Unmanned tower operation state monitoring system based on big data

Abstract

The invention relates to the technical field of unmanned control tower operation state monitoring, and discloses an unmanned control tower operation state monitoring system based on big data. The system comprises a data acquisition unit for acquiring parameters such as a communication link, weather, navigation, electric power and an aircraft trajectory of the unmanned control tower in real time; the multi-source data fusion unit is used for carrying out data fusion by using a federated learning algorithm to generate a multi-modal joint feature matrix; the dynamic modeling unit is used for constructing a state transition probability graph based on a dynamic Bayesian network and outputting a real-time state evolution sequence; the real-time anomaly detection unit is used for calculating anomaly confidence through an isolated forest algorithm and an LSTM auto-encoder; the active control instruction generation unit is used for generating an equipment regulation and control instruction set according to the abnormal confidence in combination with reinforcement learning; and a knowledge graph reasoning unit is also arranged to optimize the state transition probability graph. According to the system, the operation state of the unmanned control tower can be comprehensively and accurately monitored and regulated in real time, the operation safety and stability are improved, and the operation risk is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring the operating status of unmanned tower control, and particularly to a monitoring system for the operating status of unmanned tower control based on big data. Background Art

[0002] With the rapid development of the aviation industry, unmanned tower control shows great potential in improving airport operation efficiency and reducing labor costs. However, the safety monitoring of its operating status faces many challenges. The traditional manned tower control monitoring mode relies on manual duty, and the staff can directly observe and operate various devices. However, unmanned tower control lacks real-time human intervention. Once a device fails or operates abnormally, it cannot be detected and processed in time, which may lead to serious safety accidents. For example, a communication link failure may cause the interruption of information transmission between the tower control and the aircraft, paralyzing the command and dispatch; when the meteorological conditions change suddenly, if accurate meteorological data cannot be obtained in time, the takeoff and landing safety of the aircraft will be threatened.

[0003] In terms of data collection and processing, there are obvious deficiencies in the existing technologies. The data collected by different types of sensors and devices have different formats and inconsistent standards, forming data islands, which are difficult to integrate and analyze effectively. For example, there is a lack of correlation analysis between communication link status parameters, meteorological sensor monitoring parameters, navigation equipment operation parameters, etc., and it is impossible to grasp the operating conditions of the unmanned tower control as a whole. At the same time, in the face of a large amount of real-time data, traditional data processing methods are inefficient and cannot meet the real-time requirements, and it is difficult to quickly and accurately identify potential abnormal situations.

[0004] For anomaly detection and processing, existing monitoring systems usually adopt simple threshold judgment methods, with high false alarm rates and missed alarm rates. When multiple parameters change slightly at the same time, this method cannot comprehensively judge whether there is an anomaly, and it is easy to ignore potential fault hazards. Moreover, after an anomaly is detected, there is a lack of effective active control strategies, and it is impossible to accurately regulate the equipment in time to properly solve the problem. For example, when the accuracy of the navigation equipment decreases, it is impossible to quickly formulate and execute effective calibration measures, affecting the navigation accuracy and flight safety of the aircraft.

[0005] In addition, existing unmanned tower control monitoring systems lack in-depth exploration of the correlation relationships between devices. Each device does not operate in isolation, but is interrelated and interacts with each other. However, existing technologies cannot make full use of this correlation relationship for comprehensive risk assessment and prediction, and it is difficult to discover potential fault risks in advance, and it cannot provide strong support for the maintenance and management of the equipment. Therefore, there is an urgent need for a monitoring system for the operating status of unmanned tower control based on big data to solve the above problems and ensure the safe and stable operation of the unmanned tower control. Summary of the Invention

[0006] The object of the present invention is to provide an unmanned tower operation status monitoring system based on big data to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: An unmanned tower operation status monitoring system based on big data, the system includes: A data acquisition unit, which is used to collect the communication link status parameters, meteorological sensor monitoring parameters, navigation equipment operation parameters, power equipment load parameters and aircraft dynamic trajectory parameters of the unmanned tower in real time through a stream data processing framework; A multi-source data fusion unit, which is used to perform cross-domain feature extraction and heterogeneous data fusion on the communication link status parameters, meteorological sensor monitoring parameters, navigation equipment operation parameters, power equipment load parameters and aircraft dynamic trajectory parameters through a federated learning algorithm to generate a multi-modal joint feature matrix; A dynamic modeling unit, which is used to construct a probability graph of the operation state transition of the unmanned tower based on a dynamic Bayesian network with the multi-modal joint feature matrix as the input and output a real-time state evolution sequence; A real-time anomaly detection unit, which is used to calculate the time series anomaly score of the real-time state evolution sequence through an isolation forest algorithm and an LSTM autoencoder to generate an anomaly confidence level; An active control instruction generation unit, which is used to generate a device regulation instruction set according to the anomaly confidence level in combination with a reinforcement learning policy network.

[0008] Preferably, the execution steps of the real-time anomaly detection unit include: Dividing the real-time state evolution sequence into time window segments and extracting the statistical feature vectors within each time window; Calculating the reconstruction error of the statistical feature vectors through the LSTM autoencoder to obtain a first anomaly score; Calculating the path length deviation of the statistical feature vectors through the isolation forest algorithm to obtain a second anomaly score; Performing weighted fusion on the first anomaly score and the second anomaly score to generate the anomaly confidence level.

[0009] Preferably, the execution steps of the data acquisition unit include: Receiving the signal strength, bit error rate and link delay data in the communication link status parameters in real time through the Apache Kafka stream processing platform and the communication protocol interface of the unmanned tower; Collecting the wind speed, visibility and pressure gradient data in the meteorological sensor monitoring parameters through distributed sensor nodes; Extract the gyroscope drift value, GPS positioning deviation, and radar scanning frequency data in the operating parameters of the navigation device through the device log parsing engine.

[0010] Preferably, the execution steps of the multi-source data fusion unit further include: Construct a federated learning global model, where each tower device acts as a local client and uploads the encrypted local feature gradients; Perform a secure weighted average of the local feature gradients through the aggregation server to update the global model parameters; Send the updated global model parameters to each client to iteratively generate the multi-modal joint feature matrix.

[0011] Preferably, the execution steps of the dynamic modeling unit include: Define the node variables of the dynamic Bayesian network according to the historical operation data, including communication link stability, meteorological interference level, navigation accuracy attenuation factor, power load fluctuation index, and aircraft trajectory offset; Calculate the conditional probability distribution of the node variables through the Markov chain Monte Carlo method; Update the edge weights of the state transition probability graph based on the real-time data stream to generate the real-time state evolution sequence.

[0012] Preferably, the execution steps of the real-time anomaly detection unit further include: Set the hidden layer dimension of the LSTM autoencoder to 30% of the input feature dimension and use a Dropout layer to prevent overfitting; Introduce an adaptive subtree splitting strategy in the isolation forest algorithm to dynamically adjust the anomaly segmentation threshold according to the data distribution.

[0013] Preferably, the execution steps of the active control instruction generation unit include: Construct the state space of the reinforcement learning policy network, including the anomaly confidence level, device health index, and environmental risk level; Define the reward function as the entropy reduction amplitude of the state transition matrix after device regulation; Iteratively update the policy network parameters through the proximal policy optimization algorithm and output the device regulation instruction set.

[0014] Preferably, the generation steps of the device regulation instruction set further include: Generate an emergency shutdown instruction for high-priority anomaly events and trigger the redundant device switching protocol; Generate parameter dynamic compensation instructions for medium- and low-priority anomaly events, including communication link power adjustment, navigation device calibration coefficient correction, and power load balanced distribution.

[0015] Preferably, the system further includes: A knowledge graph reasoning unit for constructing an unmanned tower equipment association graph based on a graph neural network and extracting key node influencing factors; Among them, the execution steps of the knowledge graph reasoning unit include: Mapping the communication link state parameters, navigation equipment operation parameters, and power equipment load parameters into graph nodes; Calculating the dependence strength between nodes through a graph attention mechanism to generate the key node influencing factors; Inputting the key node influencing factors into the dynamic modeling unit to optimize the state transition probability graph.

[0016] Preferably, the present invention further includes an electronic device, and the device includes: A memory for storing a computer program of the above-mentioned big data-based unmanned tower operation state monitoring system; A processor for executing the computer program to realize real-time monitoring and control of the operation state of the unmanned tower.

[0017] Compared with the prior art, the beneficial effects of the present invention are: The big data-based unmanned tower operation state monitoring system of the present invention has many significant beneficial effects. In terms of data acquisition and fusion, the system uses a stream data processing framework and a federated learning algorithm to collect multi-source heterogeneous data in real time and perform efficient fusion. The data acquisition unit can obtain parameters such as communication links, meteorology, navigation, power, and aircraft trajectories through various methods, breaking data islands and enabling various types of data to work together. The multi-source data fusion unit uses a federated learning algorithm to complete cross-domain feature extraction and heterogeneous data fusion while protecting data privacy, generating a multi-modal joint feature matrix. This enables the system to have a more comprehensive and accurate understanding of the operation state of the unmanned tower, providing a solid data foundation for subsequent analysis and decision-making. For example, after fusing the communication link state parameters with the meteorological sensor monitoring parameters, the impact of meteorological conditions on communication quality can be analyzed, and corresponding measures can be taken in advance to ensure the stability of communication.

[0018] It has outstanding dynamic modeling and anomaly detection capabilities. The dynamic modeling unit is based on a dynamic Bayesian network. Using the fused data as input, it constructs a running state transition probability graph and outputs a real-time state evolution sequence, which can accurately capture the dynamic changes in the system running state. The real-time anomaly detection unit combines the isolation forest algorithm and the LSTM autoencoder to calculate the anomaly score for the state evolution sequence and generate an anomaly confidence level. This collaborative approach of multiple algorithms greatly improves the accuracy and timeliness of anomaly detection. By dividing the sequence into time window segments to extract statistical feature vectors, calculating the anomaly scores with the two algorithms respectively and then fusing them, various complex anomaly situations can be effectively identified. For example, during the operation of a navigation device, even if the parameter changes are relatively subtle, potential anomalies can be detected in a timely manner. Compared with traditional threshold judgment methods, the false alarm rate and missed alarm rate are greatly reduced.

[0019] The active control instruction generation mechanism is efficient and practical. The active control instruction generation unit generates a device regulation instruction set according to the anomaly confidence level, in combination with a reinforcement learning policy network. A reasonable state space and reward function are constructed, and the policy network parameters are iteratively updated through the proximal policy optimization algorithm, making the generated regulation instructions accurate and effective. For different priority anomaly events, the system will take different measures. When a high-priority anomaly event occurs, an emergency shutdown instruction is quickly generated and a redundant device switching protocol is triggered to ensure the continuous operation of key functions; for medium- and low-priority anomaly events, parameter dynamic compensation instructions are generated, such as adjusting the communication link power, correcting the navigation device calibration coefficient, and balancing the power load distribution, etc., to repair the anomaly in a timely manner without affecting the overall operation and avoid the problem from deteriorating.

[0020] The knowledge graph reasoning unit further optimizes the system performance. This unit constructs an unmanned tower equipment association graph based on a graph neural network, extracts the key node influencing factors, and then inputs them into the dynamic modeling unit to optimize the state transition probability graph. This enables the system to deeply explore the association relationships between devices and conduct risk assessment and prediction from a global perspective. For example, by analyzing the association between the communication link node and the navigation device node, the navigation anomaly risk that may be caused by a communication failure can be predicted in advance, and maintenance can be arranged in advance or preventive measures can be taken to improve the reliability and stability of the system. Description of the Drawings

[0021] Figure 1 It is the working principle diagram of the unmanned tower operation state monitoring system based on big data described in the present invention; Figure 2 It is the working flow chart of the generative adversarial network completion model and data sharing; Figure 3 It is the working flow chart of the multi-objective reinforcement learning algorithm reward function and scenario migration; Figure 4 It is the working flow chart of the spatio-temporal alignment technology and anomaly data detection. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0023] See also Figures 1-4 The present invention provides an unmanned tower operation status monitoring system based on big data, which aims to achieve real-time, accurate monitoring and effective control of the unmanned tower operation status. The specific implementation scheme of the present invention is described in detail below.

[0024] The system primarily consists of a data acquisition unit, a multi-source data fusion unit, a dynamic modeling unit, a real-time anomaly detection unit, and an active control command generation unit. In actual operation, these units work together to ensure the stable operation of the unmanned tower.

[0025] The data acquisition unit uses a streaming data processing framework to collect real-time data from the unmanned tower's communication link status parameters, meteorological sensor monitoring parameters, navigation equipment operating parameters, power equipment load parameters, and aircraft dynamic trajectory parameters. For example, through specific sensors and data interfaces, it connects to various devices in the unmanned tower to continuously obtain these key parameters, providing basic data for subsequent analysis and processing.

[0026] The multi-source data fusion unit uses a federated learning algorithm to extract cross-domain features from various collected parameters and fuse heterogeneous data to generate a multimodal joint feature matrix. Federated learning algorithms can achieve data fusion across different data sources while protecting data privacy, thereby increasing the value of data utilization.

[0027] The dynamic modeling unit, based on a dynamic Bayesian network, uses a multimodal joint feature matrix as input to construct a probability graph of the unmanned tower's operational state transitions and output a real-time state evolution sequence. The dynamic Bayesian network effectively captures dynamic changes in system states, providing a basis for subsequent anomaly detection.

[0028] The real-time anomaly detection unit uses the Isolation Forest algorithm and LSTM autoencoder to calculate a time series anomaly score for the real-time state evolution sequence and generate anomaly confidence. The combination of these two algorithms can more accurately identify anomalies in the system, improving the accuracy and reliability of anomaly detection.

[0029] The active control instruction generation unit generates a device regulation instruction set according to the anomaly confidence level and in combination with the reinforcement learning policy network. When an anomaly is detected, this unit quickly generates corresponding regulation instructions to adjust the devices of the unmanned tower, ensuring the normal operation of the system.

[0030] Through the collaborative work of the above units, the present invention can achieve comprehensive and real-time monitoring and precise regulation of the operating state of the unmanned tower, improving the operating safety and stability of the unmanned tower. The technical solution of the present invention will be further described in detail below through specific embodiments. Embodiment 1

[0031] The real-time anomaly detection unit is responsible for real-time monitoring of the operating state of the unmanned tower and promptly discovering potential anomalies.

[0032] First, the real-time state evolution sequence is divided into time window segments, and the statistical feature vectors within each time window are extracted. Assume the time window length is , and within this time window, the real-time state evolution sequence data collected is . The statistical feature vectors include common statistics such as the mean and variance. The formula for calculating the mean is: where represents the mean in the statistical feature vector, represents the state data at the -th moment within the time window, is the time window length. The formula for calculating the variance is: where represents the variance in the statistical feature vector. By calculating the reconstruction error of the statistical feature vectors through the LSTM autoencoder, the first anomaly score is obtained. The LSTM autoencoder can learn the feature representation of the statistical feature vectors in the normal state. When the input statistical feature vector deviates from the normal state, the reconstruction error will increase. Let the reconstruction error be , the larger it is, the higher the first anomaly score, that is, the greater the possibility of an anomaly.

[0033] By calculating the path length deviation of the statistical feature vectors through the Isolation Forest algorithm, the second anomaly score is obtained. The Isolation Forest algorithm determines the anomaly degree of data points by constructing a tree structure.

[0034] The first anomaly score and the second anomaly score are weighted and fused to generate the anomaly confidence level. Let the weight of the first anomaly score be , the weight of the second anomaly score be , and , the formula for calculating the anomaly confidence level is: where represents the anomaly confidence level, is the first anomaly score, is the second anomaly score, and is the weight coefficient.

[0035] In terms of data acquisition, the data acquisition unit receives the signal strength, bit error rate, and link delay data in the communication link status parameters in real time through the communication protocol interface of the Apache Kafka stream processing platform with the unmanned tower. At the same time, it collects the wind speed, visibility, and barometric gradient data in the meteorological sensor monitoring parameters through the distributed sensor nodes; extracts the gyroscope drift value, GPS positioning deviation, and radar scan frequency data in the navigation equipment operation parameters through the device log parsing engine. The collected data provides basic data support for the real-time anomaly detection unit, ensuring the accuracy and reliability of anomaly detection. Embodiment 2

[0036] The role of the multi-source data fusion unit is to fuse data from different data sources with different formats and characteristics to generate a more valuable multi-modal joint feature matrix, providing high-quality data support for subsequent dynamic modeling and anomaly detection. The specific methods include: Construct a federated learning global model. In this model, each tower device acts as a local client and uploads the encrypted local feature gradients. Suppose there are local clients, and each client owns local data , trains the local data through a certain machine learning algorithm (such as a neural network) locally to obtain the local model , and calculates the local feature gradient . Here, is the gradient vector of the local model parameters, which reflects the influence direction and degree of the local data on the model parameters.

[0037] The local feature gradients are securely weighted averaged by the aggregation server to update the global model parameters. The aggregation server will assign a weight to each local feature gradient according to factors such as the importance or data volume of each client. The weight satisfies and . The update formula for the global model parameters is:

[0038] where represents the updated global feature gradient, which combines the feature gradient information of each local client. Based on this updated global feature gradient, the aggregation server updates the parameters of the global model to obtain a new global model .

[0039] The updated global model parameters are sent to each client, and a multi-modal joint feature matrix is iteratively generated. After each client receives the updated global model parameters, it further trains and optimizes the model using local data. In this process, the model continuously learns the associations and features between data from different data sources, gradually generating a multi-modal joint feature matrix. Suppose the multi-modal joint feature matrix generated after iterations is , and in each iteration, the client updates according to the local data and the global model parameters to make it more accurately reflect the fusion features between different data. As the number of iterations increases, the quality of the multi-modal joint feature matrix will continuously improve, providing a more accurate data basis for subsequent dynamic modeling and anomaly detection. Through such a federated learning process, both the privacy of the data is ensured, and the efficient fusion of multi-source data is achieved, enhancing the system's monitoring ability of the operating status of the unmanned tower. Embodiment 3

[0040] The dynamic modeling unit models the operating status of the unmanned tower by constructing a dynamic Bayesian network, generating a real-time state evolution sequence, providing a key basis for real-time anomaly detection and the generation of active control instructions. The specific method includes: Define the node variables of the dynamic Bayesian network according to the historical operation data, including communication link stability, meteorological interference level, navigation accuracy attenuation factor, power load fluctuation index, and aircraft trajectory offset. Communication link stability is used to measure the reliability of the communication link during data transmission; the meteorological interference level reflects the impact of meteorological conditions on the operation of the unmanned tower; the navigation accuracy attenuation factor reflects the decline in the accuracy of navigation equipment over time; the power load fluctuation index describes the change in the load of power equipment; the aircraft trajectory offset represents the deviation of the actual trajectory of the aircraft from the predetermined trajectory. These node variables can comprehensively reflect the key features of the operating status of the unmanned tower.

[0041] Calculate the conditional probability distribution of the node variables through the Markov chain Monte Carlo method. The Markov chain Monte Carlo method is a random sampling algorithm based on Markov chains, used to approximately calculate the probability distribution in a high-dimensional space. Suppose there are nodes and in the dynamic Bayesian network, the value of node is , the value of node is , then the conditional probability distribution of node under the condition of given node It can be estimated by the Markov Chain Monte Carlo method. Specifically, a Markov chain is constructed such that its stationary distribution is the target conditional probability distribution , and then samples are drawn from this Markov chain to obtain a series of samples, and the estimated value of is calculated based on these samples. For example, when calculating the conditional probability distribution of the communication link stability under the influence of the meteorological interference level, a Markov chain is constructed using historical data. After multiple samplings and statistics, the probabilities of the communication link being in different stable states under different meteorological interference levels are obtained.

[0042] Update the edge weights of the state transition probability graph based on the real-time data stream to generate a real-time state evolution sequence. The real-time data stream is continuously input into the dynamic modeling unit, and the edge weights between the nodes in the dynamic Bayesian network are updated according to the new data. The edge weights reflect the strength of the dependence relationship between the nodes. The larger the weight, the closer the association between the two nodes. Let the edge weight between node and node be . When new real-time data indicates that the dependence relationship between node and node has changed, is adjusted accordingly. As the edge weights are updated, the dynamic Bayesian network evolves continuously, thus generating a real-time state evolution sequence that reflects the changes in the real-time operating state of the unmanned tower. This real-time state evolution sequence provides an accurate reference for subsequent real-time anomaly detection, helping the system to timely discover potential abnormal situations. Through the above steps, the dynamic modeling unit can effectively perform dynamic modeling and analysis on the operating state of the unmanned tower, providing strong support for the stable operation of the entire monitoring system. Embodiment 4

[0043] The real-time anomaly detection unit is used to ensure the safe and stable operation of the unmanned tower, and the active control instruction generation unit is the execution unit that takes effective measures when an abnormal situation occurs. The two cooperate closely to jointly ensure the reliability of the system.

[0044] In terms of the real-time anomaly detection unit, the hidden layer dimension of the LSTM autoencoder is set to 30% of the input feature dimension, and a Dropout layer is used to prevent overfitting. When the LSTM autoencoder reconstructs the statistical feature vector, the hidden layer plays a key role in feature extraction and transformation. Assume that the input feature dimension is , then the hidden layer dimension is set to 。Such a setting is determined after a large number of experiments and data analyses, which can not only effectively extract features but also avoid the overfitting problem caused by an overly complex model. The Dropout layer randomly discards a part of neurons during the training process, so that the model will not overly rely on certain specific neuron connections, thereby improving the generalization ability of the model. Let the dropout probability of the Dropout layer be , and it usually takes values in . For example, take . During the training process, for each neuron in the LSTM autoencoder, with probability , its output is set to 0, which can reduce the co-adaptation phenomenon between neurons and enhance the robustness of the model.

[0045] Introduce an adaptive sub-tree splitting strategy in the Isolation Forest algorithm to dynamically adjust the anomaly segmentation threshold according to the data distribution. When constructing the tree structure in the traditional Isolation Forest algorithm, a fixed splitting strategy and threshold are adopted, which may not be able to well adapt to complex data distributions. The adaptive sub-tree splitting strategy can dynamically select the optimal splitting point and adjust the anomaly segmentation threshold according to the distribution of data points. Suppose at a certain moment, the data distribution in the dataset presents a certain specific pattern. Through the analysis of the data, the algorithm calculates the splitting gain of each potential splitting point and selects the point with the largest splitting gain as the splitting point. At the same time, according to the degree of data dispersion and distribution range, the anomaly segmentation threshold is dynamically adjusted. Through this adaptive strategy, the Isolation Forest algorithm can more accurately identify abnormal data points and improve the accuracy of anomaly detection.

[0046] In terms of the active control instruction generation unit, construct the state space of the reinforcement learning policy network, including the anomaly confidence level, the device health index, and the environmental risk level. The anomaly confidence level is an indicator reflecting the possibility of abnormal situations output by the real-time anomaly detection unit; the device health index is used to measure the operating health status of each device in the unmanned tower, and is obtained by comprehensively evaluating the performance parameters of the device; the environmental risk level is a level division of potential risks to the operation of the unmanned tower determined according to factors such as meteorological conditions and the surrounding electromagnetic environment. Define the reward function as the entropy reduction amplitude of the state transition matrix after device regulation. Entropy is an indicator to measure the uncertainty of the system. The greater the entropy reduction amplitude of the state transition matrix, the more the uncertainty of the system is reduced after device regulation, and the better the regulation effect. Let the state transition matrix be , and its entropy is , and the calculation formula of the reward function is: Among them, represents the entropy of the state transition matrix before device regulation, It represents the entropy of the state transition matrix after device regulation. The parameters of the policy network are iteratively updated through the Proximal Policy Optimization (PPO) algorithm, and a device regulation instruction set is output. The PPO algorithm can quickly optimize the policy network while ensuring the stability of policy updates, enabling it to generate optimal device regulation instructions based on different state space information. During this process, the parameters of the policy network are continuously adjusted according to the information feedback by the reward function to improve the effect of device regulation and ensure the stable operation of the unmanned tower. Embodiment 5

[0047] This embodiment mainly elaborates on the execution steps of the knowledge graph reasoning unit and the detailed content of the generation of the device regulation instruction set. The knowledge graph reasoning unit provides an analysis method based on knowledge association for the unmanned tower operation state monitoring system, and the generation of the device regulation instruction set is directly related to the effective control of the unmanned tower devices in abnormal situations.

[0048] The knowledge graph reasoning unit is used to construct an association graph of unmanned tower devices based on a graph neural network and extract key node influencing factors. First, the communication link state parameters, navigation device operation parameters, and power device load parameters are mapped to graph nodes. For example, parameters such as the signal strength and bit error rate of the communication link are used as the attributes of the communication link-related nodes; parameters such as the gyroscope drift value and GPS positioning deviation of the navigation device are used as the attributes of the navigation device nodes; parameters such as the load fluctuation index of the power device are used as the attributes of the power device nodes. These nodes are connected to each other through their physical connection relationships or logical association relationships to form a device association graph.

[0049] Next, the dependence strength between nodes is calculated through the graph attention mechanism to generate key node influencing factors. The graph attention mechanism can automatically learn the importance weights between nodes and highlight the nodes that have a greater impact on the target node. By calculating the dependence strength between all nodes, key node influencing factors are obtained. The key node influencing factors can reflect which nodes have more important influence in the entire device association graph, providing an important basis for subsequent system analysis and optimization.

[0050] The key node influencing factors are input into the dynamic modeling unit to optimize the state transition probability graph. When constructing the state transition probability graph, the dynamic modeling unit considers the key node influencing factors and can more accurately reflect the transition relationship between the operation states of the unmanned tower. For example, if a certain communication link node is determined as a key node and its state change has a greater impact on the overall system state transition, then in the state transition probability graph, the edge weights related to this node will be adjusted accordingly according to its influencing factor, making the state transition probability graph more in line with the actual situation.

[0051] In terms of generating device control instruction sets, an emergency shutdown instruction is generated for high-priority abnormal events, and a redundant device switching protocol is triggered. When the abnormal confidence level exceeds the preset high-priority threshold and is determined to be a high-priority abnormal event through system comprehensive judgment, an emergency shutdown instruction is immediately generated to stop the operation of relevant devices to prevent the further expansion of faults. At the same time, the redundant device switching protocol is triggered to start standby devices to ensure the continuous operation of the key functions of the unmanned tower. For example, when a serious fault occurs in the navigation device, which may lead to a significant decrease in the navigation accuracy of the aircraft and endanger flight safety, the system will quickly issue an emergency shutdown instruction to stop the faulty navigation device and automatically switch to the standby navigation device to ensure the continuity of the navigation function.

[0052] Parameter dynamic compensation instructions are generated for medium- and low-priority abnormal events, including communication link power adjustment, navigation device calibration coefficient correction, and power load balanced distribution. When medium- and low-priority abnormal events occur, the parameters of relevant devices are adjusted to compensate for the impact of the abnormality. In terms of the communication link, if a decrease in signal strength is detected, according to the degree of decrease and communication requirements, the communication link power is adjusted according to the formula where is the adjusted power, is the power before adjustment, is the power adjustment amount calculated according to the abnormal situation. In this way, the signal strength is enhanced to ensure communication quality. For navigation devices, if a GPS positioning deviation is found, the calibration coefficient is corrected according to the deviation value and device characteristics. Suppose the calibration coefficient correction formula is where is the corrected calibration coefficient, is the calibration coefficient before correction, is the correction factor determined according to the navigation device type and accuracy requirements, is the GPS positioning deviation value, thereby improving the positioning accuracy of the navigation device. In terms of power equipment, when uneven power load is detected, a load balancing algorithm is used to redistribute the power load. For example, a certain proportion of the power of high-load devices is distributed to low-load devices to ensure the overall stable operation of the power equipment and avoid equipment failures caused by local overload. Through the generation and execution of these instructions for different-priority abnormal events, various abnormal situations during the operation of the unmanned tower can be effectively addressed, ensuring its stable and reliable operation.

[0053] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0054] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An unmanned tower operation status monitoring system based on big data, characterized in that Including: A data acquisition unit, which is used to collect the communication link status parameters, meteorological sensor monitoring parameters, navigation equipment operation parameters, power equipment load parameters and aircraft dynamic trajectory parameters of the unmanned tower in real time through a stream data processing framework; A multi-source data fusion unit, which is used to perform cross-domain feature extraction and heterogeneous data fusion on the communication link status parameters, meteorological sensor monitoring parameters, navigation equipment operation parameters, power equipment load parameters and aircraft dynamic trajectory parameters through a federated learning algorithm to generate a multi-modal joint feature matrix; A dynamic modeling unit, which is used to construct a state transition probability graph of the unmanned tower operation based on a dynamic Bayesian network with the multi-modal joint feature matrix as the input and output a real-time state evolution sequence; A real-time anomaly detection unit, which is used to calculate the temporal anomaly score of the real-time state evolution sequence through an isolation forest algorithm and an LSTM autoencoder to generate an anomaly confidence level; An active control instruction generation unit, which is used to generate a device regulation instruction set according to the anomaly confidence level in combination with a reinforcement learning policy network.

2. The system according to claim 1, wherein The execution steps of the real-time anomaly detection unit include: Dividing the real-time state evolution sequence into time window segments and extracting the statistical feature vectors within each time window; Calculating the reconstruction error of the statistical feature vectors through the LSTM autoencoder to obtain a first anomaly score; Calculating the path length deviation of the statistical feature vectors through the isolation forest algorithm to obtain a second anomaly score; Performing weighted fusion on the first anomaly score and the second anomaly score to generate the anomaly confidence level.

3. The system according to claim 2, wherein The execution steps of the data acquisition unit include: Receiving in real time the signal strength, bit error rate and link delay data in the communication link status parameters through the communication protocol interface of the Apache Kafka stream processing platform and the unmanned tower; Collecting the wind speed, visibility and pressure gradient data in the meteorological sensor monitoring parameters through distributed sensor nodes; Extracting the gyroscope drift value, GPS positioning deviation and radar scanning frequency data in the navigation equipment operation parameters through a device log parsing engine.

4. The system according to claim 3, wherein The execution steps of the multi-source data fusion unit further include: Constructing a federated learning global model, where each tower device acts as a local client and uploads the encrypted local feature gradients; Performing a secure weighted average on the local feature gradients through an aggregation server to update the parameters of the federated learning global model; Sending the updated parameters of the federated learning global model to each client and iteratively generating the multi-modal joint feature matrix.

5. The system according to claim 4, wherein The execution steps of the dynamic modeling unit include: Defining the node variables of the dynamic Bayesian network according to historical operation data, including communication link stability, meteorological interference level, navigation accuracy attenuation factor, power load fluctuation index and aircraft trajectory offset; Calculating the conditional probability distribution of the node variables through the Markov chain Monte Carlo method; Update the edge weights of the state transition probability graph based on real-time data streams to generate the real-time state evolution sequence.

6. The system according to claim 5, wherein The execution steps of the real-time anomaly detection unit further include: Set the hidden layer dimension of the LSTM autoencoder to 30% of the input feature dimension, and adopt a Dropout layer to prevent overfitting; Introduce an adaptive subtree splitting strategy in the Isolation Forest algorithm to dynamically adjust the anomaly segmentation threshold according to the data distribution.

7. The system according to claim 1, wherein The execution steps of the active control instruction generation unit include: Construct the state space of the reinforcement learning policy network, including the anomaly confidence level, device health index, and environmental risk level; Define the reward function as the entropy reduction amplitude of the state transition matrix after device regulation; Iteratively update the parameters of the reinforcement learning policy network through the Proximal Policy Optimization algorithm and output the device regulation instruction set.

8. The system according to claim 7, wherein The generation steps of the device regulation instruction set further include: Generate an emergency shutdown instruction for high-priority anomaly events and trigger the redundant device switching protocol; Generate parameter dynamic compensation instructions for medium- and low-priority anomaly events, including communication link power adjustment, navigation device calibration coefficient correction, and power load balancing allocation.

9. The system according to claim 1, wherein, Further include: A knowledge graph reasoning unit for constructing an unmanned tower device association graph based on a graph neural network and extracting key node influencing factors; Among them, the execution steps of the knowledge graph reasoning unit include: Map the communication link state parameters, navigation device operation parameters, and power device load parameters to graph nodes; Calculate the dependence strength between nodes through a graph attention mechanism to generate the key node influencing factors; Input the key node influencing factors into the dynamic modeling unit to optimize the state transition probability graph.

10. An electronic device, characterized in that, Include: A memory for storing the computer program of the big data-based unmanned tower operation state monitoring system according to any one of claims 1-9; A processor for executing the computer program to realize real-time monitoring and regulation of the unmanned tower operation state.

Images