Intelligent communication distribution system based on big data

Abstract

The present application relates to the field of digital processing of archives, and aims to solve the problems of low recognition accuracy of multi-source heterogeneous archives and non-compliance of generated electronic data, and provides an intelligent system. The present application contains modules such as archive knowledge graph construction, improved semantic enhancement recognition, and improved knowledge guided generation. Through preprocessing of multi-source heterogeneous original archives, relying on the archive knowledge graph containing entity, relationship and rule layer, using improved algorithm to accurately recognize archive entity and associated information, and then generating electronic data that conforms to archive specifications and has semantic association, and after compliance verification, it is stored in the archive. The present application improves the accuracy and compliance of electronic data processing of archives, adapts to multi-modal archive scenarios, and helps the intelligent management of digital archives.

Description

Technical Field

[0001] This invention relates to the fields of intelligent communication technology and big data processing, specifically to an intelligent communication distribution system based on big data. Background Technology

[0002] With the rapid development of 5G and advanced sensor technologies, drones have been widely used in power grid inspection, becoming an important tool for data collection and monitoring in smart grids. In drone-assisted power grid inspection scenarios, the system needs to transmit two types of heterogeneous data: one is short packet data of 20-50 bytes generated by transmission tower sensors and meters (used for periodic monitoring of equipment status), which has extremely high requirements for transmission latency and reliability; the other is long packet data of images and videos acquired by drones equipped with high-definition cameras (used for equipment appearance inspection), which has a larger requirement for transmission bandwidth.

[0003] Currently, the communication allocation of UAV-assisted power line inspection systems faces specific technical problems. Specifically, existing communication allocation systems do not fully consider the transmission differences between heterogeneous data packets of varying lengths. A uniform resource allocation strategy either prioritizes the low latency requirements of short packets, leading to congestion and bandwidth waste in long packet data transmission, or focuses on bandwidth supply for long packets, resulting in the loss of information in short packets. The increased complexity of the data allocation makes it impossible to meet the dual requirements of power line inspection for data timeliness and resource utilization. At the same time, existing allocation algorithms are mostly based on fixed models or traditional reinforcement learning frameworks, which are difficult to adapt to the dynamic changes in the channel caused by the high mobility of UAVs (such as Ricean channel gain fluctuations). They also have problems such as slow convergence speed and insufficient global optimization ability, which further exacerbates the irrationality of resource allocation.

[0004] Furthermore, in existing technologies, centralized communication allocation methods require frequent status interactions between UAVs and ground control centers, generating significant signaling overhead and consuming valuable spectrum resources. At the same time, the decision-making delay is high, failing to meet the microsecond-level response requirements for emergency fault data in power line inspections. On the other hand, purely edge allocation methods lack global resource scheduling capabilities, easily leading to local resource overload and low overall utilization.

[0005] Currently, there is no effective solution to the specific communication allocation problems encountered in the aforementioned drone-assisted power grid inspection scenario. Existing communication allocation systems cannot simultaneously address the needs of heterogeneous data transmission, adaptability to dynamic channel changes, and resource utilization efficiency, severely impacting the efficiency of power grid inspections and the security of power grid monitoring. Therefore, there is an urgent need for a big data-based intelligent communication allocation system to solve the core technical pain points in this specific scenario. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent communication distribution system based on big data to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent communication allocation system based on big data, comprising a data acquisition module, a big data preprocessing module, a heterogeneous data priority dynamic allocation module, a channel adaptive resource scheduling module, a communication execution module, and a data feedback module; the data acquisition module acquires multi-source raw data during the UAV inspection process, the multi-source raw data including long and short packet heterogeneous data and channel status data, wherein the short packet data includes sensor status data and fault alarm data, the long packet data includes inspection images and video data, and the channel status data includes channel gain, noise power, and interference intensity; the big data preprocessing module performs denoising, normalization, and feature extraction processing on the acquired multi-source raw data set to obtain standardized data. Heterogeneous data and channel state characteristic data are used in the transmission process. The heterogeneous data priority dynamic allocation module, based on preprocessed standardized heterogeneous data, calculates the dynamic priority of long and short packet data using an improved algorithm to determine the data transmission order. The channel adaptive resource scheduling module allocates communication bandwidth and power resources according to data priority and channel state characteristic data using an improved algorithm. The communication execution module completes the transmission of heterogeneous data using non-orthogonal multiple access (NOMA) technology according to the priority order and resource allocation results. The data feedback module collects data during transmission in real time (including transmission delay, bit error rate, AoI value, and resource utilization), and feeds it back to the heterogeneous data priority dynamic allocation module and the channel adaptive resource scheduling module to achieve dynamic adjustment and optimization. The heterogeneous data priority dynamic allocation module includes a data feature extraction unit, an urgency assessment unit, a channel adaptability calculation unit, an improved priority calculation unit, and a priority sorting unit. The data feature extraction unit extracts key features of the preprocessed heterogeneous data (fault level and data generation time for short packets, resolution and data volume for long packets). The urgency assessment unit evaluates the urgency of standardized heterogeneous data. The channel adaptability calculation unit calculates the adaptability of different types of standardized heterogeneous data to the current channel state. The improved priority calculation unit calculates the dynamic priority of various types of standardized heterogeneous data using an improved algorithm. The priority sorting unit determines the standardized heterogeneous data transmission order based on the priority calculation results. The channel adaptive resource scheduling module includes a channel state prediction unit, a resource demand analysis unit, an improved resource scheduling algorithm unit, a resource allocation execution unit, and a scheduling optimization unit. The channel state prediction unit predicts the channel state in the near future based on big data. The resource demand analysis unit analyzes the resource demands of various standardized heterogeneous data based on data priority sequences and data characteristics. The improved resource scheduling algorithm unit calculates bandwidth and power allocation schemes through improved algorithms. The resource allocation execution unit executes the resource allocation schemes and coordinates with the communication execution module to complete data transmission. The scheduling optimization unit optimizes resource allocation parameters based on feedback data.

[0008] Preferably, the specific implementation logic of the data acquisition module is as follows: Step A1, Deployment and Initialization of Data Acquisition Module: Integrate and deploy the various components of the data acquisition module on the UAV, link it with the UAV flight control system and edge computing unit, and complete the initial configuration; Step A2, Determining the Data Collection Scope and Parameters: Based on the requirements of the UAV power line inspection task, clarify the data collection scope and core parameters to ensure that the collected data meets the system requirements. Heterogeneous data acquisition scope: Short packet data focuses on sensor status data and fault alarm data (such as line short circuit, equipment overheating and other fault signals) of key parts of the transmission tower (conductors, insulators, tower foundations), with the data length strictly controlled between 20-50 bytes (consistent with the characteristics of short packet data); Long packet data focuses on image and video data of the appearance of transmission lines and transmission towers. The camera resolution is adjusted according to the key inspection areas (such as suspected fault areas), prioritizing the acquisition of 1080P high-definition images, and using 720P resolution for non-key areas to balance data quality and acquisition efficiency; Channel data acquisition range: Synchronously acquire the channel gain of the communication link between the UAV and the ground control center. Noise power Interference intensity It covers the entire flight trajectory of UAV inspection, ensuring the continuity of channel state data and providing complete data support for subsequent channel state prediction and channel adaptability calculation; Acquisition triggering mechanism: A combination of periodic acquisition and event-triggered acquisition is adopted. Short packet data and channel data are acquired periodically at a preset frequency. In addition to periodic acquisition, when the sensor detects a fault signal (event trigger), the acquisition frequency of the camera is automatically increased to 5 frames / second to focus on capturing images of the fault area and ensure the timeliness of fault-related long packet data. Step A3: Synchronous Acquisition of Multi-Source Raw Data: The UAV flies along the preset inspection route, and all acquisition components work together to achieve synchronous acquisition of multi-source raw data, avoiding data timing deviations. Short packet data acquisition: Sensors capture the status of transmission tower equipment in real time. When a data change is detected (such as a sudden temperature rise or fault signal triggering), the data is immediately acquired and packaged into a short packet, with the data generation time marked. (In accordance with the time parameter requirements of the urgency assessment unit), temporarily stored in the drone's local cache; Long-term data acquisition: High-definition cameras simultaneously capture images and videos of the inspection area, and automatically label the resolution of each captured image / video frame. Data volume (Consistent with the parameter requirements for channel adaptability calculation and resource requirement analysis), encapsulated as long packet data and temporarily stored in local cache; Channel data acquisition: The channel state monitoring module acquires the channel parameters of the communication link in real time. Each time data is acquired, the acquisition time is marked and correlated with the heterogeneous data acquired at the same time to ensure the correspondence between data and channel state in subsequent calculations and avoid the accuracy of channel adaptability calculation and resource scheduling due to timing deviation. Step A4: Data Preprocessing and Preliminary Verification: After temporarily storing the collected multi-source raw data in the UAV's local cache, perform simple preliminary preprocessing and verification to reduce invalid data from being transmitted to subsequent modules: Preliminary denoising: Apply simple mean filtering to the short packet data collected by the sensors to remove obvious outliers (such as data exceeding the reasonable range due to sensor malfunction); perform smoothing processing on the channel data to filter out abnormal fluctuations caused by instantaneous channel changes; Data verification: Verify the short packet data length (ensuring it is 20-50 bytes), the resolution and data volume labeling integrity of the long packet data, and the rationality of the channel data parameters (such as channel gain). Remove invalid data (such as blank images and channel data with missing parameters); Data encapsulation: Encapsulate the verified short packet data, long packet data, and channel data in a unified format, and label the collection time and UAV location information (to assist in subsequent channel status correction) to form a standardized multi-source raw data set; Step A5: Real-time data transmission to the big data preprocessing module: The data acquisition module transmits the encapsulated, standardized multi-source raw data set to the big data preprocessing module in real time via the internal link of the UAV edge computing unit. During the transmission, a simple error control mechanism (such as checksum) is used to ensure that the data transmission is lossless and error-free. At the same time, a data backup interface is reserved. When the transmission link is temporarily interrupted, the acquired data is temporarily stored in local storage and retransmitted after the link is restored, ensuring the continuity of data and providing complete and reliable raw data input for the subsequent noise reduction, normalization, and feature extraction work of the big data preprocessing module.

[0009] Preferably, the deployment of the data acquisition module in step A1 specifically includes the following: Sensor deployment: Temperature sensors, humidity sensors, and fault detection sensors (used to capture fault signals from power transmission tower equipment) are installed on the UAV fuselage and inspection mounting points. The sensor sampling frequency is set to 10Hz, and the sampling accuracy matches the power inspection standards (e.g., temperature error ≤ ±0.5℃) to ensure the real-time acquisition of short packet data; High-definition camera deployment: Equipped with a high-definition adjustable camera, supporting switching between 1080P, 720P, and 480P resolutions. The lens angle and focal length are dynamically adjusted according to the inspection distance, and the acquisition frequency is [not specified]. The frame rate is set to 1 frame / second (image data) and 25 frames / second (video data) to ensure the clarity and integrity of long packet data; Channel status monitoring module deployment: an integrated channel monitoring unit is used to capture the channel gain, noise power, and interference intensity of the Ricean channel in real time. The acquisition frequency is matched with the channel status change rate and set to 5Hz to adapt to the channel dynamic fluctuations caused by the high mobility of the UAV; Initialization calibration: after starting the UAV, all acquisition components are calibrated to ensure that the sensor zero drift error is within the preset range, the camera is accurately focused, and the channel monitoring module and the ground communication link are normal, so as to avoid abnormal acquisition data.

[0010] Preferably, the specific implementation steps of the heterogeneous data priority dynamic allocation module are as follows: Step B1, Data Feature Input: The big data preprocessing module will input the standardized heterogeneous data, i.e., the short packet data set. Long package dataset Channel gain of channel state characteristic data Noise power The data feature extraction unit is input into the heterogeneous data priority dynamic allocation module. Step B2, Data Feature Extraction: The data feature extraction unit extracts key features from the short packet data: fault level. (Value range 1-5, 5 being the highest fault level), data generation time Key features to extract from long packet data: data volume Resolution (Value range: 1080P, 720P, 480P, corresponding to quantization values ​​of 3, 2, and 1); simultaneously extract the channel gain from the channel state features. This is used for subsequent adaptation calculations; Step B3, Data Urgency Assessment: The urgency assessment unit adopts a tiered assessment method to calculate the urgency of short packet data and long packet data separately. For short packet data Its urgency The formula for calculation is: ,in The current system time. The time decay coefficient (valued at 0.01-0.05, dynamically adjusted according to the inspection scenario) reflects that the higher the fault level and the more recent the generation time of short packet data, the higher the urgency, which meets the requirement of prioritizing the transmission of urgent fault data in power inspection. For long packet data Its urgency The formula for calculation is: ,in To support the maximum long packet data volume, this formula reflects that the higher the resolution and the smaller the data volume of the long packet data, the higher the urgency, balancing the transmission efficiency and quality requirements of long packet data; Step B4, Channel Adaptability Calculation: The channel adaptability calculation unit calculates the adaptability between the standardized heterogeneous data and the current channel state. A higher adaptability indicates higher reliability and lower resource consumption for data transmission under the current channel. For short packet data, the focus is on low-latency adaptability. The formula for calculation is: ,in The channel gain of the current channel. The data length for short packets is fixed at 20-50 bytes. The signal-to-interference-plus-noise ratio (SINR) of the channel is used. The higher the SINR and the shorter the data length, the better the adaptability, which meets the requirements of low-latency transmission of short packet data. For long packet data, the focus is on bandwidth adaptation and adaptability. The formula for calculation is: ,in The higher the signal-to-interference-plus-noise ratio, the more abundant the available bandwidth, and the smaller the data volume, the higher the adaptability, meeting the requirements for efficient transmission of long packet data. Step B5, Dynamic Priority Calculation: The improved priority calculation unit adopts an improved AoI-weighted collaborative algorithm, which integrates urgency, channel adaptability, and information age. Calculate the dynamic priority of standardized heterogeneous data. (Short package) (Long bag), the formula is: ; ,in , , Here are the weighting coefficients for short packet data, and , Prioritize matters of urgency; , , The weighting coefficients for long-package data. , Prioritize channel adaptability, among which The information about age in short packet data. The information age for long-package data, This reflects that the higher the freshness of the data, the higher its priority; Step B6, Priority Sorting: The priority sorting unit sorts the priorities of all short and long data packets uniformly, generating a data transmission sequence according to the order of priority from high to low. The sequence is then sent to the channel adaptive resource scheduling module as the core basis for resource allocation. Step B7, Dynamic Adjustment: The data feedback module will adjust the data during the transmission process. Values ​​and transmission latency are fed back to the heterogeneous data priority dynamic allocation module, which adjusts the weighting coefficients in real time. , , and , , When short packet data When the value exceeds the threshold (e.g., 500ms), increase. The value of is increased when the long packet data transmission delay exceeds the threshold (e.g., 2 seconds). The value of is determined to ensure the dynamic adaptability of priority allocation.

[0011] Preferably, the channel adaptive resource scheduling module is implemented using the following steps: Step C1, Input Data Reception: Receive the data transmission sequence sent by the heterogeneous data priority dynamic allocation module. And standardized channel state data sent by the big data preprocessing module, including historical channel gain sequences. Noise power Interference intensity ; Step C2, Channel State Prediction: The channel state prediction unit uses an LSTM neural network, based on historical channel gain sequences. Predicting the future Each time slot ( Channel gain (100ms per time slot) The prediction formula is: ,in These are the parameters of the LSTM neural network (obtained through training with historical channel data). At the same time, combined with the flight speed of the drone The channel gain prediction value is corrected using the following formula: ,in, The time interval for each time slot (100ms). The distance between the UAV and the ground control center is used as the reference point. This correction step addresses the problem of excessive channel prediction errors caused by the high maneuverability of the UAV. Step C3, Resource Requirements Analysis: The resource requirements analysis unit analyzes the data transmission sequence... Analyze the characteristics of various types of data to understand their bandwidth and power requirements: For short packet data Calculate the minimum bandwidth requirement based on the formula for the maximum coding rate of short packet communication. : ,in Maximum allowable latency for short data packets; minimum power requirement The formula for calculation is: For long packet data Calculate the bandwidth requirements based on the data volume and transmission time requirements. With power demand : ; ,in Transmission time for long packet data (dynamically allocated based on priority; higher priority means shorter transmission time). Step C4, Improved resource scheduling algorithm calculation: The improved resource scheduling algorithm unit adopts an improved dual-time-scale resource scheduling algorithm (improved DDPG-Dueling DQN cooperative algorithm), which divides resource allocation into coarse-grained bandwidth allocation with a large time scale (1s) and fine-grained power allocation with a small time scale (100ms), solving the problems of slow convergence speed and excessively large state space of traditional algorithms; Step C5, Resource Allocation Execution: The resource allocation execution unit calculates the bandwidth allocation scheme based on the improved resource scheduling algorithm. With power distribution scheme The data is sent to the communication execution module, which, in conjunction with NOMA technology, enables the parallel transmission of long and short packet data (short packet data and long packet data are transmitted on the same time and frequency resources, and are distinguished by power domain multiplexing). The ground control center uses serial interference cancellation (SIC) technology to decode the data, first decoding the short packet data with higher priority, and then decoding the long packet data. Step C6, Scheduling Optimization: The data feedback module feeds back the resource utilization, bit error rate, and transmission delay during transmission to the scheduling optimization unit, which then adjusts the algorithm parameters (such as weighting coefficients) in real time. , , Penalty coefficient When resource utilization is below a threshold (e.g., 70%), adjust the bandwidth allocation ratio to increase bandwidth supply for high-priority data; when the bit error rate is above a certain threshold... At the same time, increase the transmission power of the corresponding data to ensure the optimization and stability of resource allocation.

[0012] Preferably, the specific implementation steps of the improved resource scheduling algorithm in step C4 are as follows: Step C4.1, Large-scale coarse-grained bandwidth allocation (based on the improved DDPG algorithm): Based on the total bandwidth... As constraints, to maximize system resource utilization and minimize average resource utilization. To achieve this, bandwidth is allocated to different types of data, including short packet data. Total bandwidth of long packet data ,satisfy The policy network update formula for the improved DDPG algorithm is: ,in For policy network parameters; The gradient of the objective function of the policy network; The input state of the policy network is specifically a standardized set of channel state characteristics and data characteristics, including historical channel gain, noise power, current available bandwidth, total amount of long and short packets and priority distribution, etc., which is the core basis for output bandwidth allocation actions. for Value functions; Let this be the policy function, representing the state at input. Below, output the optimal bandwidth allocation action on a large time scale. ; For experience replay pool; Subsequently Introducing channel prediction factor into the value function Improved channel adaptability of bandwidth allocation, corrected The value function is: ,in Resource utilization rate The average age of the system's information. , , These are the weighting coefficients; Step C4.2, Fine-grained power allocation on a small time scale (based on an improved Dueling DQN algorithm): Building upon the bandwidth allocation on a large time scale, [the following steps are performed]. , Each data point is allocated its own transmission power, ensuring that the minimum bandwidth and power requirements of each data point are met. An improved Dueling DQN algorithm is then used to... Value decoupling into state value function With advantage function The formula is: ; The improved Dueling DQN algorithm introduces a power constraint factor. (System maximum transmission power) and bit error rate constraints By adding a bit error rate penalty term to the dominance function to avoid transmission failures caused by excessively low power allocation, the modified dominance function is as follows: ,in The reward value is calculated based on resource utilization and transmission latency. The penalty coefficient is... This represents the actual bit error rate.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention solves the specific communication allocation problem in the scenario of UAV-assisted power inspection, namely the problems of heterogeneous data transmission conflict of long and short packets, poor adaptability to dynamic channel changes, low resource utilization, and high signaling overhead. Through the collaborative work of two creative modules, the present invention achieves accurate allocation of heterogeneous data and efficient utilization of resources, ensuring the timeliness and reliability of power inspection data.

[0014] The heterogeneous data priority dynamic allocation module of this invention adopts an improved AoI-weighted collaborative algorithm, breaking through the traditional fixed priority mode. It comprehensively considers urgency, channel adaptability, and information age to dynamically adjust data priority, enabling short packet data to achieve higher priority. The reduced value lowers the transmission latency of long packet data, thus avoiding conflicts in heterogeneous data transmission.

[0015] The channel adaptive resource scheduling module of this invention adopts an improved dual-time-scale resource scheduling algorithm, which combines LSTM channel prediction and UAV flight state correction to improve channel adaptability, increase resource utilization, shorten algorithm convergence speed, and reduce signaling overhead caused by centralized scheduling, thus meeting the needs of high-mobility UAV scenarios. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the system structure of the present invention; Figure 2 This is a schematic diagram of the data acquisition module workflow of the present invention; Figure 3 This is a schematic diagram of the workflow of the heterogeneous data priority dynamic allocation module of the present invention; Figure 4 This is a schematic diagram of the workflow of the channel adaptive resource scheduling module of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1-4 This invention provides a technical solution: an intelligent communication allocation system based on big data, comprising a data acquisition module, a big data preprocessing module, a heterogeneous data priority dynamic allocation module, a channel adaptive resource scheduling module, a communication execution module, and a data feedback module; the data acquisition module acquires multi-source raw data during UAV inspection, including heterogeneous data of long and short packets and channel status data, wherein short packet data includes sensor status data and fault alarm data, long packet data includes inspection images and video data, and channel status data includes channel gain, noise power, and interference intensity; the big data preprocessing module processes the acquired multi-source raw dataset... The data undergoes denoising, normalization, and feature extraction to obtain standardized heterogeneous data and channel state feature data. A dynamic priority allocation module for heterogeneous data, based on the preprocessed standardized heterogeneous data, calculates the dynamic priority of long and short packet data using an improved algorithm to determine the data transmission order. A channel adaptive resource scheduling module allocates communication bandwidth and power resources according to data priority and channel state feature data using an improved algorithm. The communication execution module, following the priority order and resource allocation results, uses non-orthogonal multiple access (NOMA) technology to complete the transmission of heterogeneous data. A data feedback module collects data in real time during the transmission process (including transmission delay, bit error rate, etc.). The data (value, resource utilization rate) is fed back to the heterogeneous data priority dynamic allocation module and the channel adaptive resource scheduling module to achieve dynamic adjustment and optimization.

[0019] The implementation logic of each module is explained below: The specific implementation logic of the data acquisition module is as follows: Step A1, Data Acquisition Module Deployment and Initialization: Integrate and deploy all components of the data acquisition module on the UAV, linking it with the UAV flight control system and edge computing unit to complete the initial configuration; Step A1, the deployment of the acquisition module specifically includes the following: Sensor Deployment: Install temperature sensors, humidity sensors, and fault detection sensors (for capturing fault signals of transmission tower equipment) on the UAV fuselage and inspection mounting points. The sensor sampling frequency is set to 10Hz, and the sampling accuracy matches the power inspection standard (e.g., temperature error ≤ ±0.5℃) to ensure the real-time acquisition of short packet data; High-Definition Camera Deployment: Equip with a high-definition adjustable camera, supporting 1080P, 720P, and 480P resolutions. The system offers three resolution options (0P, 0P, 0P) with dynamic adjustment of lens angle and focal length based on inspection distance. The acquisition frequency is set to 1 frame / second (image data) and 25 frames / second (video data) to ensure the clarity and integrity of long packet data. The channel status monitoring module integrates a channel monitoring unit to capture the channel gain, noise power, and interference intensity of the Ricean channel in real time. The acquisition frequency is matched to the channel status change rate and set to 5Hz to adapt to the dynamic channel fluctuations caused by the high mobility of the drone. Initialization calibration involves calibrating all acquisition components after drone startup to ensure that sensor zero-drift error is within the preset range, camera focus is accurate, and the channel monitoring module and ground communication link are normal, preventing abnormal data acquisition. Step A2, Determining the Data Collection Scope and Parameters: Based on the requirements of the UAV power line inspection task, clarify the data collection scope and core parameters to ensure that the collected data meets the system requirements. Heterogeneous data acquisition scope: Short packet data focuses on sensor status data and fault alarm data (such as line short circuit, equipment overheating and other fault signals) of key parts of the transmission tower (conductors, insulators, tower foundations), with the data length strictly controlled between 20-50 bytes (consistent with the characteristics of short packet data); Long packet data focuses on image and video data of the appearance of transmission lines and transmission towers. The camera resolution is adjusted according to the key inspection areas (such as suspected fault areas), prioritizing the acquisition of 1080P high-definition images, and using 720P resolution for non-key areas to balance data quality and acquisition efficiency; Channel data acquisition range: Synchronously acquire the channel gain of the communication link between the UAV and the ground control center. Noise power Interference intensity It covers the entire flight trajectory of UAV inspection, ensuring the continuity of channel state data and providing complete data support for subsequent channel state prediction and channel adaptability calculation; Acquisition triggering mechanism: A combination of periodic acquisition and event-triggered acquisition is adopted. Short packet data and channel data are acquired periodically at a preset frequency. In addition to periodic acquisition, when the sensor detects a fault signal (event trigger), the acquisition frequency of the camera is automatically increased to 5 frames / second to focus on capturing images of the fault area and ensure the timeliness of fault-related long packet data. Step A3: Synchronous Acquisition of Multi-Source Raw Data: The UAV flies along the preset inspection route, and all acquisition components work together to achieve synchronous acquisition of multi-source raw data, avoiding data timing deviations. Short packet data acquisition: Sensors capture the status of transmission tower equipment in real time. When a data change is detected (such as a sudden temperature rise or fault signal triggering), the data is immediately acquired and packaged into a short packet, with the data generation time marked. (In accordance with the time parameter requirements of the urgency assessment unit), temporarily stored in the drone's local cache; Long-term data acquisition: High-definition cameras simultaneously capture images and videos of the inspection area, and automatically label the resolution of each captured image / video frame. Data volume (Consistent with the parameter requirements for channel adaptability calculation and resource requirement analysis), encapsulated as long packet data and temporarily stored in local cache; Channel data acquisition: The channel state monitoring module acquires the channel parameters of the communication link in real time. Each time data is acquired, the acquisition time is marked and correlated with the heterogeneous data acquired at the same time to ensure the correspondence between data and channel state in subsequent calculations and avoid the accuracy of channel adaptability calculation and resource scheduling due to timing deviation. Step A4: Data Preprocessing and Preliminary Verification: After temporarily storing the collected multi-source raw data in the UAV's local cache, perform simple preliminary preprocessing and verification to reduce invalid data from being transmitted to subsequent modules: Preliminary denoising: Apply simple mean filtering to the short packet data collected by the sensors to remove obvious outliers (such as data exceeding the reasonable range due to sensor malfunction); perform smoothing processing on the channel data to filter out abnormal fluctuations caused by instantaneous channel changes; Data verification: Verify the short packet data length (ensuring it is 20-50 bytes), the resolution and data volume labeling integrity of the long packet data, and the rationality of the channel data parameters (such as channel gain). Remove invalid data (such as blank images and channel data with missing parameters); Data encapsulation: Encapsulate the verified short packet data, long packet data, and channel data in a unified format, and label the collection time and UAV location information (to assist in subsequent channel status correction) to form a standardized multi-source raw data set; Step A5: Real-time data transmission to the big data preprocessing module: The data acquisition module transmits the encapsulated, standardized multi-source raw data set to the big data preprocessing module in real time via the internal link of the UAV edge computing unit. During the transmission, a simple error control mechanism (such as checksum) is used to ensure that the data transmission is lossless and error-free. At the same time, a data backup interface is reserved. When the transmission link is temporarily interrupted, the acquired data is temporarily stored in local storage and retransmitted after the link is restored, ensuring the continuity of data and providing complete and reliable raw data input for the subsequent noise reduction, normalization, and feature extraction work of the big data preprocessing module.

[0020] The heterogeneous data priority dynamic allocation module includes a data feature extraction unit, an urgency assessment unit, a channel adaptability calculation unit, an improved priority calculation unit, and a priority sorting unit. The data feature extraction unit extracts key features of the preprocessed heterogeneous data (fault level and data generation time for short packets, resolution and data volume for long packets). The urgency assessment unit evaluates the urgency of standardized heterogeneous data. The channel adaptability calculation unit calculates the adaptability of different types of standardized heterogeneous data to the current channel state. The improved priority calculation unit calculates the dynamic priority of various types of standardized heterogeneous data using an improved algorithm. The priority sorting unit determines the standardized heterogeneous data transmission order based on the priority calculation results. The specific implementation steps of the heterogeneous data priority dynamic allocation module are as follows: Step B1, Data Feature Input: The big data preprocessing module will input the standardized heterogeneous data, i.e., the short packet data set. Long package dataset Channel gain of channel state characteristic data Noise power The data feature extraction unit is input into the heterogeneous data priority dynamic allocation module; whereby... Represents the total number of short packet data; each All data include key features such as fault level and data generation time (extracted by the big data preprocessing module), corresponding to sensor monitoring data or fault alarm data of a single power transmission tower device in drone inspection, with a fixed data length of 20-50 bytes. Represents the total number of long packet data; each All contain key features such as resolution and data volume (extracted by the big data preprocessing module), corresponding to a single frame image or a video data segment in UAV inspection, with resolutions supporting 1080P, 720P, and 480P (in line with the feature definition of long packet data in the invention). Step B2, Data Feature Extraction: The data feature extraction unit extracts key features from the short packet data: fault level. (Value range 1-5, 5 being the highest fault level), data generation time Key features to extract from long packet data: data volume Resolution (Value range: 1080P, 720P, 480P, corresponding to quantization values ​​of 3, 2, and 1); simultaneously extract the channel gain from the channel state features. This is used for subsequent adaptation calculations; Step B3, Data Urgency Assessment: The urgency assessment unit adopts a tiered assessment method to calculate the urgency of short packet data and long packet data separately. For short packet data Its urgency The formula for calculation is: ,in The current system time. The time decay coefficient (valued at 0.01-0.05, dynamically adjusted according to the inspection scenario) reflects that the higher the fault level and the more recent the generation time of short packet data, the higher the urgency, which meets the requirement of prioritizing the transmission of urgent fault data in power inspection. For long packet data Its urgency The formula for calculation is: ,in To support the maximum long packet data volume, this formula reflects that the higher the resolution and the smaller the data volume of the long packet data, the higher the urgency, balancing the transmission efficiency and quality requirements of long packet data; Step B4, Channel Adaptability Calculation: The channel adaptability calculation unit calculates the adaptability between the standardized heterogeneous data and the current channel state. A higher adaptability indicates higher reliability and lower resource consumption for data transmission under the current channel. For short packet data, the focus is on low-latency adaptability. The formula for calculation is: ,in The channel gain of the current channel. The data length for short packets is fixed at 20-50 bytes. The signal-to-interference-plus-noise ratio (SINR) of the channel is used. The higher the SINR and the shorter the data length, the better the adaptability, which meets the requirements of low-latency transmission of short packet data. For long packet data, the focus is on bandwidth adaptation and adaptability. The formula for calculation is: ,in The higher the signal-to-interference-plus-noise ratio, the more abundant the available bandwidth, and the smaller the data volume, the higher the adaptability, meeting the requirements for efficient transmission of long packet data. Step B5, Dynamic Priority Calculation: The improved priority calculation unit adopts an improved AoI-weighted collaborative algorithm, which integrates urgency, channel adaptability, and information age. Calculate the dynamic priority of standardized heterogeneous data. (Short package) (Long bag), the formula is: ; ,in , , Here are the weighting coefficients for short packet data, and , Prioritize matters of urgency; , , The weighting coefficients for long-package data. , Prioritize channel adaptability, among which The information about age in short packet data. The information age for long-package data, This reflects that the higher the freshness of the data, the higher its priority; It should be noted that the aforementioned weighting coefficients are not fixed values, but are obtained through a three-step method of "sample training - constraint calibration - dynamic adaptation" based on the core requirements of the UAV power line inspection scenario and historical data after big data preprocessing. The specific steps are as follows, and the acquisition process is deeply bound to the working logic of the two inventive modules of this invention to ensure that the parameters adapt to the scenario requirements: Step 1: Sample Data Collection and Preprocessing. Based on the standardized data output by the big data preprocessing module, extract historical drone inspection data from the past 3-6 months, including the fault level distribution and urgency statistics of short package data. The system records the resolution distribution, data volume statistics, and channel adaptability of long packet data, as well as feedback data such as transmission delay, resource utilization, and bit error rate for the corresponding time period. It constructs a training sample set for weight coefficients, removes abnormal data (such as invalid data caused by channel mutations or equipment failures), and performs normalization processing to ensure the validity of the sample data.

[0021] Step 2: Initial weight coefficient training. Based on the core requirements of this invention (short packet priority urgency, long packet priority channel adaptability), the training objective is set as follows: the training objective for short packet data is to "minimize the data with high urgency". The training objective for long packet data is to "maximize the transmission efficiency and resource utilization of high-channel-adaptability data"; a gradient descent algorithm is used, with feedback data (transmission delay, transmission latency, and transmission time) from the training sample set as the training data. Value and resource utilization rate) are used as evaluation indicators, and the initial weighting coefficients are set as follows (initial value setting: short package). , , Long bag , , The training is performed iteratively, with the coefficient values ​​adjusted in each iteration, until the evaluation metric reaches a preset threshold (e.g., the data with the highest urgency of short packets). The initial weight coefficients are obtained by taking values ​​≤300ms and ensuring the resource utilization rate of the data with the highest long packet channel adaptability is ≥80%.

[0022] Step 3: Constraint Calibration and Dynamic Adaptation. Based on weighted coefficient constraints (short package) , Long bag , The initial coefficients obtained from training are calibrated to ensure that the coefficients meet the constraints (if the coefficients do not meet the constraints after training, such as short packets). Then adjust and The values ​​are selected such that the sum is 1 and the magnitude relationship is satisfied. After calibration, the coefficients are imported into the heterogeneous data priority dynamic allocation module, combined with the real-time feedback from the data feedback module. The AoI value, transmission latency, and resource utilization data are dynamically adjusted: when the AoI value of short packet data exceeds a threshold (e.g., 500ms), the value is increased. (Information age weighting), reduce (Channel adaptability weight), while maintaining The sum is 1; when the long packet data transmission delay exceeds the threshold (e.g., 2s), the value increases. (Channel Adaptability Weight), Reduce (Urgency weight), while maintaining The sum of all values ​​is 1, ensuring that the weighting coefficients always adapt to the real-time scenario requirements. Step B6, Priority Sorting: The priority sorting unit sorts the priorities of all short and long data packets uniformly, generating a data transmission sequence according to the order of priority from high to low. The sequence is then sent to the channel adaptive resource scheduling module as the core basis for resource allocation. Step B7, Dynamic Adjustment: The data feedback module will adjust the data during the transmission process. Values ​​and transmission latency are fed back to the heterogeneous data priority dynamic allocation module, which adjusts the weighting coefficients in real time. , , and , , When short packet data When the value exceeds the threshold (e.g., 500ms), increase. The value of is increased when the long packet data transmission delay exceeds the threshold (e.g., 2 seconds). The value of ensures the dynamic adaptability of priority allocation. The channel adaptive resource scheduling module includes a channel state prediction unit, a resource demand analysis unit, an improved resource scheduling algorithm unit, a resource allocation execution unit, and a scheduling optimization unit. The channel state prediction unit predicts the channel state in the near future based on big data. The resource demand analysis unit analyzes the resource demands of various standardized heterogeneous data based on data priority sequences and data characteristics. The improved resource scheduling algorithm unit calculates bandwidth and power allocation schemes through improved algorithms. The resource allocation execution unit executes the resource allocation schemes and coordinates with the communication execution module to complete data transmission. The scheduling optimization unit optimizes resource allocation parameters based on feedback data.

[0023] The specific implementation steps of the channel adaptive resource scheduling module are as follows: Step C1, Input Data Reception: Receive the data transmission sequence sent by the heterogeneous data priority dynamic allocation module. And standardized channel state data sent by the big data preprocessing module, including historical channel gain sequences. Noise power Interference intensity ; Step C2, Channel State Prediction: The channel state prediction unit uses an LSTM neural network, based on historical channel gain sequences. Predicting the future Each time slot ( Channel gain (100ms per time slot) The prediction formula is: ,in These are the parameters of the LSTM neural network (obtained through training with historical channel data). At the same time, combined with the flight speed of the drone The channel gain prediction value is corrected using the following formula: ,in, The time interval for each time slot (100ms). The distance between the UAV and the ground control center is used as the reference point. This correction step addresses the problem of excessive channel prediction errors caused by the high maneuverability of the UAV. Step C3, Resource Requirements Analysis: The resource requirements analysis unit analyzes the data transmission sequence... Analyze the characteristics of various types of data to understand their bandwidth and power requirements: For short packet data Calculate the minimum bandwidth requirement based on the formula for the maximum coding rate of short packet communication. : ,in Maximum allowable latency for short data packets; minimum power requirement The formula for calculation is: For long packet data Calculate the bandwidth requirements based on the data volume and transmission time requirements. With power demand : ; ,in Transmission time for long packet data (dynamically allocated based on priority; higher priority means shorter transmission time). Step C4: Improved Resource Scheduling Algorithm Calculation: The improved resource scheduling algorithm unit adopts an improved dual-timescale resource scheduling algorithm (improved DDPG-Dueling DQN cooperative algorithm), which divides resource allocation into coarse-grained bandwidth allocation at a large timescale (1s) and fine-grained power allocation at a small timescale (100ms), solving the problems of slow convergence speed and excessively large state space of traditional algorithms; the specific implementation steps are as follows: Step C4.1, Large-scale coarse-grained bandwidth allocation (based on the improved DDPG algorithm): Based on the total bandwidth... As constraints, to maximize system resource utilization and minimize average resource utilization. To achieve this, bandwidth is allocated to different types of data, including short packet data. Total bandwidth of long packet data ,satisfy The policy network update formula for the improved DDPG algorithm is: ,in For policy network parameters; The gradient of the objective function of the policy network; The input state of the policy network is specifically a standardized set of channel state characteristics and data characteristics, including historical channel gain, noise power, current available bandwidth, total amount of long and short packets and priority distribution, etc., which is the core basis for output bandwidth allocation actions. for Value functions; Let this be the policy function, representing the state at input. Below, output the optimal bandwidth allocation action on a large time scale. ; For experience replay pool; Subsequently Introducing channel prediction factor into the value function Improved channel adaptability of bandwidth allocation, corrected The value function is: ,in Resource utilization rate The average age of the system's information. , , These are the weighting coefficients; Step C4.2, Fine-grained power allocation on a small time scale (based on an improved Dueling DQN algorithm): Building upon the bandwidth allocation on a large time scale, [the following steps are performed]. , Each data point is allocated its own transmission power, ensuring that the minimum bandwidth and power requirements of each data point are met. An improved Dueling DQN algorithm is then used to... Value decoupling into state value function With advantage function The formula is: ; The improved Dueling DQN algorithm introduces a power constraint factor. (System maximum transmission power) and bit error rate constraints By adding a bit error rate penalty term to the dominance function to avoid transmission failures caused by excessively low power allocation, the modified dominance function is as follows: ,in The reward value is calculated based on resource utilization and transmission latency. The penalty coefficient is... This represents the actual bit error rate. It should be noted here that The reward value is based primarily on system resource utilization. With data transmission latency The calculation (aligning with the optimization requirements of fine-grained power allocation on a small time scale, consistent with the goal of the improved dual-time-scale resource scheduling algorithm) involves the following specific calculation steps: Step 1: Parameter Acquisition and Standardization. Obtain the current status from the data feedback module. Execute power distribution action The two core parameters that follow are resource utilization rate. (The ratio of actual bandwidth / power used by the system to total bandwidth / total power), transmission delay (Short package) Long bag (The time delay threshold is consistent with that set in the resource demand analysis); the min-max normalization algorithm, consistent with that of the big data preprocessing module, is used to map the two parameters to the 0-1 range, eliminate the difference in dimensions, and ensure the uniformity of calculation.

[0024] Step 2: Parameter Weighting. Based on the core objective of small-timescale power allocation (balancing resource utilization and transmission latency, while adapting to the requirements of short packets and low latency), set the weighting coefficients: resource utilization weight. Short packet transmission delay weight Long packet transmission delay weight (satisfy Priority will be given to ensuring short packet latency and resource utilization.

[0025] Step 3: Calculation of individual indicators. ① Resource utilization rate incentive item: , The higher the value, the greater the reward, incentivizing the algorithm to improve resource utilization; ② Short packet latency reward: ,in (Maximum allowable latency for short packets), the shorter the latency, the greater the reward value, which aligns with the low latency requirements of short packets; ③ Latency reward for long packets: ,in (Maximum allowed latency for long packets), the shorter the latency, the greater the reward value.

[0026] Step 4: Total Reward Value Synthesis. Add the three individual reward values ​​together to obtain the total reward value. The value ranges from 0 to 1; if the short packet delay exceeds the limit ( ) or resource utilization is too low ( If the reward value is less than 0.2 points, an additional penalty point will be deducted from the total reward value to ensure that the reward value accurately reflects the power distribution action. The advantages and disadvantages of the function provide reliable input for the dominant function.

[0027] Step C5, Resource Allocation Execution: The resource allocation execution unit calculates the bandwidth allocation scheme based on the improved resource scheduling algorithm. With power distribution scheme The data is sent to the communication execution module, which, in conjunction with NOMA technology, enables the parallel transmission of long and short packet data (short packet data and long packet data are transmitted on the same time and frequency resources, and are distinguished by power domain multiplexing). The ground control center uses serial interference cancellation (SIC) technology to decode the data, first decoding the short packet data with higher priority, and then decoding the long packet data. Step C6, Scheduling Optimization: The data feedback module feeds back the resource utilization, bit error rate, and transmission delay during transmission to the scheduling optimization unit, which then adjusts the algorithm parameters (such as weighting coefficients) in real time. , , Penalty coefficient When resource utilization is below a threshold (e.g., 70%), adjust the bandwidth allocation ratio to increase bandwidth supply for high-priority data; when the bit error rate is above a certain threshold... At the same time, increase the transmission power of the corresponding data to ensure the optimization and stability of resource allocation. The specific implementation steps are as follows: Step 1: Feedback Data Reception and Analysis. The data feedback module collects three types of core data in real time during the transmission process (resource utilization rate...). Bit error rate Transmission delay After standardizing the data (to match the normalization standard of the big data preprocessing module), it is sent to the scheduling optimization unit in real time. The scheduling optimization unit parses the received data and matches it with the optimization objectives of large-scale bandwidth allocation (maximizing resource utilization, minimizing average bandwidth utilization). Constraints on small timescale power allocation (bit error rate ≤ (Meeting minimum power requirements for data) and simultaneously associating the corresponding data transmission sequence. This clarifies abnormal transmission parameters for various data types (short / long packets, different priorities).

[0028] Step 2: Multi-dimensional threshold determination. The scheduling optimization unit presets threshold ranges for three types of core data. These thresholds are based on the requirements of the UAV power line inspection scenario and the system hardware performance (total bandwidth). Maximum transmission power The training objective settings for the improved algorithm described above, specifically the following thresholds: resource utilization threshold. (Above this threshold indicates unreasonable resource allocation and wasted bandwidth / power), Bit error rate threshold Transmission delay threshold (short packet data) Long package data Consistent with the latency requirements in the resource demand analysis above, a value higher than this threshold indicates that resource allocation cannot meet the data timeliness requirements. The scheduling optimization unit compares the parsed real-time data with the corresponding threshold to determine whether the parameter adjustment process needs to be initiated. Specifically, there are three judgment results: only a single indicator is abnormal, two or more indicators are abnormal, and all indicators are normal (no adjustment is required).

[0029] Step 3: Scenario-specific parameter adjustment. Based on the judgment results, the scheduling optimization unit adjusts the key parameters of the improved dual-timescale resource scheduling algorithm (the weight coefficients of the large-timescale improved DDPG algorithm). , , Penalty coefficient of the improved Dueling DQN algorithm with small time scale Targeted adjustments will be made, along with iterations of bandwidth and power allocation schemes, to ensure that the adjusted parameters match the algorithm logic and scenario requirements. Specifically, this will be done across three scenarios: Scenario 1: Resource utilization only (Bandwidth / Power Waste). This indicates that the coarse-grained bandwidth allocation ratio on a large time scale is unreasonable, or that there is redundancy in the power allocation on a small time scale. It is necessary to prioritize adjusting the bandwidth allocation ratio on the large time scale, while simultaneously fine-tuning the algorithm's weighting coefficients. , , ① Increase The value of (e.g., adjusted from 0.4 to 0.5), decrease The value of (e.g., adjusted from 0.2 to 0.1) makes the improved DDPG algorithm... The value function focuses more on optimizing resource utilization; ② Adjust the bandwidth allocation ratio to reduce the bandwidth supply for low-priority data (such as the low-resolution, large-volume portion of long packet data), and allocate redundant bandwidth to high-priority data (such as short packet fault data and long packet high-resolution image data), while ensuring the minimum bandwidth requirement for short packet data. Bandwidth requirements for long packet data The following conditions are met: ③ If power allocation is redundant (e.g., some data transmission power exceeds the minimum power requirement); And the error rate is much lower than By appropriately reducing the transmission power of the corresponding data and allocating redundant power to high-priority data, the overall resource utilization rate can be improved.

[0030] Scenario 2: Bit error rate only (Insufficient transmission reliability). This indicates that the fine-grained power allocation on a small time scale is unreasonable, and insufficient power supply leads to transmission errors. The power allocation scheme on the small time scale should be adjusted first, while the penalty coefficient should be fine-tuned. ① Increase the penalty coefficient The value of (e.g., adjusting from 0.3 to 0.5) increases the weight of the bit error rate penalty term in the advantage function of the improved Dueling DQN algorithm, preventing subsequent power allocation from being too low; ② Locate data with excessive bit error rate and prioritize increasing the transmission power of high-priority data (such as short packet fault data) to ensure its transmission reliability. If the total system power does not reach the limit... Simultaneously increase the transmission power of corresponding long packet data; if the total system power has reached its limit, appropriately reduce the transmission power of low-priority long packet data, allocating limited power to high-priority data with excessive bit error rate, while ensuring that the power of all data after adjustment is not lower than the minimum power requirement. .

[0031] Scenario 3: Transmission latency exceeds the standard (short packet) or long bag (This refers to an issue where multiple indicators are abnormal simultaneously.) If the short packet latency exceeds the standard, it indicates insufficient bandwidth / power supply for short packet data. Priority should be given to ensuring short packet data: increase the bandwidth allocation ratio for short packet data, increase transmission power, and adjust... (average Adjusting the weight value (e.g., from 0.4 to 0.5) makes the algorithm more focused on reducing short packets. Values ​​and transmission latency; if the long packet latency exceeds the standard, it indicates insufficient bandwidth allocation for long packet data. Appropriately increase the bandwidth supply for high-priority long packet data and adjust accordingly. The value of (e.g., adjusted from 0.1 to 0.2), combined with the channel prediction factor. Optimize bandwidth allocation; if multiple indicators are abnormal at the same time (such as low resource utilization and high bit error rate), adjust the power allocation first to solve the bit error rate problem, and then optimize the bandwidth allocation to improve resource utilization. Avoid adjustment conflicts and ensure the orderly adjustment of parameters.

[0032] Step 4: Parameter Verification and Scheme Iteration. After the parameters are adjusted, the scheduling optimization unit will use the new algorithm parameters ( , , , The data is sent to the improved resource scheduling algorithm unit, which recalculates the bandwidth and power allocation scheme. Simultaneously, the scheduling optimization unit receives subsequent feedback data from the data feedback module in real time to verify whether the adjusted indicators meet the threshold requirements. If all indicators meet the requirements, the current parameters and allocation scheme are maintained. If any indicators are still abnormal, steps two and three are repeated until all indicators reach the preset threshold. If the requirements still cannot be met after multiple adjustments (e.g., due to channel mutations), an alarm mechanism is triggered to notify the ground control center to adjust the UAV's flight status (e.g., reduce flight speed, shorten flight distance) to help improve channel quality and ensure the stability of resource allocation.

[0033] For the communication execution module: it adopts conventional NOMA communication technology to complete data transmission and decoding, without any innovation; its core function is to coordinate with the two innovative modules according to the allocation scheme of the channel adaptive resource scheduling module to achieve efficient and parallel transmission of heterogeneous data of long and short packets, ensuring that the data is accurately delivered to the ground control center. The specific working steps are as follows: Step 1: Receiving the Allocation Scheme. This step involves receiving the bandwidth and power allocation scheme sent by the channel adaptive resource scheduling module. This includes the specific bandwidth allocation values ​​and transmission power parameters for various short and long data packets, as well as the data transmission sequence generated by the heterogeneous data priority dynamic allocation module, clarifying the data transmission order and resource parameters.

[0034] Step 2: Data power domain multiplexing. In accordance with the characteristics of NOMA technology, data of different priorities are multiplexed in the power domain on the same time and frequency resources. Higher transmission power is allocated to high-priority short packets, and lower transmission power is allocated to low-priority long packets. Different data are distinguished by power differences to avoid transmission interference and meet the needs of short packet priority scenarios.

[0035] Step 3: Parallel Data Transmission. Based on the allocated bandwidth and power parameters, the data transmission process is initiated, and various types of data are transmitted sequentially according to the transmission sequence. Short packet data and long packet data are transmitted in parallel, ensuring both the low latency requirement of short packet data and the efficient transmission of long packet data, thus adapting to the high mobility scenarios of UAV inspection.

[0036] Step 4: Data Reception and Decoding. The ground control center receives the transmitted data and uses Serial Interference Cancellation (SIC) technology for decoding. It strictly follows the data priority order, first decoding the high-priority short packets to eliminate their interference with the low-priority long packets, and then decoding the long packets to ensure decoding accuracy.

[0037] Step 5: Transmission Status Feedback. The data transmission completion status and decoding results are fed back to the data feedback module. This allows the data feedback module to combine parameters such as transmission delay and bit error rate to support the dynamic adjustment of the heterogeneous data priority allocation module and the channel adaptive resource scheduling module, thus realizing a closed-loop transmission system.

[0038] For the data feedback module: it collects various parameters during the transmission process in real time and feeds them back to the heterogeneous data priority dynamic allocation module and the channel adaptive resource scheduling module. The specific working steps are as follows: Step 1: Determine the acquisition parameters. Clarify the acquisition scope and core parameters to fully match the adjustment requirements of the heterogeneous data priority dynamic allocation module and the channel adaptive resource scheduling module. This includes transmission latency during the communication execution module's transmission process (divided into short packets and long packets, corresponding to the latency threshold requirements of the heterogeneous data priority dynamic allocation module), bit error rate (corresponding to the bit error rate constraints of the channel adaptive resource scheduling module), and information age (…). Value, which is the core factor for priority calculation, and resource utilization rate (which is the optimization objective for resource scheduling).

[0039] Step 2: Real-time parameter acquisition. The feedback unit deployed at the ground control center synchronously acquires the transmission status data of the communication execution module, matching the acquisition frequency with channel status changes and transmission rhythm (set to 5Hz) to ensure data real-time performance; at the same time, it associates the data transmission sequence and labels the transmission parameters corresponding to each type of data (short packet / long packet, different priorities) to avoid parameter and data mismatch and provide a basis for subsequent precise adjustments.

[0040] Step 3: Data Preprocessing and Filtering. The collected raw feedback data undergoes simple preprocessing to remove outliers caused by transient interference. A normalization standard consistent with the big data preprocessing module is used to map parameters to a uniform range, ensuring that the feedback data format is consistent with the input data format of the core module, thus avoiding the impact of differences in data units on adjustment accuracy.

[0041] Step 4: Precise Feedback by Module. Based on the parameter type, the feedback data is directed to the heterogeneous data priority dynamic allocation module and the channel adaptive resource scheduling module. The values ​​and transmission delays are fed back to the heterogeneous data priority dynamic allocation module to adjust the weight coefficients; the resource utilization and bit error rate are fed back to the channel adaptive resource scheduling module to optimize the algorithm parameters (weight coefficients, penalty coefficients) and resource allocation scheme to achieve precise adaptation.

[0042] Step 5: Feedback-based closed-loop maintenance. Continuously collect the adjusted transmission parameters and provide real-time feedback to the corresponding modules to verify the adjustment effect. If the adjusted parameters still do not reach the preset threshold, trigger the adjustment process again. This forms a complete closed loop with the two creative modules and the communication execution module, ensuring dynamic optimization of system resource allocation and priority allocation, and meeting the high mobility requirements of UAV inspection scenarios.

[0043] This invention relates to the field of archival digitization, aiming to solve the problems of low accuracy in identifying multi-source heterogeneous archives and non-compliant electronic data generation, and to provide an intelligent system. This invention includes modules for constructing an archival knowledge graph, improved semantic enhancement recognition, and improved knowledge-guided generation. By preprocessing multi-source heterogeneous original archives, relying on an archival knowledge graph containing entity, relation, and rule layers, an improved algorithm is used to accurately identify archival entities and related information, thereby generating semantically relevant electronic data that conforms to archival standards. After compliance verification, the data is archived and stored. This invention improves the accuracy and compliance of archival electronic data processing, adapts to multimodal archival scenarios, and facilitates intelligent management of digital archives.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart communication distribution system based on big data, characterized in that, The system includes a data acquisition module, a big data preprocessing module, a heterogeneous data priority dynamic allocation module, a channel adaptive resource scheduling module, a communication execution module, and a data feedback module. The data acquisition module collects multi-source raw data during the UAV inspection process. This multi-source raw data includes heterogeneous data of varying lengths and channel status data. Short packet data includes sensor status data and fault alarm data, while long packet data includes inspection images and video data. Channel status data includes channel gain, noise power, and interference intensity. The big data preprocessing module performs denoising, normalization, and feature extraction on the collected multi-source raw data set to obtain standardized heterogeneous data. The system includes: channel state characteristic data; a heterogeneous data priority dynamic allocation module that calculates the dynamic priority of long and short packet data based on preprocessed standardized heterogeneous data using an improved algorithm to determine the data transmission order; a channel adaptive resource scheduling module that allocates communication bandwidth and power resources based on data priority and channel state characteristic data using an improved algorithm; a communication execution module that completes the transmission of heterogeneous data using non-orthogonal multiple access technology according to the priority order and resource allocation results; and a data feedback module that collects data during transmission in real time and feeds it back to the heterogeneous data priority dynamic allocation module and the channel adaptive resource scheduling module to achieve dynamic adjustment and optimization. The heterogeneous data priority dynamic allocation module includes a data feature extraction unit, an urgency assessment unit, a channel adaptability calculation unit, an improved priority calculation unit, and a priority sorting unit. The data feature extraction unit extracts key features from the preprocessed heterogeneous data; the urgency assessment unit evaluates the urgency of the standardized heterogeneous data; the channel adaptability calculation unit calculates the adaptability of different types of standardized heterogeneous data to the current channel state; the improved priority calculation unit calculates the dynamic priority of various types of standardized heterogeneous data using an improved algorithm; and the priority sorting unit determines the standardized heterogeneous data transmission order based on the priority calculation results. The channel adaptive resource scheduling module includes a channel state prediction unit, a resource demand analysis unit, an improved resource scheduling algorithm unit, a resource allocation execution unit, and a scheduling optimization unit. The channel state prediction unit predicts the channel state in the near future based on big data. The resource demand analysis unit analyzes the resource demands of various standardized heterogeneous data based on data priority sequences and data characteristics. The improved resource scheduling algorithm unit calculates bandwidth and power allocation schemes through improved algorithms. The resource allocation execution unit executes the resource allocation schemes and coordinates with the communication execution module to complete data transmission. The scheduling optimization unit optimizes resource allocation parameters based on feedback data.

2. The intelligent communication distribution system based on big data according to claim 1, characterized in that: The specific implementation logic of the data acquisition module is as follows: Step A1, Deployment and Initialization of Data Acquisition Module: Integrate and deploy the various components of the data acquisition module on the UAV, link it with the UAV flight control system and edge computing unit, and complete the initial configuration; Step A2, Determining the Data Collection Scope and Parameters: Based on the requirements of the UAV power line inspection task, clarify the data collection scope and core parameters to ensure that the collected data meets the system requirements. Heterogeneous data acquisition scope: Short packet data focuses on sensor status data and fault alarm data of key parts of the transmission tower, with the data length strictly controlled between 20-50 bytes; Long packet data focuses on image and video data of the appearance of the transmission line and transmission tower. Channel data acquisition range: Synchronously acquire the channel gain of the communication link between the UAV and the ground control center. Noise power Interference intensity It covers the entire flight trajectory of UAV inspection, ensuring the continuity of channel state data and providing complete data support for subsequent channel state prediction and channel adaptability calculation; Acquisition triggering mechanism: A combination of periodic acquisition and event-triggered acquisition is adopted. Short packet data and channel data are acquired periodically at a preset frequency. In addition to periodic acquisition, long packet data will automatically increase the camera acquisition frequency to 5 frames / second when the sensor detects a fault signal, focusing on capturing images of the fault area to ensure the timeliness of fault-related long packet data. Step A3: Synchronous Acquisition of Multi-Source Raw Data: The UAV flies along the preset inspection route, and all acquisition components work together to achieve synchronous acquisition of multi-source raw data, avoiding data timing deviations. Short packet data acquisition: Sensors capture the status of transmission tower equipment in real time. When a data change is detected, the data is immediately acquired and packaged into short packets, with the data generation time marked. Temporarily stored in the drone's local cache; Long-term data acquisition: High-definition cameras simultaneously capture images and videos of the inspection area, and automatically label the resolution of each captured image / video frame. Data volume Encapsulate the data into a long packet and temporarily store it in the local cache; Channel data acquisition: The channel state monitoring module acquires the channel parameters of the communication link in real time. Each time data is acquired, the acquisition time is marked and correlated with the heterogeneous data acquired at the same time to ensure the correspondence between data and channel state in subsequent calculations and avoid the accuracy of channel adaptability calculation and resource scheduling due to timing deviation. Step A4: Data Preprocessing and Preliminary Verification: After the collected multi-source raw data is temporarily stored in the UAV's local cache, simple preliminary preprocessing and verification are performed to reduce invalid data being transmitted to subsequent modules: Preliminary denoising: Simple mean filtering is applied to the short packet data collected by the sensors to remove obvious outliers; the channel data is smoothed to filter abnormal fluctuations caused by instantaneous channel changes; Data verification: The length of short packet data, the resolution and data volume labeling integrity of long packet data, and the rationality of channel data parameters are verified, and invalid data is removed; Data encapsulation: The verified short packet data, long packet data, and channel data are encapsulated in a unified format, labeled with the collection time and UAV location information, forming a standardized multi-source raw data set; Step A5: Real-time data transmission to the big data preprocessing module: The data acquisition module transmits the packaged standardized multi-source raw data set to the big data preprocessing module in real time through the internal link of the UAV edge computing unit. A simple error control mechanism is used during the transmission process to ensure that the data transmission is free of loss and errors. At the same time, a data backup interface is reserved so that when the transmission link is temporarily interrupted, the acquired data will be temporarily stored in local storage and retransmitted after the link is restored.

3. The intelligent communication distribution system based on big data according to claim 1, characterized in that: The deployment of the data acquisition module in step A1 specifically includes the following: Sensor deployment: Temperature sensors, humidity sensors, and fault detection sensors are installed on the drone body and inspection attachment points. The sensor sampling frequency is set to 10Hz, and the sampling accuracy matches the power inspection standard to ensure the real-time acquisition of short package data; High-definition camera deployment: A high-definition adjustable camera is installed, supporting switching between 1080P, 720P, and 480P resolutions. The lens angle and focal length are dynamically adjusted according to the inspection distance. The acquisition frequency is set to 1 frame / second and 25 frames / second to ensure the clarity and integrity of long package data; Channel status monitoring module deployment: Integrates a channel monitoring unit to capture the channel gain, noise power, and interference intensity of the Ricean channel in real time. The acquisition frequency is matched with the channel status change rate and set to 5Hz to adapt to the dynamic fluctuations in the channel caused by the high mobility of the UAV; Initialization calibration: After starting the UAV, all acquisition components are calibrated to ensure that the sensor zero drift error is within the preset range, the camera is accurately focused, and the communication link between the channel monitoring module and the ground is normal, so as to avoid abnormal acquisition data.

4. The intelligent communication distribution system based on big data according to claim 1, characterized in that: The specific implementation steps of the heterogeneous data priority dynamic allocation module are as follows: Step B1, Data Feature Input: The big data preprocessing module will input the standardized heterogeneous data, i.e., the short packet data set. Long package dataset Channel gain of channel state characteristic data Noise power The data feature extraction unit is input into the heterogeneous data priority dynamic allocation module. Step B2, Data Feature Extraction: The data feature extraction unit extracts key features from the short packet data: fault level. Data generation time Key features to extract from long packet data: data volume Resolution Simultaneously, extract the channel gain from the channel state features. This is used for subsequent adaptation calculations; Step B3, Data Urgency Assessment: The urgency assessment unit uses a tiered assessment method to calculate the urgency of short packet data and long packet data separately. For short packet data Its urgency The formula for calculation is: ,in The current system time. This is the time decay coefficient; For long packet data Its urgency The formula for calculation is: ,in To support the maximum long packet data size; Step B4, Channel Adaptability Calculation: The channel adaptability calculation unit calculates the adaptability between the standardized heterogeneous data and the current channel state. A higher adaptability indicates higher reliability and lower resource consumption for data transmission under the current channel. For short packet data, the focus is on low-latency adaptability. The formula for calculation is: ,in The channel gain of the current channel. For short packet data length, The signal-to-interference-plus-noise ratio (SINR) of the channel is used. The higher the SINR and the shorter the data length, the better the adaptability, which meets the requirements of low-latency transmission of short packet data. For long packet data, the focus is on bandwidth adaptation and adaptability. The formula for calculation is: ,in The higher the signal-to-interference-plus-noise ratio, the more abundant the available bandwidth, and the smaller the data volume, the higher the adaptability, meeting the requirements for efficient transmission of long packet data. Step B5, Dynamic Priority Calculation: The improved priority calculation unit adopts an improved AoI-weighted collaborative algorithm, which integrates urgency, channel adaptability, and information age. Calculate the dynamic priority of standardized heterogeneous data. , The formula is: ; ,in , , Here are the weighting coefficients for short packet data, and , Prioritize matters of urgency; , , The weighting coefficients for long-package data. , Prioritize channel adaptability, among which The information about age in short packet data. The information age of the long package data; Step B6, Priority Sorting: The priority sorting unit sorts the priorities of all short and long data packets uniformly, generating a data transmission sequence according to the order of priority from high to low. The sequence is then sent to the channel adaptive resource scheduling module as the core basis for resource allocation. Step B7, Dynamic Adjustment: The data feedback module will adjust the data during the transmission process. Values ​​and transmission latency are fed back to the heterogeneous data priority dynamic allocation module, which adjusts the weighting coefficients in real time. , , and , , When short packet data When the value exceeds the threshold, increase. The value of is increased when the long packet data transmission delay exceeds the threshold. The value of is determined to ensure the dynamic adaptability of priority allocation.

5. The intelligent communication distribution system based on big data according to claim 1, characterized in that: The specific implementation steps of the channel adaptive resource scheduling module are as follows: Step C1, Input Data Reception: Receive the data transmission sequence sent by the heterogeneous data priority dynamic allocation module. And standardized channel state data sent by the big data preprocessing module, including historical channel gain sequences. Noise power Interference intensity ; Step C2, Channel State Prediction: The channel state prediction unit uses an LSTM neural network, based on historical channel gain sequences. Predicting the future Channel gain per time slot The prediction formula is: ,in For the parameters of the LSTM neural network, At the same time, combined with the flight speed of the drone The channel gain prediction value is corrected using the following formula: ,in, The time interval for each time slot, The distance between the drone and the ground control center; Step C3, Resource Requirements Analysis: The resource requirements analysis unit analyzes the data transmission sequence... Analyze the characteristics of various types of data to understand their bandwidth and power requirements: For short packet data Calculate the minimum bandwidth requirement based on the formula for the maximum coding rate of short packet communication. : ,in Maximum allowable latency for short data packets; minimum power requirement The formula for calculation is: For long packet data Calculate the bandwidth requirements based on the data volume and transmission time requirements. With power demand : ; ,in Transmission time for long packet data (dynamically allocated based on priority; higher priority means shorter transmission time). Step C4, Improved resource scheduling algorithm calculation: The improved resource scheduling algorithm unit adopts an improved dual-time-scale resource scheduling algorithm, which divides resource allocation into coarse-grained bandwidth allocation at a large time scale and fine-grained power allocation at a small time scale. Step C5, Resource Allocation Execution: The resource allocation execution unit calculates the bandwidth allocation scheme based on the improved resource scheduling algorithm. With power distribution scheme The data is sent to the communication execution module, which, in conjunction with NOMA technology, enables the parallel transmission of long and short packet data. The ground control center uses serial interference cancellation technology to decode the data, first decoding the short packet data with higher priority, and then decoding the long packet data. Step C6, Scheduling Optimization: The data feedback module feeds back the resource utilization, bit error rate, and transmission delay during transmission to the scheduling optimization unit. The scheduling optimization unit adjusts the algorithm parameters in real time. When the resource utilization is lower than the specified level, it adjusts the bandwidth allocation ratio to increase the bandwidth supply for high-priority data; when the bit error rate is higher than the specified level, it adjusts the bandwidth allocation ratio to increase the bandwidth supply for high-priority data. At the same time, increase the transmission power of the corresponding data to ensure the optimization and stability of resource allocation.

6. The intelligent communication distribution system based on big data according to claim 1, characterized in that: The specific implementation steps of the improved resource scheduling algorithm in step C4 are as follows: Step C4.1, Large-scale coarse-grained bandwidth allocation: Based on total bandwidth As constraints, to maximize system resource utilization and minimize average resource utilization. To achieve this, bandwidth is allocated to different types of data, including short packet data. Total bandwidth of long packet data ,satisfy The policy network update formula for the improved DDPG algorithm is: ,in For policy network parameters; The gradient of the objective function of the policy network; The input state of the policy network is specifically the standardized set of channel state features and data features; for Value functions; Let this be the policy function, representing the state at input. Below, output the optimal bandwidth allocation action on a large time scale. ; For experience replay pool; Subsequently Introducing channel prediction factor into the value function Improved channel adaptability of bandwidth allocation, corrected The value function is: ,in Resource utilization rate The average age of the system's information. , , These are the weighting coefficients; Step C4.2, Fine-grained power allocation on a small time scale: Based on the bandwidth allocation on a large time scale, ... , Each data point is allocated its own transmission power, ensuring that the minimum bandwidth and power requirements of each data point are met. An improved Dueling DQN algorithm is then used to... Value decoupling into state value function With advantage function The formula is: ; The improved Dueling DQN algorithm introduces a power constraint factor. With bit error rate constraints By adding a bit error rate penalty term to the dominance function to avoid transmission failures caused by excessively low power allocation, the modified dominance function is as follows: ,in The reward value is calculated based on resource utilization and transmission latency. The penalty coefficient is... This represents the actual bit error rate.

Images