device." By incorporating multi-dimensional scenario features into the decision tree model and establishing a mapping between scenario and interference probabilities based on multi-node splitting rules, the limitations of SVM are corrected. The following tree structure design is adopted, with a maximum depth limit of 6 layers. The node splitting criterion is Gini impurity, generating 137 decision nodes and 68 leaf nodes. Based on the input features, the signal sequentially passes through each node of the decision tree for judgment, and finally, anomaly probability scores are output at the leaf nodes. The model achieved an accuracy rate of 91.5% in identifying illegal vessels operating at night and 89.3% in identifying illegal equipment around wind farms.

[0052] CNNs can automatically learn local patterns in spectrograms through convolutional layers, making them particularly suitable for identifying complex interference signals such as frequency hopping interference and narrowband burst interference. These interference signals exhibit unique texture features in their spectrograms, which are difficult to extract effectively using traditional methods. A convolutional neural network (CNN) is used to perform a short-time Fourier transform (STFT) on the time-domain signal, converting it into a 224×224 grayscale spectrogram, which is then input into the CNN model for processing. Texture features are extracted through three convolutional layers, focusing on identifying frequency hopping and narrowband burst interference. The network structure is designed as follows: Convolutional layer 1: 32 3×3 convolutional kernels to extract edge and line features from the spectrogram; Convolutional layer 2: 64 3×3 convolutional kernels to extract higher-level features; Convolutional layer 3: 128 3×3 convolutional kernels to extract complex spectral features; Fully connected layer: Decisions are made through a fully connected layer with 256 neurons, and the probability distribution of the three types of signals is output using the Softmax function; The specific workflow involves inputting a spectrogram for feature extraction and pattern recognition, and outputting a signal category and anomaly probability score. Using this model, the accuracy rate for frequency hopping interference identification is 96.7%, and the accuracy rate for narrowband burst interference identification is 93.4%.

[0053] Compared to traditional manual monitoring, the AI ​​recognition system, which integrates static parameter classification, dynamic scene discrimination, and complex morphology capture, significantly improves the efficiency and accuracy of radio interference signal identification while drastically reducing labor costs. A pilot application in a certain water area demonstrated a 96.5% accuracy rate, representing a 5400-fold increase in interference detection speed and a 27.5 percentage point improvement in accuracy compared to traditional manual monitoring methods.

[0054] As one implementation method, in S4 to S6, the probability score of the support vector machine model outputting an abnormal signal is: The decision tree model outputs a probability score for an abnormal signal. The probability score of the convolutional neural network model outputting an abnormal signal is: ; Following S7, the following steps are included: dynamically weighting and summing the probability scores of the three models' outputs being anomalous signals according to the following formula, and finally calculating a comprehensive anomalous probability score P: P= Based on the comprehensive anomaly probability score P, a tiered early warning system is implemented, which includes the following steps: When 0.30 ≤ P < 0.50, it is judged as a mild abnormal situation, and a level one warning is implemented. A warning message is sent and a level one handling strategy is executed. Specifically, it is necessary to remind the management department and regulators to pay attention.

[0055] When 0.50 ≤ P < 0.80, it is judged as a moderate abnormal situation, and a level-two warning is implemented. A warning message is sent and a level-two handling strategy is executed. Specifically, the person in charge of the user unit needs to be notified to conduct regular verification and provide a rectification plan within a specified period.

[0056] When P ≥ 0.80, it is determined to be a serious abnormal situation, and a three-level early warning is implemented. An early warning message is sent and a three-level handling strategy is executed. Specifically, the department leader and other responsible persons are notified to trigger the emergency procedure.

[0057] Each warning message must be accompanied by detailed data, including: a spectrum diagram of the abnormal signal (in image form), the time and specific location of the anomaly (e.g., latitude and longitude, device ID), the type of equipment involved, its operating status, and the status of related upstream and downstream equipment (e.g., device ID, power, frequency). For each warning, the system will record the feedback results, including rectification status, processing procedures, and outcomes, ensuring continuous tracking of regulatory work. Regulatory departments and equipment users must complete the warning processing within the specified time and upload the results to the platform to ensure that problems are addressed promptly.

[0058] Upon triggering a Level 3 warning, the emergency response process is automatically initiated. Relevant departments (such as maritime authorities and equipment manufacturers) must respond immediately and take measures. Depending on the specific interference situation (e.g., ship communications are affected), it may be necessary to call in backup frequencies, shut down relevant equipment, or conduct on-site inspections. A detailed emergency response report is generated for each warning situation, including emergency response measures, implementation status, and final results, and uploaded to the data platform. An accident investigation and follow-up processing are initiated, and the rectification results are reported back to ensure the problem is thoroughly resolved.

[0059] In this embodiment, the probability scores of the outputs of the three models, SVM, decision tree, and CNN, which are abnormal signals are dynamically weighted and summed to calculate the comprehensive abnormal probability score P. Based on this, a graded early warning is implemented. According to the severity and urgency of the warning, different levels of warning information are pushed to personnel at different levels, which shortens the accident handling time, reduces the accident handling cost, and ensures that the emergency response process is triggered in a timely manner, thus mitigating the risk of interference and reducing false alarms and missed alarms.

[0060] In this embodiment, historical early warning data is periodically reviewed and analyzed to evaluate the accuracy and response efficiency of the early warning system, and to analyze the performance of the early warning model under different scenarios. The model is regularly updated by combining new monitoring data and practical operational feedback to improve identification accuracy and prediction precision. More deep learning models are introduced to enhance the ability to identify complex signal features and reduce false alarms and missed alarms.

[0061] like Figure 2As shown in this embodiment, by deploying monitoring stations, at least three monitoring methods for radio signals within the regulated waters can be established. These three methods are implemented in parallel. First, there are 24-hour fixed monitoring stations, prioritized in typical ports and areas with concentrated wind farms. These fixed stations generate spectrum occupancy datasets and store historical monitoring data. Second, there are mobile monitoring stations, utilizing patrol boats, drones, and mobile monitoring vehicles to create dynamic monitoring of the regulated waters, enabling flexible responses to interference monitoring needs. Third, monitoring is conducted through AI-based early warning systems. Machine learning algorithms intelligently identify interference and classify warnings accordingly. Specifically, a three-tiered early warning system is adopted, with each tier sending warning messages to different departments and personnel. This three-tiered processing strategy ensures reasonable resource allocation and timely triggering of emergency procedures.

[0062] In this embodiment, the support vector machine model in S4 takes eight-dimensional signal features as input: frequency offset, peak power spectral density, modulation depth, modulation frequency stability, signal bandwidth occupancy, duty cycle, signal-to-noise ratio, and harmonic distortion coefficient. The abnormal signals output by the support vector machine model include co-channel interference signals and intermodulation interference signals. The decision tree model in S5 takes four-dimensional scene features as input: signal occurrence time, geographical location type, associated ship type, and surrounding equipment density. The abnormal signals output by the convolutional neural network model in S6 include frequency hopping interference signals and narrowband burst interference signals. Each model focuses on different feature dimensions and outputs independent confidence scores. These results are fused through a weighted voting mechanism to enhance the accuracy and reliability of the identification.

[0063] In some implementations, the system also includes dynamic prediction and adaptive optimization steps, using time-series analysis of historical data to predict future disturbance trends. Simultaneously, model parameters are automatically adjusted based on real-time feedback data to achieve continuous system optimization and adaptive learning.

[0064] In this embodiment, step S2 includes the following steps: S2-1. Perform time synchronization and spatial consistency checks on the radio signal information data collected by fixed monitoring stations and mobile monitoring stations. The radio signal information data includes spectrum data, AIS data, and environmental data. Specifically, in S2-1, the spatial consistency of the information data is checked using the following formula. If the condition is met, the data passes the consistency check; otherwise, it is marked as abnormal and the abnormal data is removed. in Representing the first Signal power and frequency of a fixed monitoring station Representing the first The signal power and frequency of each mobile monitoring station These are the power difference threshold and the frequency difference threshold, respectively. For the first The number of fixed monitoring stations to the first The distance between mobile monitoring stations.

[0065] These two inequalities determine whether the data from different monitoring stations are consistent; the power difference threshold varies with the distance between stations. The frequency difference increases linearly because the greater the distance, the greater the difference in propagation loss. The frequency difference threshold is fixed at 100Hz because frequency is not affected by distance; if the condition is met, the data passes the consistency check; otherwise, it is marked as abnormal. Optionally, this formula can be used to verify the reasonableness of the data when both the fixed station and the patrol boat detect the same interference signal simultaneously.

[0066] Specifically, the differences in timestamps of all data involved in the fusion Based on GPS clock synchronization, spatiotemporal alignment of data is ensured. Data exceeding this threshold is not included in the same batch fusion.

[0067] S2-2, Utilization The criteria and Mahalanobis distance are used to remove outlier data after testing; specifically, the following steps are included: 1) Single-dimensional anomaly detection (3σ criterion) It is the average of measurements from multiple stations, and σ is the standard deviation. Data points that deviate from the mean by more than three times the standard deviation are considered outliers. This is a classic statistical anomaly detection method, applicable to single parameters such as power or frequency.

[0068] 2) Multidimensional anomaly detection It is a multi-dimensional feature vector, such as frequency, power, bandwidth, and modulation method. It is a multidimensional mean vector, and Σ is the covariance matrix. It's the Mahalanobis distance, which takes into account the correlation between the dimensions. When hour, It is the number of feature dimensions, which is judged as an anomaly with a 95% confidence level.

[0069] S2-3. Dynamically assign data weights to each monitoring station according to the following formula: in, It is the calculated number of The data weight of each monitoring station Is this the first The equipment reliability coefficient of each monitoring station is the highest among fixed monitoring stations, which provide 24 / 7 monitoring and have high stability. When each monitoring station is a fixed monitoring station =0.9, the performance of the mobile monitoring vehicle equipment is comparable to that of the fixed monitoring station, but considering the change in location, the reliability is slightly reduced. When the monitoring station is a mobile monitoring vehicle =0.85, the patrol boat uses portable equipment and is greatly affected by the environment, the first When the monitoring station is a patrol boat =0.7, the miniaturization of drone equipment limits its endurance and stability, the first When a monitoring station is used by a drone =0.6; Is this the first The distance attenuation factor of each monitoring station is determined by the distance between the monitoring station and the monitored object, where n is the total number of monitoring stations; S2-4. Construct state equations based on the system evolution models of each monitoring station and observation equations based on the measurement models of each monitoring station. Then, fuse the predicted values ​​of the information data derived from the state equations with the actual measured values ​​of the monitoring stations using Kalman filtering to obtain optimal estimates, including power and frequency. Specifically, this includes the following steps: Constructing state equations (system evolution model): This equation describes how the state of the monitoring system changes over time. It is the system state at time k, including information such as frequency, power, and location of interference sources; It is a state transition matrix, which describes the natural evolution of states; It controls inputs such as known environmental changes. It is process noise, representing the uncertainty of the model.

[0070] Constructing the observation equation (measurement model): This equation describes the relationship between the actual measured values ​​and the true state of each monitoring station. It is the set of observation data from each monitoring station at time k. It is the observation matrix, which maps the state space to the observation space. This is observation noise, representing the error of the measuring equipment.

[0071] The state at the next moment is predicted by the state equation, and the actual measurement is obtained by the observation equation. Kalman filtering is then used to fuse the prediction and measurement to obtain the optimal estimate.

[0072] S2-5. The data from multiple monitoring stations that have collected the same radio signal information are weighted and averaged according to the following formula, and a spectrum situation diagram is output based on the fusion result. in, It is the fusion power value. It is the fusion frequency value. The data weights calculated for S2-3, For the first The optimal power estimate for each monitoring station. For the first The optimal frequency estimate for each monitoring station.

[0073] S2-6. The reliability of the fusion result is evaluated and classified according to the following formula; in, Represents the reliability assessment results. This indicates the percentage of valid sites participating in the integration out of the total number of sites. The complement representing the relative standard deviation of the fusion result. This indicates the spatial uniformity of the monitoring stations, ranging from 0.6 to 1.0. The more uniform the station distribution, the better. The larger; when and It is determined to be highly reliable and can be used to generate spectrum situation maps; when and It is determined to be of medium reliability, and other information is needed to help determine whether it can be used to generate a spectrum situation map; when or It was determined to be of low reliability and could not be used to generate a spectrum situation map, but additional monitoring resources were needed for the same radio signal information.

[0074] S2-7 also includes the step of adaptively optimizing and adjusting the fusion result according to the following formula; in, This is the current adaptive adjustment value of the output. This is the initially set baseline adjustment threshold. This is the ratio of recent fluctuations to baseline fluctuations. This ratio increases when current environmental disturbances intensify, and approaches 1 when the environment is stable. =0.3 is an adjustment coefficient that controls the adaptation speed. The threshold is dynamically adjusted according to the environment. When the environment is complex, the threshold is relaxed to avoid false alarms, and when the environment is stable, the threshold is tightened to improve sensitivity.

[0075] This embodiment also proposes a system for implementing the above-mentioned method for identifying and locating maritime radio interference through multi-source data fusion. The functional architecture specifically includes: The data acquisition layer is responsible for receiving multi-source heterogeneous data of radio signal information collected by fixed monitoring stations and mobile monitoring stations; The data transmission layer uses communication network technology to transmit the data acquired by the data acquisition layer to the data processing layer; The data processing layer standardizes and normalizes the multi-source heterogeneous data transmitted from the data transmission layer, eliminating redundancy and contradictions between data, and performs Kalman filtering fusion; it constructs a rule base and matches real-time acquired data against the rule base one by one, identifying abnormal signals through a multi-model collaborative architecture; it uses positioning technology to determine the direction and location of interference signals, pinpointing the location of the interference source; and... The data storage layer is responsible for storing monitoring data and processing results; classifying and querying data, distinguishing between real-time data and historical data, to facilitate later querying and analysis.

[0076] like Figure 3 As shown, in this embodiment, multi-source data is first collected by fixed monitoring stations and mobile monitoring stations. The data acquisition layer receives data from general monitoring units, dedicated maritime frequency band monitoring units, and AIS monitoring units in the fixed monitoring stations, as well as data collected by patrol boats, drones, and mobile monitoring vehicles in the mobile monitoring stations. This data is then transmitted to the data processing layer through the data transmission layer for multi-source data fusion processing, including the following steps: First, data spatiotemporal alignment. Specifically, pulse interference data from fixed monitoring stations is removed using the 3σ criterion and Mahalanobis distance. Linear interpolation fills in short-term missing data from drones. Then, spectrum data, AIS data, and environmental data are normalized to JSON format, and spatiotemporal alignment is achieved based on GPS clocks. Next, precision feature fusion is performed. Specifically, core features from each data source are extracted, a fusion model is constructed based on the Kalman filter algorithm, and state equation prediction and observation equation updates are used. Furthermore, data weights are dynamically allocated, and weighted average fusion ensures complementary coverage of each monitoring station. The fused spectrum situation map is output, providing a reliable data foundation for subsequent identification and positioning.

[0077] By collaboratively identifying interference signals using multiple models, extracting the features of the interference signals with high precision, performing cross-validation, calculating the location based on the fused data, and outputting the precise location result of the interference signal to support subsequent regulatory actions.

[0078] like Figure 4As shown in this embodiment, during the data acquisition stage of the above-mentioned intelligent identification and positioning system for radio interference on water, the data collected by each party is optimized by the data training and optimization module. The training and optimization module receives real-time measurement data of the monitored target, updates the model parameters of the module with the measured data as the label value, and outputs identification and positioning results after the real-time measurement data participates in the aforementioned multi-source data fusion processing, interference identification and positioning calculation. The data training and optimization module is optimized based on the results, so as to realize the continuous and automatic optimization of the data.

[0079] In this embodiment, suitable machine learning algorithms, such as Support Vector Machines (SVM), decision trees, and neural networks, are selected for signal classification and anomaly identification. Historical data can also be used to train the selected model, optimize parameters, adjust hyperparameters, and evaluate the model's accuracy and recall on the validation set. The trained model is validated using cross-validation and performance evaluation (such as AUC, accuracy, F1-score, etc.) to ensure its effectiveness. New monitoring data is collected regularly to update the training set and continuously optimize the model to adapt to changes in the spectral environment. An adaptive learning mechanism is incorporated, enabling the model to self-update based on new signal patterns and continuously improve recognition accuracy.

[0080] In this embodiment, abnormal signals are located and interference is predicted. A trained anomaly recognition model is used to compare real-time monitoring signals to identify anomalies such as frequency shifts, sudden power increases or decreases, and unknown modulation methods. Real-time monitoring data is matched with a historical normal signal feature database, and the model's output is used to determine whether the signal is abnormal. Once an abnormal signal is detected, the system records the time of the anomaly, its spectrum, device ID, and other information, and generates an alarm message.

[0081] In this embodiment, historical operating data of the equipment (such as years of use, operating time, and usage trends of surrounding spectrum) can also be used for interference prediction. The prediction model is applied to the current equipment to predict the risk of future interference. For example, the probability of interference is assessed by combining factors such as spectrum usage trends and equipment aging. If interference is predicted, an early warning is issued, and relevant parameters of the equipment (such as frequency, power, and years of operation) are recorded.

[0082] In summary, this embodiment constructs a fusion model based on the Kalman filter algorithm, dynamically allocates data weights, and outputs a fused spectrum situation map through state equation prediction and observation equation updating. This provides a reliable data foundation for subsequent identification and positioning. Through multi-model collaboration, accurate identification of maritime radio interference signals is achieved. Furthermore, using the maritime radio interference signal identification and positioning system as a core data node, it can provide spectrum environment data support for intelligent system applications such as "intelligent collision avoidance of ships" and "dynamic channel management," realizing the transformation of major waterway transportation from "passive supervision" to "active early warning."

[0083] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or basic characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects. The scope of the invention is defined by the appended claims rather than the foregoing description. Therefore, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.

[0084] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for multi-source data fusion-based radio interference identification and positioning on water, characterized in that, Comprise the following steps: S1, using mobile monitoring station plus 24 hours fixed monitoring station to collect radio signal information in the supervision water area; S2, extracting the frequency spectrum characteristics of the radio signal information collected by each monitoring station, predicting the trend change of the frequency spectrum characteristics through Kalman filtering algorithm and outputting the frequency spectrum trend chart; S3, constructing a rule base, screening out suspected abnormal signals from the frequency spectrum trend chart through threshold judgment and rule matching; S4, calling a support vector machine model, taking multi-dimensional static signal features as input, constructing a classification hyperplane through a radial basis kernel function, and distinguishing normal signals and abnormal signals from the suspected abnormal signals; S5, calling a decision tree model, taking multi-dimensional scene features as input, and identifying abnormal signals through scene judgment of suspected abnormal signals; S6, calling a convolutional neural network model, converting time domain signals into frequency spectrum charts through short time Fourier transform, extracting texture features through multi-layer convolution, and identifying abnormal signals from suspected abnormal signals; S7, cross verifying the abnormal signals identified by the model in combination with multi-source data from each monitoring station, screening out interference signals and positioning the interference source through TDOA.

2. The multi-source data fusion based radio interference identification and localization method according to claim 1, characterized in that, In S4 to S6, the support vector machine model outputs a probability score of the result being an abnormal signal as , the decision tree model outputs a probability score of the result being an abnormal signal as , and the convolutional neural network model outputs a probability score of the result being an abnormal signal as . After S7, the following steps are included, the probability scores of the abnormal signals output by the three models are dynamically weighted and summed according to the following formula, and a comprehensive abnormal probability score P is finally calculated: P= According to the comprehensive abnormal probability score P, a graded early warning is implemented, comprising the following steps: When 0.30 ≤ P < 0.50, it is determined that the abnormal condition is mild, a first-level warning is implemented, a warning message is sent and a first-level processing strategy is executed; When 0.50 ≤ P < 0.80, it is determined that the abnormal condition is moderate, a second-level warning is implemented, a warning message is sent and a second-level processing strategy is executed; When P ≥ 0.80, it is determined that the abnormal condition is serious, a third-level warning is implemented, a warning message is sent and a third-level processing strategy is executed.

3. The multi-source data fusion based radio interference identification and localization method according to claim 1, characterized in that, In S4, the support vector machine model takes 8-dimensional signal features of frequency offset, power spectral density peak value, modulation depth, modulation frequency stability, signal bandwidth occupation rate, duty cycle, signal-to-noise ratio and harmonic distortion coefficient as input, and the abnormal signals output by the support vector machine model include co-frequency interference signals and intermodulation interference signals, in S5, the decision tree model takes 4-dimensional scene features of signal occurrence period, geographical location type, associated ship type and peripheral equipment density as input, and in S6, the abnormal signals output by the convolutional neural network model include frequency hopping interference signals and narrowband burst interference signals.

4. The multi-source data fusion based radio frequency interference identification and positioning method for water surface according to claim 1, characterized in that, The fixed monitoring station is equipped with a spectrum analyzer, a 24-hour monitoring antenna, and an AIS device for 24-hour uninterrupted collection of radio signal information of the supervised water area, the mobile monitoring station includes a patrol ship, a mobile monitoring vehicle, and a drone; the patrol ship is equipped with a portable spectrum analyzer, a directional monitoring antenna, a satellite communication device, an AIS device, and a communication facility for monitoring the use of ship radio equipment during the cruise; the mobile monitoring vehicle is equipped with a spectrum monitoring device and a direction finding device for spectrum monitoring and direction finding in response to the supervision needs of the coastal port, the wind farm, and the area around the shore-based radio facility; the drone is equipped with a miniaturized spectrum monitoring module and a high-definition camera for spectrum data collection and image shooting in the monitoring blind area.

5. The multi-source data fusion based radio frequency interference identification and localization method according to claim 1, characterized in that, The S2 includes the following steps: S2-1, time synchronization verification and spatial consistency verification are performed on the radio signal information data collected by the fixed monitoring station and the mobile monitoring station, the radio signal information data including spectrum data, AIS data, and environmental data; S2-2, using Criteria and Mahalanobis distance will remove outliers after inspection; S2-3, a data weight is dynamically assigned to each monitoring station according to the following formula: in, It is the calculated number of The data weight of each monitoring station Is this the first The equipment reliability coefficient of the monitoring station, when the... When each monitoring station is a fixed monitoring station =0.9, for mobile monitoring vehicles =0.85, for cruising ships =0.7, for drones =0.6; Is this the first The distance attenuation factor of each monitoring station is determined by the distance between the monitoring station and the monitored object, where n is the total number of monitoring stations; S2-4, a state equation based on the system evolution model of each monitoring station and an observation equation based on the measurement model of each monitoring station are constructed, the predicted value of the information data derived from the state equation is subjected to Kalman filtering fusion with the actual measured value of the monitoring station, and an optimal estimated value including power and frequency is obtained; S2-5, the data of multiple monitoring stations collecting the same radio signal information are weighted and averaged according to the following formula, and a spectrum situation map is output according to the fusion result; wherein, is a fusion power value, is a fusion frequency value, is a data weight calculated for S2-3, is a power optimal estimate value for the monitoring station, is a frequency optimal estimate value for the monitoring station.

6. The multi-source data fusion based radio interference identification and localization method according to claim 5, characterized in that, Further comprising the following steps: S2-6, the reliability of the fusion result is evaluated according to the following formula, and the reliability is classified; wherein, represents the reliability evaluation result, represents the proportion of the number of effective sites participating in fusion to the total number of sites, represents the complement of the relative standard deviation of the fusion result, represents the spatial distribution uniformity of the monitoring station, taking 0.6-1.0, the more uniform the site distribution is, the greater; When and , it is determined as high reliability, and can be used to generate a spectrum situation map; When and , it is determined that the reliability is medium, and other information data needs to be combined to assist in determining whether it can be used to generate a spectrum situation map; When or , it is determined as low reliability, which cannot be used to generate the spectrum situation map, but monitoring forces are dispatched for the same radio signal information.

7. The multi-source data fusion based radio frequency interference identification and localization method according to claim 5, characterized in that, In S2-1, the information data is subjected to spatial consistency verification according to the following formula, if the condition is met, the data passes the consistency verification, otherwise, the abnormal data is marked and excluded; in Representing the first Signal power and frequency of a fixed monitoring station Representing the first The signal power and frequency of each mobile monitoring station These are the power difference threshold and the frequency difference threshold, respectively. For the first The number of fixed monitoring stations to the first The distance between mobile monitoring stations.

8. The multi-source data fusion based radio interference identification and localization method according to claim 7, characterized in that, Further comprising the step of adaptively optimizing and adjusting the fusion result according to the following formula; wherein, is the current output adaptive adjustment value, is the initial set reference adjustment threshold value, is the ratio of recent fluctuation to baseline fluctuation, when the current environmental disturbance intensifies, this ratio increases, when the environment is stable, this ratio is close to 1, is the adjustment coefficient, which controls the adaptive speed.

9. The multi-source data fusion based radio frequency interference identification and localization method according to claim 1, wherein, The abnormal signal identified by the S4 to S6 model is verified by AIS data association, whether there is a ship radio equipment matching the parameters of the abnormal signal, and at the same time, the historical database is retrieved and manually reviewed.

10. A system for implementing the method of radio interference identification and localization at sea based on fusion of multiple sources of data as claimed in claim 1, characterized in that, Comprise: The data acquisition layer is responsible for receiving the multi-source heterogeneous data of the radio signal information collected by the fixed monitoring station and the mobile monitoring station; The data transmission layer transmits the data collected by the data acquisition layer to the data processing layer by using communication network technology; The data processing layer standardizes and normalizes the multi-source heterogeneous data transmitted by the data transmission layer, eliminates the redundancy and contradiction between the data, and performs Kalman filtering fusion; a rule base is constructed, and real-time collected data are matched with the rule base one by one, and abnormal signals are identified through a multi-model collaborative architecture; the direction finding and positioning of the interference signal are performed by using positioning technology to determine the position of the interference source; And The data storage layer is responsible for storing the monitoring data and processing results; the data are classified and queried to distinguish real-time data and historical data, facilitating later query and analysis.

Images