A multi-agent collaborative method and device for electromagnetic spectrum monitoring and analysis

Abstract

The application provides a multi-agent collaborative method and device for electromagnetic spectrum monitoring and analysis, and belongs to the technical field of artificial intelligence.The method comprises the following steps: receiving a task demand, and acquiring interface information of an electromagnetic detection tool model; a first agent analyzes the task demand to generate a preliminary task intention; a task planning book containing multiple subtasks is generated based on the preliminary task intention and the interface information; a task scheduling module of a second agent calls the electromagnetic detection tool model to execute the subtasks, and acquires intermediate result information and final label information; based on the intermediate result information and the final label information, the second agent performs feature fusion and intention reasoning to generate an intention inference result of an electromagnetic signal and an evidence chain supporting the intention inference result.Based on the method, a multi-agent collaborative device for electromagnetic spectrum monitoring and analysis is also provided.The application improves the recall rate and robustness of spectrum analysis, and improves the abnormal detection rate.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of artificial intelligence, and particularly relates to a multi-agent collaborative method and device for electromagnetic spectrum monitoring and analysis. BACKGROUND

[0002] With the breakthroughs of large-scale language models in natural language understanding and complex task generalization, autonomous agent (Agentic AI) technology based on LLM has become a new direction in the field of artificial intelligence. The core feature of such agents is to independently complete complex multi-step tasks through autonomous perception, reasoning, planning and execution. The core technical problem focuses on how to accurately parse and structure high-level user instructions into executable sub-task sequences, and how to make credible final intent reasoning and decision-making based on heterogeneous and multi-modal intermediate results. Electromagnetic spectrum monitoring technology has wide application needs in the field of public safety, including but not limited to unmanned aerial vehicle black flight monitoring, spectrum compliance supervision, public event and large venue safety monitoring, key infrastructure electromagnetic safety protection, and emergency communication support, etc. With the popularity of wireless communication, video transmission, Internet of Things and unmanned systems in urban and industrial environments, the electromagnetic environment presents characteristics of multi-source, multi-protocol, multi-bandwidth, and non-stationary, which brings significant challenges to traditional monitoring and identification.

[0003] In general fields, autonomous agents based on LLMs exhibit strong planning capabilities in tasks such as dialogue and programming. However, when applied to highly specialized industrial and technical fields such as electromagnetic spectrum monitoring, the following technical bottlenecks are faced: First, lack of domain knowledge: general LLMs lack professional knowledge in the field of electromagnetic signals (such as modulation classification, protocol identification), making it difficult to understand professional terminology and generate task planning consistent with domain logic. Second, insufficient reasoning interpretability: traditional end-to-end detection models are a "black box", and their decision-making process lacks transparency, making it impossible to provide convincing evidence chains for reasoning, which is a fatal flaw in safety-critical public safety monitoring scenarios. Third, static task flow: existing electromagnetic monitoring systems mostly use pre-set, static analysis processes that cannot be adjusted and self-adaptively planned in real time according to dynamically changing electromagnetic environments and diverse user needs, with poor flexibility.

[0004] Therefore, there is an urgent need in the art for an intelligent processing method that can deeply integrate domain knowledge, have dynamic task planning, interpretable intent reasoning, and autonomous reflection optimization capabilities. SUMMARY

[0005] To solve the above technical problems, the present application proposes a multi-agent collaborative method and device for electromagnetic spectrum monitoring and analysis, which significantly improves the recall rate and robustness of spectrum analysis, increases the abnormal detection rate and reduces the false alarm rate.

[0006] To achieve the above object, the present application adopts the following technical solutions:

[0007] A multi-agent collaborative method for electromagnetic spectrum monitoring and analysis, comprising the following steps:

[0008] Receiving task requirements for electromagnetic spectrum monitoring and analysis, and obtaining interface information of an electromagnetic detection tool model;

[0009] Analyzing the task requirements by a preset first agent to generate a preliminary task intent, and generating a task planning document containing multiple subtasks based on the preliminary task intent and the interface information;

[0010] Through a task scheduling module of a second agent, calling the electromagnetic detection tool model to execute the subtasks of the task planning document, and obtaining intermediate result information and final label information generated by the electromagnetic detection tool model;

[0011] Based on the intermediate result information and the final label information, performing feature fusion and intent reasoning by a preset second agent to generate an intent inference result of the electromagnetic signal and an evidence chain supporting the intent inference result.

[0012] The present application also provides a multi-agent collaborative device for electromagnetic spectrum monitoring and analysis, comprising at least one processor and a memory, wherein the memory stores a computer program, and the computer program is executed by the at least one processor to implement the multi-agent collaborative method for electromagnetic spectrum monitoring and analysis.

[0013] The effects provided in the summary are only the effects of the embodiments, not all the effects of the present application, and one of the above technical solutions has the following advantages or beneficial effects:

[0014] The application provides a multi-agent collaborative method and device for electromagnetic spectrum monitoring and analysis, and belongs to the technical field of artificial intelligence.

[0015] The application can output high-level intent inference conclusions and evidence chains through semantic fusion and causal reasoning of the second agent on the results of the multi-tool model. This improves the analysis level from the physical layer and the signal layer to the cognitive layer and the decision layer, and provides high-value intelligence that can be directly used for decision-making for supervisors.

[0016] The application can learn from the execution results each time and automatically optimize the subsequent task planning strategy through the reflection mechanism. This enables the system to continuously evolve and become more intelligent, and significantly improves the robustness in unfamiliar and complex electromagnetic environments.

[0017] The functions of the two agents in the application are decoupled and interact through standardized interfaces. New electromagnetic detection tool models only need to register interface information, and can be identified and called by the system. This "plug and play" architecture enables the system to quickly integrate the latest analysis algorithms, easily expand to cope with new threats, and greatly reduces the maintenance and upgrade costs of the system. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 A multi-agent collaborative method flow chart for electromagnetic spectrum monitoring and analysis is provided for the embodiment 1 of the application;

[0019] Figure 2 An architecture diagram of the multi-agent collaborative method for electromagnetic spectrum monitoring and analysis is provided for the embodiment 1 of the application;

[0020] Figure 3 The user interaction interface of the electromagnetic signal detection platform for unmanned aerial vehicle black flight detection in the embodiment 1 of the application;

[0021] Figure 4 The multi-agent collaborative device schematic diagram for electromagnetic spectrum monitoring and analysis proposed in the embodiment 2 of the application. DETAILED DESCRIPTION

[0022] To make the technical features of the present application clear, the application will be described in detail below with specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples to implement the different structures of the application. In order to simplify the disclosure of the application, the components and settings of specific examples are described below. In addition, the application can repeatedly refer to numbers and / or letters in different examples. Such repetition is for the purpose of simplification and clarity, and does not indicate the relationship between the various embodiments and / or settings discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. The application omits the description of well-known components and processing techniques and processes to avoid unnecessary limitation of the application.

[0023] Embodiment 1

[0024] The embodiment 1 of the application proposes a multi-agent collaborative method for electromagnetic spectrum monitoring and analysis, which is used to solve the technical problems existing in the prior art of electromagnetic spectrum monitoring and analysis.

[0025] Figure 1 The multi-agent collaborative method for electromagnetic spectrum monitoring and analysis proposed in the embodiment 1 of the application is a flow chart; Figure 2 The architecture diagram of the multi-agent collaborative method for electromagnetic spectrum monitoring and analysis proposed in the embodiment 1 of the application is realized; combined Figure 1 and Figure 2 Introduce the process of execution.

[0026] In step S100, a user interaction interface is set at the front end, and an agent system is set at the back end. The user interaction interface includes: 1, a user input dialog box, in which the user inputs an electromagnetic spectrum signal task background, a set of tool models that can be called, and analysis requirements; 2, a tool model set display box, in which specific tool models uploaded or replaced by the user at the back end of the system can be displayed, so as to facilitate the user and the first agent to understand the types and quantities of the callable models; 3, a task sheet display box, which displays the task sheet generated by the first agent, the operation execution flow and steps of the second agent, and visually displays the electromagnetic spectrum signal analysis process; 4, an intention inference result display box, which displays the intention inference result and the reference task sheet for executing the task. The display box needs a user interaction dialog box, which can be used by the user to evaluate the correctness of the intention inference result and the correctness of the task sheet flow referred to by the result. The intention inference reflection module inside the system can feed back and update the internal parameters of the first agent according to the part of information, so as to better complete the subsequent tasks.

[0027] The agent system in the present application includes a first agent and a second agent;

[0028] The method for obtaining the first agent is: obtaining a first basic large language model; performing supervised fine-tuning on the first basic large language model using a first training data set to obtain the first agent; wherein the first training data set contains multiple task requirement texts in the electromagnetic monitoring field, and a structured task planning sheet corresponding to each task requirement text;

[0029] The method for obtaining the second agent is: obtaining a second basic large language model; performing supervised fine-tuning on the second basic large language model using a second training data set to obtain the second agent; wherein the second training data set contains multiple electromagnetic signal feature data, and an intention inference conclusion and evidence chain description corresponding to each feature data.

[0030] In the present application, the essence of the first agent and the second agent is a large language model (LLM), which has a powerful ability to process open language tasks. However, for the field of electromagnetic spectrum analysis autonomous analysis and cognitive confrontation, the large language model still has deficiencies in terms of field terminology and abbreviation recognition. Therefore, before the first agent and the second agent are put into use, the first agent and the second agent need to be fine-tuned using specific data sets.

[0031] In step S110, the overall task requirement in the form of text is received through the user interaction interface, and the interface information of the electromagnetic detection tool model is obtained.

[0032] First, deploy electromagnetic measurement equipment in the area to be monitored, collect raw data streams, and normalize the data. The raw data streams include I / Q streams, power spectra (PSD), short-time Fourier transform (STFT) frames, radar echoes, synchronized video frames, and acoustic data. Then normalize the raw data streams into tensor input types available to the numerous electromagnetic detection tool models deployed in the system.

[0033] Users describe the text form of the signal intent inference requirements for different scenarios and different periods of the captured signal. The interface information contains the functions of each tool model and the data type requirements of the input and output.

[0034] In step S120, the preliminary task intent is generated by parsing the task requirements through the preset first agent, and the task planning book containing multiple sub-tasks is generated based on the preliminary task intent and the interface information. The specific steps include:

[0035] The semantic understanding of the task requirements generates a preliminary task intent. The preliminary task intent includes task output targets and execution steps.

[0036] Based on the preliminary task intent and in combination with the functions described in the interface information, task decomposition and tool matching are performed.

[0037] Iterative optimization of the matching results generates a structured task planning book, where the task planning book defines the task type, input data source, required electromagnetic detection tool model, execution priority, and dependency relationship between sub-tasks for each sub-task.

[0038] In step S130, the task scheduling module of the second agent calls the electromagnetic detection tool model to execute the sub-tasks of the task planning book, and obtains the intermediate result information and final label information generated by the electromagnetic detection tool model. The specific steps include:

[0039] According to the dynamic task scheduling sequence, the normalized electromagnetic data is bound and called with the required electromagnetic detection tool model for the sub-tasks. The electromagnetic detection tool model includes automatic modulation classification model, protocol recognition model, tracking and positioning model, noise filtering model, and signal enhancement model. The electromagnetic detection tool model is also a set of sub-agents of the present application.

[0040] Monitor the execution status of each electromagnetic detection tool model and collect the output intermediate result information and final label information. The intermediate result information includes text descriptions or intermediate features in the reasoning process of the electromagnetic detection tool model, and the final label information includes classification, recognition, or positioning conclusions obtained by the electromagnetic detection tool model after reasoning on the input data.

[0041] In step S140, based on the intermediate result information and the final label information, a second preset agent is used for feature fusion and intention reasoning to generate an intention inference result of the electromagnetic signal and an evidence chain supporting the intention inference result; the specific steps include:

[0042] In the calculation graph of the intention reasoning of the second agent, the intention inference result is taken as a starting point, and the gradient of the output layer to the input layer is calculated by back propagation;

[0043] The gradient value on the input node corresponding to the final label information is extracted, and the absolute value of the gradient value is taken as a quantitative index of the contribution degree of the corresponding final label information to the intention inference result;

[0044] Based on the quantitative index, all final label information is sorted, and a preset number of evidences with the highest contribution degree are screened out.

[0045] In step S150, feedback information for the intention inference result is received; based on the intention inference result and / or the feedback information, the first agent is used to reflect on the process of generating the task planning book; and based on the reflection result, the internal parameters of the first agent are updated.

[0046] The process of generating the task planning book by the first agent is reflected, specifically: based on the intention inference result, the contribution degree of each final label information to the intention inference result is determined; based on the contribution degree, the evidences are screened out; the causal relationship between the evidences and the intention inference result is analyzed, and the task planning logic is adjusted based on the analysis result.

[0047] Based on the intention inference result, the contribution degree of each final label information to the intention inference result is determined, specifically: in the calculation graph of the intention reasoning of the second agent, the intention inference result is taken as a starting point, and the gradient of the output layer to the input layer is calculated by back propagation; the gradient value on the input node corresponding to the final label information is extracted, and the absolute value of the gradient value is taken as a quantitative index of the contribution degree of the corresponding final label information to the intention inference result; based on the quantitative index, all final label information is sorted, and a preset number of evidences with the highest contribution degree are screened out.

[0048] In this application, the planning strategy evolution of the first agent, the intention knowledge evolution of the second agent, and the tool set evolution of the task scheduling module of the second agent realize the cooperation of the first agent and the second agent.

[0049] The multi-agent cooperative method for electromagnetic spectrum monitoring and analysis proposed in embodiment 1 of the application can be applied to different scenes, and the process of the application in the field scene situation awareness will be introduced as follows:

[0050] Wild field broadband monitoring refers to short-time or continuous scanning of a large bandwidth spectrum in an open or semi-open field environment, with the purpose of quickly discovering and identifying unconventional emission events, inferring their intentions for timely intervention, and analyzing the underlying reasons.

[0051] First of all, in the selection of tool models, models that are more suitable for field scene situational awareness tasks should be selected, such as automatic modulation classification models that can quickly identify ASK / FSK / QAM modulation types and are resistant to noise. The protocol recognition model selects a field-specific wireless protocol recognition model to quickly match unconventional protocols and exclude civilian legal signals. The tracking and positioning model selects a portable TDOA positioning model and a signal strength triangulation model to quickly locate the source of emission with low power consumption and adapt to field terrain. In addition, noise filtering models and signal enhancement models are used to handle complex electromagnetic environment interference in the field and improve signal quality.

[0052] In the fine-tuning step of the first intelligent agent, the data set used for fine-tuning should contain more than 100 field scene task requirement texts and corresponding task book labels, for example:

[0053] Requirement text: "Scan in the forest protection area, find unknown high-frequency signals, suspect illegal surveying and mapping drones, need to quickly analyze their intentions", task book label:

[0054] {"plan":

[0055] [{"step":1,

[0056] "description":"Filter noise in the field environment and enhance the target signal",

[0057] "tool":"Noise filtering model",

[0058] "args":{"input_data":"raw I / Q stream in the field, PSD data"}},

[0059] {"step":2,

[0060] "description":"Identify signal modulation type",

[0061] "tool":"Lightweight CNN modulation classification model",

[0062] "args":{"items":"signal after step 1 processing"}},

[0063] {"step":3,

[0064] "description":"Match drone-specific protocol",

[0065] "tool":"Field Drone Protocol Recognition Model",

[0066] "args":{"modulation_type":"Result of Step 2"}},

[0067] {"step":4,

[0068] "description":"Location of signal transmitter",

[0069] "tool":"TDOA positioning model",

[0070] "args":{"signal_features":"Results of Step 3"}}]}

[0071] In the fine-tuning step for the second agent, the dataset required for fine-tuning contains over 500 field scenarios, namely paired data of "tool model output labels and intent results". Example: Input (tool model labels): {"Modulation type": "QAM-16", "Protocol type": "a certain radar's proprietary protocol", "Location result": "core area of ​​a forest reserve", "Signal strength": "medium persistence"}. Output (intent result): "The transmission intent is illegal mapping, with a medium-high risk level. It is recommended to immediately dispatch a patrol team to the location area for intervention." The fine-tuning process adopts a LoRA lightweight fine-tuning scheme, freezing the core parameters of the LLM, training only the electromagnetic spectrum domain-related layers, and controlling the training time to within 8 hours to adapt to rapid deployment requirements.

[0072] This application involves deploying portable electromagnetic measurement equipment in the field area to be monitored during data acquisition and demand deployment, specifically including:

[0073] Core equipment: broadband spectrum analyzer (bandwidth ≥ 200MHz, supports continuous scanning), portable I / Q data acquisition unit, solar power module (for outdoor scenarios without mains power).

[0074] Auxiliary equipment: Environmental sensors (monitoring temperature, humidity, and wind speed for subsequent signal interference analysis).

[0075] 4G communication module, satellite communication module (remote data transmission backup).

[0076] The data acquisition includes I / Q streams, power spectrum (PSD), short-time Fourier transform (STFT) frames, radar echoes (for UAV / aircraft signals), synchronized video frames (monitoring footage of the deployment area to assist in intent verification), and acoustic data.

[0077] The original data is normalized to the tensor type, with the dimension unified to [32, 32, 1], and the input requirements of each electromagnetic detection tool model are adapted. The original data and the normalized data are stored synchronously to the edge server, retaining 72 hours to meet the needs of field task backtracking. Users submit overall task requirements in text form in the system, which must include three core elements (example): "Monitoring area: Western mountainous area A section; Task goal: Quickly find unconventional unmanned aerial vehicle signals; Intervention requirement: Intended results and intervention suggestions within 1 hour". Users upload the current available field electromagnetic detection tool model interface information through the system, with the format example:

[0078] {"tool_name":"Lightweight CNN modulation classification model",

[0079] "function":"Identify anti-interference signal modulation type in the field",

[0080] "input_type":"tensor[32, 32, 1]",

[0081] "output_type":"Text (modulation type + confidence)",

[0082] "response_time":"≤0.5 seconds"}

[0083] The first agent captures user demand text and tool model information, quickly understands the semantics through the fine-tuned LLM, and generates the initial task intent:

[0084] Task output goal: Complete unconventional unmanned aerial vehicle signal identification, positioning, and intent inference in the western mountainous area A section within 1 hour. Brief execution steps: Signal filtering → modulation classification → protocol identification → positioning → intent inference.

[0085] The first agent iteratively optimizes the initial intent and generates a JSON format sub-task sequence (task book) that highlights the priority, timeout threshold, and dependency relationship in the field scenario; example:

[0086] {"task_book_id":"Field monitoring",

[0087] "overall_objective":"Unconventional unmanned aerial vehicle signal intent inference in the western mountainous area A section",

[0088] "sub_tasks":

[0089] [{"task_id":"T1",

[0090] "task_type":"Signal preprocessing",

[0091] "input_data_index":"Data Acquisition Device 1-20240520-0900",

[0092] "preprocessing": "Noise filtering + signal enhancement",

[0093] "tool":"Noise Filtering Model",

[0094] "priority":"highest",

[0095] "timeout_threshold":"10 minutes",

[0096] "dependencies":[]},

[0097] {"task_id":"T2",

[0098] "task_type":"modulation classification",

[0099] "input_data_index":"Output result of T1",

[0100] "preprocessing": "None",

[0101] "tool":"Lightweight CNN modulation classification model",

[0102] "priority":"high",

[0103] "timeout_threshold":"15 minutes",

[0104] "dependencies":["T1"]},

[0105] {"task_id":"T3",

[0106] "task_type":"protocol identification",

[0107] "input_data_index":"T2 output result",

[0108] "preprocessing": "Protocol feature extraction",

[0109] "tool":"Field Drone Protocol Recognition Model",

[0110] "priority":"high",

[0111] "timeout_threshold":"15 minutes",

[0112] "dependencies": ["T2"]},

[0113] {"task_id": "T4",

[0114] "task_type": "Positioning",

[0115] "input_data_index": "Output result of T3 + GPS information of the acquisition device",

[0116] ​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​

[0131] The reasoning process is: feature fusion: the modulation type, protocol type, positioning result, signal strength and other features are weighted and fused.

[0132] Intention matching: based on the fine-tuned LLM, the field scene intention library (including illegal mapping, signal interference and illegal communication and other typical intentions) is matched.

[0133] The final result is: "the signal emission intention is illegal mapping, and the risk level is medium and high".

[0134] Evidence chain generation: output the reasoning process and core evidence, example: "1. The modulation type is QAM-16 (confidence 92%), which meets the characteristics of high-definition mapping signals of unmanned aerial vehicles; 2. The protocol matches a certain illegal unmanned aerial vehicle private protocol (confidence 88%); 3. The positioning result is located in the core protection area of the mountainous area (prohibited mapping area); 4. The signal duration is 30 minutes, and there is no legal record mark".

[0135] The scope of protection of the present application is not limited to specific numerical values in the process of field scene situation awareness, and those skilled in the art can make reasonable choices according to actual conditions.

[0136] The execution logic of the intention inference reflection module is: first, receive the intention inference result and user feedback, then reflect the task book generation process, if the intention inference is accurate, then strengthen the corresponding sub-task sequence logic (such as retaining the "signal filtering → modulation classification → protocol identification" process), if there is deviation (such as not identifying anti-interference signals), then optimize the task book (such as adding the "anti-interference signal detection" sub-task), finally, update the parameters, fine-tune the first intelligent agent parameters through gradient descent, and the update cycle is once a day (wild batch task iteration optimization); the evidence chain saving part first carries out reverse gradient calculation, that is, the reflection module calculates the contribution gradient of each input label to the result based on the intention reasoning result (such as "protocol type label contribution gradient 0.7, TOP-1"), then carries out evidence chain saving work, first automatically selects the top-3 labels with the highest contribution degree (such as protocol type, positioning result, modulation type), saves to the edge server, format is "task book ID + label information + contribution gradient + causal relationship", then carries out manual review, exports evidence chain data every week, and domain experts analyze the correctness of the causal relationship, if correct, then strengthen the reasoning path, if incorrect, then correct the task book execution steps (such as adjusting the priority of sub-tasks), at the same time, the saving period is clear, the core evidence chain is permanently saved, and the original data is saved for 30 days.

[0137] The multi-agent collaborative method for electromagnetic spectrum monitoring and analysis proposed in embodiment 1 of the present application can be applied to different scenarios. The process of the present application in unmanned aerial vehicle black flight detection is introduced as follows: the core goal is to realize real-time capture, rapid identification, accurate positioning and intention inference of unmanned aerial vehicle black flight signals within 5-10 kilometers around the airport clearance area and its periphery through multi-agent collaboration, to provide timely intervention basis for the airport air traffic control department, and to avoid black flight unmanned aerial vehicles from interfering with flight takeoff and landing and threatening aviation safety.

[0138] First, build an electromagnetic spectrum intention reasoning multi-agent system. The system architecture needs to be adapted to the high-density signal environment and safety control requirements of the airport. The front-end user interface combines the design of the airport air traffic control command center dedicated terminal and the mobile monitoring APP. The terminal supports multi-screen split display, and the APP provides real-time reception of early warning information for patrol personnel. Both have offline caching functions to deal with temporary network interruption scenarios at the airport. The core modules that the interface needs to include are as follows: the electromagnetic detection tool model display window needs to be labeled with "high sensitivity", "low false alarm rate", and "airport special" tags, and the models for unmanned aerial vehicle signal capture, protocol analysis, and accurate positioning are highlighted. The window will simultaneously display the current available interface information of the airport electromagnetic detection tool model. The format is as follows:

[0139] {"tool_name":"enhanced CNN modulation classification model",

[0140] "function":"identify 2.4GHz / 5.8GHz frequency band unmanned aerial vehicle signal modulation type, exclude civil aviation communication interference",

[0141] "input_type":"tensor[64,64,1]",

[0142] "output_type":"text (modulation type + confidence + interference exclusion explanation)",

[0143] "response_time":"≤0.3 seconds",

[0144] "applicable_scenario":"airport runway periphery, apron airspace signal detection",

[0145] "maintenance_status":"normal use"},

[0146] Figure 3The user interaction interface of the electromagnetic signal detection platform in the unmanned aerial vehicle black flight detection of embodiment 1 of the application; the interactive dialog box window supports text input and quick selection of air traffic control standard instructions, such as "suspected black flight signal near the entrance of runway No. 1", and has an emergency warning pop-up window function, and a red pop-up window is prompted when a high-risk signal is triggered; the task book display window is presented in a simple text format, highlights "task time limit requirement" and "intervention priority", and simultaneously displays the airport unmanned aerial vehicle black flight detection task book generated by the system in a complete manner, and an example of the format is as follows:

[0147] {"tool_name":"enhanced CNN modulation classification model",

[0148] "function":"identify the modulation type of the 2.4GHz / 5.8GHz frequency band unmanned aerial vehicle signal at the airport, and exclude civil aviation communication interference",

[0149] "input_type":"tensor[64,64,1]",

[0150] "output_type":"text (modulation type + confidence + interference exclusion explanation)",

[0151] "response_time":"≤0.3 seconds",

[0152] "applicable_scenario":"airport runway perimeter, apron airspace signal detection",

[0153] "maintenance_status":"normal use"},

[0154] The interactive dialog box window supports text input and quick selection of air traffic control standard instructions, such as "suspected black flight signal near the entrance of runway No. 1", and has an emergency warning pop-up window function, and a red pop-up window is prompted when a high-risk signal is triggered; the task book display window is presented in a simple text format, highlights "task time limit requirement" and "intervention priority", and simultaneously displays the airport unmanned aerial vehicle black flight detection task book generated by the system in a complete manner, and an example of the format is as follows:

[0155] {"task_book_id":"airport-drone-detection-20240610-001",

[0156] "overall_objective":"identify, locate and infer the intention of the black flight unmanned aerial vehicle signal above the north side of runway No. 3 at the airport to ensure the safety of the 10:30 take-off flight",

[0157] "task_initiator":"airport air traffic control command center",

[0158] "initiate_time":"2024-06-1009:50:00",

[0159] "required_completion_time":"2024-06-1010:00:00",

[0160] "sub_tasks":

[0161] [{"task_id":"T1",

[0162] "task_type":"signal preprocessing",

[0163] "input_data_index":"Airport Base Station 3-20240610-0948 (Data Collection Point on the North Side of Runway 3)"

[0164] "preprocessing": "Airport radar interference filtering + drone weak signal enhancement",

[0165] "tool":"Airport Signal Noise Reduction Model",

[0166] "priority":"highest",

[0167] "timeout_threshold":"3 minutes",

[0168] "dependencies":[],

[0169] "responsible_module":"Sub-smart agent set - Signal preprocessing module"},

[0170] {"task_id":"T2",

[0171] "task_type":"modulation classification",

[0172] "input_data_index":"T1 output result (preprocessed I / Q stream)",

[0173] "preprocessing":"Signal frequency band focus (2.4GHz / 5.8GHz)",

[0174] "tool":"Enhanced CNN Modulation Classification Model",

[0175] "priority":"highest",

[0176] "timeout_threshold":"4 minutes",

[0177] "dependencies":["T1"],

[0178] "responsible_module":"sub-agent set - modulation classification module",

[0179] "output_requirement":"explicit modulation type (FSK / QAM, etc.), confidence ≥ 90%, interference exclusion explanation",

[0180] {"task_id":"T3",

[0181] "task_type":"protocol identification",

[0182] "input_data_index":"T2 output results (modulation type + signal features)",

[0183] "preprocessing":"protocol feature extraction + civil aviation legal protocol comparison",

[0184] "tool":"airport drone protocol identification model",

[0185] "priority":"highest",

[0186] "timeout_threshold":"4 minutes",

[0187] "dependencies":["T2"],

[0188] "responsible_module":"sub-agent set - protocol identification module",

[0189] "output_requirement":"protocol type, brand, legal record status, suspicious feature description",

[0190] {"task_id":"T4",

[0191] "task_type":"precise positioning",

[0192] "input_data_index":"T3 output results (protocol information) + base station 3 / 4 / 5 signal arrival time + monitoring camera coordinates (number C12)",

[0193] "preprocessing":"positioning data calibration (exclude building obstruction errors)",

[0194] "tool":"TDOA+AOA fusion positioning model",

[0195] "priority":"high",

[0196] "timeout_threshold":"5 minutes",

[0197] "dependencies":["T3"],

[0198] "responsible_module":"sub-agent set - positioning module",

[0199] "output_requirement":"positioning coordinates (latitude and longitude), error range ≤10 meters, surrounding geographical identification (such as "Runway 3 end north 1.2 kilometers, close to the protective forest")",

[0200] {"task_id":"T5",

[0201] "task_type":"intention inference",

[0202] "input_data_index":"T2 (modulation type) + T3 (protocol information) + T4 (positioning result) + flight plan (10:30 departure flight HU7801)",

[0203] "preprocessing":"feature weighted fusion (positioning and flight association degree weight highest)",

[0204] "tool":"second agent",

[0205] "priority":"highest",

[0206] "timeout_threshold":"4 minutes",

[0207] "dependencies":["T4"],

[0208] "responsible_module":"second agent",

[0209] "output_requirement":"black flight type, threat level, impact range, intervention suggestion, reasoning evidence chain"}],

[0210] "emergency_contact":"Airport Security Department Li worker",

[0211] "backup_plan":"If a tool model fails, automatically switch to the backup model (e.g. switch from T2 to the airport-specific SVM modulation classification model)".

[0212] The reflection process display window focuses on the deviation points between the task book and the actual detection results, such as "some new type of black flying unmanned aerial vehicle protocol is not covered, and model training needs to be supplemented"; the intention inference result display window highlights "black flying type", "threat level" and "suggested intervention measures", such as "consumer black flying unmanned aerial vehicle - high threat - immediately start directional interference", and is associated with geographic information such as airport runway and parking apron to assist air traffic control personnel in quickly judging the impact range.

[0213] Both the first and second agents are based on high-performance LLM deployment, and the model version supporting millisecond-level response is selected for deployment on the airport local server cluster to avoid cloud transmission delay, and a backup server is configured to ensure uninterrupted system operation. The sub-agent set needs to filter electromagnetic detection tool models suitable for the airport scene, among which the automatic modulation classification model prioritizes enhanced CNN models that support multi-band unmanned aerial vehicle signal recognition, which can accurately identify the 2.4GHz and 5.8GHz frequency band signal modulation types commonly used by consumer and industrial unmanned aerial vehicles; the protocol recognition model needs to include an airport-specific unmanned aerial vehicle protocol library, covering private protocols of major black flying unmanned aerial vehicle brands and custom protocols of illegal modified devices, and can quickly exclude civil aviation legal communication, airport ground equipment and other interference signals; the tracking and positioning model uses the TDOA+AOA fusion positioning technology of multiple base stations, combined with the coordinates of the monitoring cameras deployed around the airport, to achieve accurate positioning of black flying unmanned aerial vehicles within 10 meters; the supplementary tool model needs to include an unmanned aerial vehicle signal noise reduction model and a multi-signal separation model, the former is used to filter electromagnetic interference generated by airport runway radars and ground navigation equipment, and the latter is used to separate signals when multiple black flying signals are captured simultaneously to avoid mutual interference affecting the detection results.

[0214] The first agent fine-tuning dataset needs to include more than 1500 airport air traffic control personnel actual task demand texts and corresponding task book labels, and the demand text example is "unidentified flying object found above the 3rd end of the airport runway, suspected to be a black flying unmanned aerial vehicle, needs to confirm the signal type, locate the position and judge whether it affects the landing of the flight within 5 minutes", the corresponding label task book needs to include the complete process of signal capture, modulation classification, protocol identification, positioning and intention inference, and clearly specifies the time requirements of each link, and its format can refer to the above-mentioned example of airport unmanned aerial vehicle black flying detection task book.

[0215] The second intelligent agent fine-tuning dataset needs to include more than 800 pairs of "tool model output label + intent result" data. The input example is "modulation type: FSK, protocol type: certain consumer drone private protocol; positioning result: 1 km north of runway 3; signal strength: strong; flight height: 300 meters; associated flight: 10:30 take-off HU7801", and the output intent result needs to be clear "black flight type, threat level, impact range, and intervention suggestion". The fine-tuning execution adopts an incremental training combined with LoRA scheme, which preserves the LLM basic language understanding ability while strengthening the identification of airport electromagnetic signal field terms and black flight scene task logic.

[0216] Field data collection needs to adapt to the high-density signal environment of the airport and the all-weather monitoring requirements. In terms of equipment deployment, within a range of 5-10 kilometers around the airport clearance area, fixed electromagnetic measurement base stations are deployed every 2 kilometers, equipped with ultra-wideband spectrum analyzers (bandwidth covering 100MHz-6GHz), high-gain directional antennas, and 24-hour uninterrupted power supply systems. Portable mobile monitoring equipment is deployed at both ends of the runway and around the parking apron, carried by patrol personnel to meet the detection needs of the blind area of fixed base station signal coverage; auxiliary equipment needs to include weather sensors to monitor the impact of weather factors such as wind speed and precipitation on the flight trajectory of unmanned aerial vehicles, as well as 4G / 5G dual-link communication modules to ensure real-time transmission of collected data to the local server of the airport, and satellite communication backup links to deal with ground communication interruptions caused by extreme weather.

[0217] The data collection content needs to focus on covering the commonly used signal types of black flight unmanned aerial vehicles, and the core data includes I / Q stream (sample rate increased to 5MSps to ensure the capture of high-speed unmanned aerial vehicle signals), power spectrum (PSD), and short-time Fourier transform (STFT) frames (frame length shortened to 5ms to improve signal real-time capture efficiency). The collection time, device coordinates, and corresponding airport flight schedule need to be recorded synchronously to facilitate subsequent correlation analysis of the impact of black flight signals on flights; supplementary data needs to include airport monitoring camera synchronous video frames (used to verify the matching of black flight unmanned aerial vehicle visual features and electromagnetic signal features), airport air traffic control radar data (excluding civil passenger aircraft and legal navigation aircraft signal interference), and ground personnel voice feedback (such as the location and appearance description of black flight unmanned aerial vehicles found by patrol personnel).

[0218] Data normalization processing must meet the airport's requirements for data consistency across multiple devices. The preprocessing module built into the acquisition device must normalize the raw data collected from different base stations and different types of devices into tensor types (with dimensions uniformly set to [64, 64, 1]), and simultaneously label the signal strength, signal-to-noise ratio, and interference level during data acquisition to facilitate subsequent tool models adjusting detection parameters based on data quality. Data storage adopts a three-tiered storage scheme of "real-time transmission + local backup + cloud archiving." Real-time data is transmitted to the airport command center server for immediate analysis, local backup retains 7 days of data for retrospective querying, and cloud archiving retains 3 months of data for model optimization and incident tracing. All data must be stored encrypted in accordance with airport information security management regulations.

[0219] User requests are submitted through a dedicated terminal at the airport air traffic control command center. Air traffic controllers can choose preset task templates (such as "detecting unauthorized flights around the runway", "investigating signals over the apron", and "focusing on monitoring during flight take-off and landing periods") or manually enter custom requests. The request text must include the monitoring area (accurate to the specific location of the airport), task timeframe (such as "outputting preliminary results within 3 minutes"), and key focus (such as "prioritizing the identification of modified drone signals").

[0220] The input of tool model interface information is the responsibility of airport technical maintenance personnel. Each time a tool model is added or updated, the model interface information must be uploaded to the system. In addition to the core tool model interface example mentioned above, the format of the newly added model interface information is as follows:

[0221] {"tool_name":"Multi-signal separation model",

[0222] "function":"Simultaneously separates signals from multiple unauthorized flights at airports to avoid signal superposition interfering with detection results",

[0223] "input_type":"Mixed I / Q streams + signal quantity estimation",

[0224] "output_type":"Single-target I / Q stream (sorted by signal strength) + separation confidence",

[0225] "response_time":"≤0.8 seconds",

[0226] "applicable_scenario": "Scenario in which multiple unauthorized flying targets appear simultaneously around the airport",

[0227] "maintenance_status":"Normal and available",

[0228] "version":"V2.1",

[0229] "update_time":"2024-05-20"}

[0230] The system automatically checks the compatibility of the model interface with the existing system, generates available, to-be-tested, and unavailable case status identifiers, and records the update time and maintenance person of the model, facilitating subsequent troubleshooting.

[0231] Task analysis needs to highlight the time priority and safety orientation of the airport scene. The first intelligent agent captures the task demand text submitted by air traffic control personnel and the current available tool model information in real time, quickly completes semantic understanding through the fine-tuned LLM, and focuses on extracting key information such as "whether the monitoring area involves the runway or the parking apron", "whether the task time limit is urgent", "whether it is related to the flight departure plan", etc. to generate the initial task intent. The initial task intent needs to clarify the task output target and brief execution steps. The output target needs to quantify the time limit and accuracy requirements, such as "complete the modulation classification and protocol identification of the signal above runway 1 within 2 minutes, with a positioning error of no more than 20 meters"; the brief execution steps need to adjust the order according to the task urgency, with urgent tasks (such as detecting black flying signals before flight departure) prioritizing signal capture and identification, and non-urgent tasks (such as daily inspection) adding signal feature storage and model training data accumulation steps, for example, the initial execution steps of the daily inspection task are "signal preprocessing → modulation classification → protocol identification → feature storage → abnormal signal marking", while the initial execution steps of the urgent task are "signal preprocessing (priority) → modulation classification (parallel) → protocol identification (parallel) → positioning → intent inference".

[0232] After the second intelligent agent's task scheduling module reads the subtask sequence of the airport unmanned aerial vehicle black flying detection task, it first analyzes the priority and dependency of each subtask, and adopts the high-priority task preemptive execution and independent task parallel coordination strategy according to the principles of safety priority and time efficiency in the airport scene. For example, in the "emergency black flying detection on runway 3" task, when T1 (signal preprocessing, highest priority) is executed, the system will simultaneously load the enhanced CNN modulation classification model parameters required by T2 (modulation classification) and the airport dedicated protocol library, and start the signal synchronous receiving channels of base stations 3 and 4; after T1 completes signal noise reduction and screening, T2 (modulation classification) and T3 (protocol identification) are triggered for parallel execution, at which time the task scheduling module will monitor the algorithm power occupation of the two subtasks in real time. If T2 is short of algorithm power, it will temporarily allocate spare server resources to ensure that the single subtask timeout rate is less than 1%. For daily inspection tasks, independent subtasks can be flexibly staggered to avoid resource waste and ensure the completion of the overall task within the required time.

[0233] The second intelligent agent focuses on collecting intermediate result information and final label information of each tool model when reasoning about the airport drone black flight intention. The intermediate results need to include key data specific to the airport scene, such as the T2 (multi-band modulation classification) intermediate result: "2.4 GHz frequency band modulation type is FSK, confidence 95%; 5.8 GHz frequency band has no valid signal, interference signal accounts for 12% (mainly for airport ground communication)"; T3 (protocol identification and legal verification) final label information: "protocol type matches a certain consumer-level black flight drone private protocol (no civil aviation record identification), attribution subject unknown, signal duration has reached 18 minutes, and moves towards the runway 1 end direction".

[0234] The reasoning process optimizes the design around airport security needs. First, feature fusion is performed, which includes weighting and fusing modulation type, protocol legality, positioning result, signal moving trajectory, and associated flight dynamics. Among them, the "distance between positioning result and runway or flight route", "protocol legality record state", and "signal moving direction" have the highest weight, which adapts to the core judgment needs of the airport on whether the flight is threatened, such as in the runway perimeter black flight detection task. If the positioning result shows that the black flight signal is only 200 meters away from the runway center line and is moving along the flight landing route, the feature weight will temporarily increase to 35%, triggering high-risk early warning logic preferentially. Then, intent matching is carried out based on the LLM fine-tuned by airport black flight data, which matches the airport exclusive intent library. This library contains typical intents such as "malicious interference with flight take-off and landing", "illegal shooting of airport facilities", "illegal transportation", etc. At the same time, it is associated with the airport security level (such as during the Spring Festival peak and important flight guarantee period, the matching threshold of "malicious interference" intent is reduced by 15%). Finally, the final result and evidence chain generation are completed, and the output result needs to be clear about "black flight intent type, threat level, impact range, and intervention measures". The evidence chain needs to include compliance and relevance data specific to the airport scene, such as: the intent inference result is "the signal emission intent is to maliciously interfere with flight take-off and landing, with a high threat level"; the reasoning evidence chain is "1. The modulation type is FSK, which meets the characteristics of unmanned aerial vehicle remote control signals, and the signal strength continues to increase, indicating that the unmanned aerial vehicle is approaching the airport; 2. The protocol matches a modified black flight unmanned aerial vehicle private protocol, which has no Civil Aviation Bureau record identifier, and contains "anti-interference frequency hopping" instructions in the protocol, which has the characteristics of deliberately avoiding monitoring; 3. The positioning result is located at the east side of runway 1 by 300 meters, and the moving trajectory prediction shows that it will enter the runway airspace in the next 2 minutes; 4. Associated flight dynamics: HU7801 flight (10:30 landing) has entered the five-edge approach phase, with a current height of 1000 meters, and the black flight signal may interfere with the flight navigation system; 5. The signal duration is 22 minutes, during which there is no legal signal interaction"; the intervention suggestion is "immediately start the runway 1 end directional interference device to cut off the unmanned aerial vehicle remote link; notify the airport security patrol team to go to the positioning coordinates with a anti-unmanned aerial vehicle capture net; synchronize with the air traffic control department and suggest that HU7801 flight temporarily reflies after the interference is removed".

[0235] The intention inference reflection module is embedded in the first intelligent agent. The execution logic of the intention inference reflection module in the airport scene needs to be optimized in combination with the air traffic control personnel feedback and the airport safety event handling results: first, receive the intention inference results and multi-dimensional feedback, including the effectiveness evaluation of the intervention measures submitted by the air traffic control personnel through the command terminal, the on-site disposal report of the security department, and the flight operation record; then the reflection task book generation process, if the intention inference is accurate and the intervention measures are effective (such as successfully avoiding flight interference), the corresponding sub-task sequence logic is strengthened, for example, the emergency task flow of "signal preprocessing → parallel modulation classification and protocol identification → positioning → trajectory prediction → intention inference" is retained, and the characteristic data of this case (such as "anti-interference frequency hopping protocol feature" and "approach lane black flight trajectory") is added to the first intelligent agent fine-tuning data set; if there is a deviation (such as mistakenly judging a legal plant protection unmanned aerial vehicle as a black flight, resulting in temporary closing of the apron), the deviation reason is analyzed in depth, if the task book does not include "airport surrounding legal operation unmanned aerial vehicle reporting list", the "T3-1 (legal reporting comparison)" step is added in the sub-task sequence, and the airport air traffic control reporting database interface is called for secondary verification; if the tool model has low recognition accuracy for "plant protection unmanned aerial vehicle special protocol", the model needs to be supplemented and the standby model calling rule is updated.

[0236] The parameter updating link adapts to the characteristics of 24-hour airport operation, selects the flight take-off and landing gap, fine-tunes the first intelligent agent parameters through gradient descent, and focuses on optimizing the "airport different area sub-task priority allocation", "flight correlation feature weight", "new type of black flight protocol identification logic" and other modules. After each parameter update, 10 groups of historical typical case tests need to be performed to ensure that the task book generation accuracy is not less than 95%, and then it can be put into use the next day.

[0237] The evidence chain preservation strategy needs to meet the requirements of airport security event tracing and civil aviation supervision. First, the reverse gradient calculation is performed. The reflection module calculates the contribution gradient of each input label to the intention inference based on the difference between the intention reasoning result and the actual disposal result. For example, in the case of malicious interference with the flight, the "positioning result and runway distance" label contributes a gradient of 0.85 (TOP-1), the "protocol anti-interference feature" label contributes a gradient of 0.78 (TOP-2), and the "flight approach state" label contributes a gradient of 0.72 (TOP-3). Then, the evidence chain preservation work is carried out. The top 3 labels with the highest contribution degree are automatically selected, combined with the sub-task execution log, air traffic control instruction record, and on-site disposal video segment index, and sorted into structured data according to the "airport security event evidence specification". The format example is as follows: "task book ID: airport-drone-detection-20240610-001; label information 1: positioning result (300 meters east of runway No. 1, in the approach channel), contribution gradient 0.85, causal relationship: this position directly threatens the safety of the landing flight, and is the core basis for determining the "malicious interference" intention; label information 2: protocol type (a modified black fly private protocol containing anti-interference frequency hopping instructions), contribution gradient 0.78, causal relationship: the protocol has features that deliberately avoid monitoring, excluding the possibility of "mistaken entry"; label information 3: flight status (HU7801 flight five-edge approach, height 1000 meters), contribution gradient 0.72, causal relationship: the flight is in a critical flight phase, and the black fly signal interference risk is extremely high; preservation time: 2024-06-1011:30:00; storage location: airport security database-black fly event partition". Artificial review is performed by the airport security management department and the air traffic control technology team. The evidence chain data of all black fly detection tasks in this week is exported from 16:00 to 18:00 every Friday, and the compliance of the intention inference logic with the civil aviation safety specification, the compliance of the intervention measures with the airport emergency disposal process, and the completeness and traceability of the data record are checked. If the evidence chain is verified correctly, the reasoning path is included in the airport black fly intention inference optimal practice library, and the same type of task is preferentially called in the future; if there is an error, the task book execution steps are corrected (such as adding a suspicious feature labeling sub-task after T4), and the data collection link person is held accountable to ensure the integrity of the evidence chain. The preservation period strictly follows the supervision requirements of the Civil Aviation Administration: the core evidence chain is permanently preserved, the ordinary black fly event evidence chain is preserved for 5 years, the original collected data is preserved for 90 days, and after the expiration, the destruction record needs to be retained for 10 years for inspection.

[0238] The scope of protection of the present application is not limited to specific numerical values in the process of unmanned aerial vehicle black fly detection, and those skilled in the art can make reasonable choices according to actual conditions.

[0239] The scope of protection of the present application is not limited to the two examples of field scene situation awareness and unmanned aerial vehicle black flight detection listed in embodiment 1, and a person skilled in the art can make reasonable choices according to actual conditions.

[0240] The multi-agent collaborative method for electromagnetic spectrum monitoring and analysis proposed in embodiment 1 of the present application decomposes the overall task into multiple sub-tasks such as perception, classification, feature fusion, and anomaly detection, and processes them in parallel or series by the encapsulated second agent, thereby overcoming the performance decline of a single end-to-end model when facing multi-source, multi-format, and non-steady signals, significantly improving the recall rate and robustness of spectrum analysis, and improving the detection rate and reducing the false alarm rate of anomaly detection.

[0241] Embodiment 2

[0242] The present application also proposes a device, Figure 4 The present application also proposes a device,

[0243] At the hardware level, the electronic device 400 includes a processor 410, and optionally, an internal bus 420, a network interface 430, and a memory. The memory can include a memory such as a high-speed random access memory (RAM), and can also include a non-volatile memory such as at least one disk memory. Of course, the electronic device can also include other hardware required by the business.

[0244] The processor 410, the network interface 430, and the memory can be connected to each other through the internal bus 420, which can be an industry standard architecture (ISA) bus, a peripheral component interconnect (PCI) bus, or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one bidirectional arrow is used in the figure, but it does not mean that there is only one bus or one type of bus. The memory is used to store programs. Specifically, the program can include program code, and the program code includes computer operation instructions. The memory can include a memory 440 and a non-volatile memory 450, and provide instructions and data to the processor 410.

[0245] The processor 410 reads the corresponding computer program from the non-volatile memory 450 into the memory 440 and then runs, and forms a device for positioning a target user at a logical level. The processor 410 executes the program stored in the memory, and is specifically used for executing:

[0246] In step S100, a user interaction interface is set at the front end, and an agent system is set at the back end.

[0247] In step S110, a general task requirement in a text form is received through the user interaction interface, and interface information of an electromagnetic detection tool model is acquired.

[0248] In step S120, a first agent is used to analyze the task requirement to generate a preliminary task intention, and based on the preliminary task intention and the interface information, a task planning document containing a plurality of subtasks is generated.

[0249] In step S130, a task scheduling module of a second agent is used to call the electromagnetic detection tool model to execute the subtasks of the task planning document, and intermediate result information and final label information generated by the electromagnetic detection tool model are acquired.

[0250] In step S140, based on the intermediate result information and the final label information, a second agent is used to perform feature fusion and intention reasoning to generate an intention inference result of an electromagnetic signal and an evidence chain supporting the intention inference result.

[0251] In step S150, feedback information for the intention inference result is received, the process of generating the task planning document is reflected through the first agent based on the intention inference result and / or the feedback information, and internal parameters of the first agent are updated based on a reflection result.

[0252] Figure 1The method can be applied to the processor 410 or implemented by the processor 410. The processor can be an integrated circuit chip having a signal processing capability. In the implementation process, each step of the above method can be completed by integrated logic circuits or instructions in the form of software in the hardware of the processor. The processor described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. Each method, step and logic block diagram disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as a hardware code processor for execution, or a combination of hardware and software modules in the code processor for execution. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register or other mature storage medium in the art. The storage medium is located in the memory, and the processor reads the information in the memory and combines the hardware to complete the steps of the above method.

[0253] Of course, in addition to the software implementation, the electronic device of the present application does not exclude other implementation manners, such as logic devices or a combination of software and hardware, etc. That is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0254] The related part of the multi-agent collaborative device for electromagnetic spectrum monitoring and analysis provided in Embodiment 2 of the present application can refer to the detailed description of the corresponding part of the multi-agent collaborative method for electromagnetic spectrum monitoring and analysis provided in Embodiment 1 of the present application, which will not be repeated here.

[0255] It is to be noted that, in the present text, the terms such as first and second, and the like are used merely to differentiate one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between such entities or operations. Moreover, the terms "comprising", "containing", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements in the list. Without more limitations, an element defined by an expression "comprising a..." does not exclude the existence of additional identical elements in the process, method, article, or apparatus including the element. In addition, the above-described technical solutions provided by the embodiments of the present application have not been described in detail, which are consistent with the implementation principles of the corresponding technical solutions in the prior art, so as not to be too verbose.

[0256] The above describes the specific embodiments of the present application in conjunction with the accompanying drawings, but is not a limitation on the scope of protection of the present application. For those skilled in the art, other different forms of modifications or variations can be made on the basis of the above description. Here, it is not necessary and impossible to exhaust all the embodiments. Various modifications or variations made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the scope of protection of the present application.

Claims

1. A multi-agent cooperative method for electromagnetic spectrum monitoring and analysis, characterized in that, Includes the following steps: Receive task requirements for electromagnetic spectrum monitoring and analysis, and obtain interface information for electromagnetic detection tool models; The task requirements are analyzed by a pre-defined first intelligent agent to generate a preliminary task intent; Based on the initial task intent and the interface information, a task planning document containing multiple sub-tasks is generated; The task scheduling module of the second intelligent agent calls the electromagnetic detection tool model to execute the sub-tasks of the task plan and obtains the intermediate result information and final label information generated by the electromagnetic detection tool model. The intermediate result information and the final label information are input into a preset second intelligent agent; The second intelligent agent performs feature fusion and intent reasoning on the input intermediate result information and final label information to generate the intent inference result of the electromagnetic signal and the evidence chain supporting the intent inference result. Specifically: The intermediate result information and the final label information are fused into a feature vector; Based on a predefined domain knowledge rule base, corresponding weight coefficients are assigned to different feature categories in the feature vector to generate a weighted comprehensive feature vector; The weighted feature vector is matched and reasoned with a pre-set scene intent knowledge base to generate intent inference results for the behavior of electromagnetic signal emitting sources. Based on the weighted integrated feature vector and the matching reasoning process, an evidence chain supporting the intent inference result is generated. The evidence chain includes at least the feature source, the reasoning logic path, and the contribution of each piece of evidence to the final conclusion.

2. The method according to claim 1, characterized in that, The task requirements are analyzed by a pre-defined first intelligent agent to generate a preliminary task intent; and a task plan containing multiple sub-tasks is generated based on the preliminary task intent and the interface information. Specifically: The task requirements are semantically understood to generate a preliminary task intent; the preliminary task intent includes the task output target and execution steps. Based on the initial task intent and in conjunction with the functions described in the interface information, task decomposition and tool matching are performed. The matching results are iteratively optimized to generate a structured task plan, which defines the task type, input data source, electromagnetic detection tool model to be called, execution priority, and dependencies between subtasks for each subtask.

3. The method according to claim 1, characterized in that, The task scheduling module of the second intelligent agent calls the electromagnetic detection tool model to execute the sub-tasks of the task plan, and obtains the intermediate result information and final label information generated by the electromagnetic detection tool model, specifically: According to the dynamic task scheduling sequence, the normalized electromagnetic data is bound and invoked with the electromagnetic detection tool model required by the sub-task; the electromagnetic detection tool model includes an automatic modulation classification model, a protocol identification model, a tracking and positioning model, a noise filtering model, and a signal enhancement model; Monitor the execution status of each electromagnetic detection tool model and collect its output intermediate result information and final label information; wherein, the intermediate result information includes text descriptions or intermediate features in the reasoning process of the electromagnetic detection tool model, and the final label information includes the classification, identification or location conclusions drawn by the electromagnetic detection tool model after reasoning on the input data.

4. The method according to claim 1, characterized in that, The method further includes: Receive feedback information regarding the intent inference result; Based on the intent inference results and / or the feedback information, the first intelligent agent reflects on the process of generating the task plan. Based on the reflection results, the internal parameters of the first agent are updated.

5. The method according to claim 4, characterized in that, The process of generating the task plan is reflected upon by the first intelligent agent, specifically as follows: Based on the intent inference results, determine the contribution of each final tag information to the intent inference results; Evidence was selected based on the stated contribution level; Analyze the causal relationship between the evidence and the inferred intent, and adjust the task planning logic based on the analysis results.

6. The method according to claim 5, characterized in that, Based on the intent inference result, the contribution of each final tag information to the intent inference result is determined, specifically as follows: In the computational graph of the second agent performing intent reasoning, the intent reasoning result is used as the starting point to backpropagate and calculate the gradient of its output layer with respect to the input layer. Extract the gradient value on the input node corresponding to the final label information, and use the absolute value of the gradient value as a quantitative indicator of the contribution of the corresponding final label information to the intent inference result; Based on the quantitative indicators, all final label information is sorted, and a preset number of evidence with the highest contribution is selected.

7. The method according to any one of claims 1 to 6, characterized in that, The method for obtaining the first intelligent agent is as follows: obtain the first basic large language model; use the first training dataset to perform supervised fine-tuning on the first basic large language model to obtain the first intelligent agent; wherein, the first training dataset contains multiple task requirement texts in the field of electromagnetic monitoring, and a structured task planning document corresponding to each task requirement text. The method for obtaining the second intelligent agent is as follows: obtaining a second basic large language model; using a second training dataset to perform supervised fine-tuning on the second basic large language model to obtain the second intelligent agent; wherein, the second training dataset contains multiple electromagnetic signal feature data, as well as the intent inference conclusion and evidence chain description corresponding to each feature data.

8. The method according to claim 7, characterized in that, The method further includes: achieving collaboration between the first and second intelligent agents through the evolution of the planning strategy of the first intelligent agent, the evolution of the intent knowledge of the second intelligent agent, and the evolution of the toolset of the task scheduling module of the second intelligent agent, specifically: The first intelligent agent updates its parameters based on the reflection results, optimizing its understanding and planning capabilities for electromagnetic monitoring tasks; During the reasoning process, the second intelligent agent incrementally updates the scene intent knowledge base it invokes based on the reflected electromagnetic signal intent pattern. The task scheduling module of the second intelligent agent can dynamically integrate the updated electromagnetic detection tool model according to the updated task plan.

9. A multi-agent collaborative device for electromagnetic spectrum monitoring and analysis, comprising at least one processor and a memory, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the at least one processor, it implements a multi-agent cooperative method for electromagnetic spectrum monitoring and analysis as described in any one of claims 1 to 8.

Images