A UAV abnormality analysis method and system

Abstract

The present invention provides a method and system for analyzing anomalies in unmanned aerial vehicles (UAVs), belonging to the field of data processing technology. The method comprises: receiving a monitoring data stream from a UAV sensor, performing data shunting processing on the monitoring data stream to generate at least one sub-data stream; determining an anomaly diagnosis result for any sub-data stream based on the data characteristics of any sub-data stream, and calling a corresponding machine learning model from a model library based on the anomaly diagnosis result; inputting the data characteristics of any sub-data stream into the machine learning model to obtain a model analysis result for any sub-data stream; and obtaining a comprehensive analysis result for the UAV based on the model analysis results of all sub-data streams. The present invention improves the accuracy and reliability of analysis results for large quantities of real-time monitoring data through a collaborative mechanism of data shunting processing, dynamic calling of multiple models, and intelligent result fusion.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for anomaly analysis of unmanned aerial vehicles (UAVs). Background Art

[0002] Against the backdrop of rapid development in the low-altitude economy, drones are increasingly being used in logistics delivery, agricultural and forestry inspection, and emergency rescue. However, with the increasing complexity of drone systems, anomaly monitoring and health management (PHM) are facing unprecedented challenges.

[0003] Currently, UAV PHM systems mainly use a single model to monitor and diagnose anomalies in specific components or sensors. However, when dealing with large volumes of heterogeneous data from multiple sources, a single model cannot adapt to complex operating conditions (such as compound faults and environmental abrupt changes). When faced with multi-dimensional coupled anomaly patterns, the accuracy and reliability of the analysis results are difficult to guarantee.

[0004] Therefore, improving the accuracy and reliability of analytical results has become an urgent technical problem to be solved. Summary of the Invention

[0005] This invention provides a method, system, electronic device, and storage medium for anomaly analysis of unmanned aerial vehicles (UAVs), which addresses the shortcomings of low accuracy and reliability of analysis results in existing technologies, thereby improving the accuracy and reliability of analysis results.

[0006] This invention provides a method for anomaly analysis of unmanned aerial vehicles (UAVs), comprising the following steps:

[0007] Receive monitoring data streams from UAV sensors and perform data splitting processing on the monitoring data streams to generate at least one sub-data stream;

[0008] Based on the data characteristics of any sub-data stream, an anomaly diagnosis result for that sub-data stream is determined, and a corresponding machine learning model is called from the model library based on the anomaly diagnosis result; the model library includes a fault diagnosis model and a lifetime prediction model, wherein the fault diagnosis model is built based on a Ghost network, or the fault diagnosis model is built based on a first autoencoder and a long short-term memory network, wherein the first autoencoder is composed of a Fourier analysis network, and the Fourier analysis network includes at least one FAN layer; the lifetime prediction model is built based on a second autoencoder, a long short-term memory network, and a fully connected layer;

[0009] The data features of any sub-data stream are input into the machine learning model to obtain the model analysis result of any sub-data stream;

[0010] Based on the model analysis results of all the sub-data streams, the comprehensive analysis results of the UAV are obtained.

[0011] According to the present invention, a method for anomaly analysis of unmanned aerial vehicles (UAVs) is provided, wherein the model analysis result includes a first diagnostic result; the step of inputting the data features of any sub-data stream into the machine learning model to obtain the model analysis result of the any sub-data stream specifically includes:

[0012] When the abnormal diagnosis result is a fault diagnosis type, the fault diagnosis model is called from the model library;

[0013] The fault diagnosis features are obtained by extracting features from the data features through the feature extraction network in the fault diagnosis model.

[0014] The fault diagnosis features are classified by the feature classification layer in the fault diagnosis model to obtain the first diagnosis result;

[0015] The feature extraction network includes a feature extraction module, a convolutional layer, and a fully connected layer connected in sequence; the feature extraction module includes multiple Ghost Bottleneck modules connected in sequence; the Ghost Bottleneck module includes an extension layer and a compression layer connected in sequence; the input of the extension layer and the output of the compression layer are added together to obtain the output of the Ghost Bottleneck module.

[0016] According to the present invention, a method for anomaly analysis of unmanned aerial vehicles (UAVs) is provided, wherein the model analysis result includes a second diagnostic result; the step of inputting the data features of any sub-data stream into the machine learning model to obtain the model analysis result of the any sub-data stream specifically includes:

[0017] When the abnormal diagnosis result is a fault diagnosis type, the fault diagnosis model is called from the model library;

[0018] Periodic feature extraction is performed on the data features through any one of the FAN layers in the first autoencoder;

[0019] By stacking all the FAN layers, the periodic feature is obtained;

[0020] The periodic features are sequentially modeled using the gating mechanism of the Long Short-Term Memory network to obtain the first external state at the current moment;

[0021] The first external state at the current moment is input into the fully connected layer to obtain the second diagnostic result.

[0022] According to the present invention, a method for anomaly analysis of unmanned aerial vehicles (UAVs) is provided, wherein the model analysis result includes a lifetime prediction result; the step of inputting the data features of any sub-data stream into the machine learning model to obtain the model analysis result of the any sub-data stream specifically includes:

[0023] When the abnormal diagnosis result is of the lifespan prediction type, the lifespan prediction model is called from the model library;

[0024] The data features are extracted by the encoder and decoder in the second autoencoder to obtain effective features.

[0025] The effective features are sequentially modeled using the gating mechanism of the Long Short-Term Memory network to obtain the second external state at the current moment;

[0026] The second external state at the current moment is input into the fully connected layer to obtain the lifetime prediction result.

[0027] According to the UAV anomaly analysis method provided by the present invention, the step of obtaining the comprehensive analysis result of the UAV based on the model analysis results of all the sub-data streams specifically includes:

[0028] When the abnormal diagnosis result is a fault diagnosis type, all the model analysis results are weighted and fused to obtain a comprehensive analysis result.

[0029] When the abnormal diagnosis result is of the life prediction type, all the model analysis results are directly output.

[0030] According to a method for anomaly analysis of unmanned aerial vehicles (UAVs) provided by the present invention, the step of receiving monitoring data streams from UAV sensors and performing data splitting processing on the monitoring data streams to generate at least one sub-data stream specifically includes:

[0031] The stream processing engine, configured based on a preset communication protocol, connects to the drone's sensor data source and receives sensor data streams in real time.

[0032] The sensor data stream is cleaned and its format is standardized to generate a standardized monitoring data stream.

[0033] The monitoring data stream is input into a distributed message queue for buffering and storage, and the monitoring data stream is segmented to generate at least one sub-data stream;

[0034] The distributed message queue is used to persistently store the received data.

[0035] According to the present invention, a method for anomaly analysis of unmanned aerial vehicles (UAVs) is provided, the method further includes:

[0036] The model analysis results are associated with the corresponding sub-data streams and pushed to at least two types of servers through different message topic channels.

[0037] The present invention also provides an unmanned aerial vehicle (UAV) anomaly analysis system, comprising the following modules:

[0038] The first processing module is used to receive the monitoring data stream from the UAV sensor and perform data splitting processing on the monitoring data stream to generate at least one sub-data stream;

[0039] The second processing module is used to determine the anomaly diagnosis result of any sub-data stream based on the data characteristics of any sub-data stream, and to call the corresponding machine learning model from the model library based on the anomaly diagnosis result.

[0040] The third processing module is used to input the data features of any sub-data stream into the machine learning model to obtain the model analysis result of any sub-data stream;

[0041] The fourth processing module is used to obtain the comprehensive analysis result of the UAV based on the model analysis results of all the sub-data streams.

[0042] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement any of the above-described methods for analyzing unmanned aerial vehicle anomalies.

[0043] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the UAV anomaly analysis method as described above.

[0044] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-described UAV anomaly analysis methods.

[0045] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0046] By receiving monitoring data streams from UAV sensors and performing data splitting, effective management and efficient processing of multi-source heterogeneous data are achieved. Based on the data characteristics of any sub-data stream, anomaly diagnosis results are determined. The corresponding machine learning model is then retrieved from the model library based on these results. The data features of any sub-data stream are input into the machine learning model to obtain the model analysis results for that sub-data stream. This achieves dimensionality reduction of high-dimensional data and extraction of key features. Furthermore, the most suitable model is determined based on the anomaly diagnosis results, enabling accurate analysis of data features. By analyzing the model results of all sub-data streams, a comprehensive analysis result for the UAV is obtained. All model analysis results are processed using either weighted fusion or direct output, achieving the fusion and efficient output of multi-model results, thereby improving the accuracy and reliability of the analysis results. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0048] Figure 1 This is one of the flowcharts of the UAV anomaly analysis method provided by the present invention.

[0049] Figure 2 This is the second flowchart of the UAV anomaly analysis method provided by the present invention.

[0050] Figure 3 This is the third flowchart of the UAV anomaly analysis method provided by the present invention.

[0051] Figure 4 This is the fourth flowchart of the UAV anomaly analysis method provided by the present invention.

[0052] Figure 5 This is the fifth flowchart of the UAV anomaly analysis method provided by the present invention.

[0053] Figure 6 This is the sixth flowchart of the UAV anomaly analysis method provided by the present invention.

[0054] Figure 7 This is a schematic diagram of the structure of the UAV anomaly analysis system provided by the present invention.

[0055] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0057] It should be noted that in the description of this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. The terms "upper," "lower," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0058] The terms "first," "second," etc., used in this invention are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0059] The following is combined Figures 1-8 This invention describes the UAV anomaly analysis method, system, electronic device, and storage medium provided by the present invention.

[0060] Figure 1 This is one of the flowcharts illustrating the UAV anomaly analysis method provided by the present invention, such as... Figure 1 As shown, including but not limited to the following steps:

[0061] Step 101: Receive the monitoring data stream from the UAV sensor and perform data splitting processing on the monitoring data stream to generate at least one sub-data stream.

[0062] In the operation of the UAV anomaly analysis system, step 101, as the core link of data access and preprocessing, realizes the reception and distribution of UAV sensor monitoring data streams in this embodiment. Its core lies in solving the technical problem that traditional PHM systems cannot effectively handle large-volume real-time data streams. The specific implementation process is as follows:

[0063] In one possible implementation, Figure 2 This is the second flowchart of the UAV anomaly analysis method provided by the present invention, as shown below. Figure 2 As shown, step 101 specifically includes steps 201-203:

[0064] Step 201: The stream processing engine, configured based on a preset communication protocol, connects to the UAV's sensor data source and receives sensor data streams in real time.

[0065] Step 202: Perform data cleaning and format standardization on the sensor data stream to generate a standardized monitoring data stream;

[0066] Step 203: Input the monitoring data stream into a distributed message queue for buffering and storage, and divide the monitoring data stream to generate at least one sub-data stream; wherein, the distributed message queue is used to persistently store the received data.

[0067] Specifically, firstly, a stream processing engine (such as Flink) is configured based on a preset communication protocol to establish a real-time connection with the drone sensor data source. This design solves the problem that traditional batch processing systems cannot meet real-time requirements. Through the real-time data retrieval mechanism of the stream processing engine, the timeliness of monitoring data is ensured. In actual deployment, the stream processing engine retrieves the raw data stream from the drone sensor data source at fixed time intervals (such as 100ms), effectively avoiding data backlog and delay.

[0068] Next, data cleaning and format standardization are performed on the received raw data stream. Because UAV sensor data is multi-source and heterogeneous (including controller status data, environmental monitoring data, equipment operation data, etc.), this embodiment employs a unified data format conversion rule to convert various types of sensor data into a standardized JSON format data stream. This process not only solves the problem of inconsistent formats among multi-source data but also ensures data quality through data verification mechanisms (such as range checks and null value handling), providing a reliable data foundation for subsequent processing.

[0069] The standardized monitoring data stream is then input into a distributed message queue system (such as Pulsar) for buffering and storage. This design solves the problem of data loss in traditional systems during network anomalies. Through the persistent storage mechanism of the message queue, even if a network interruption occurs, the data can be completely preserved and processed again after the network is restored. At the same time, the message queue system adopts a multi-partition storage strategy, distributing the data stream evenly across different partitions according to timestamps, significantly improving data throughput.

[0070] Finally, the monitoring data stream is segmented to generate at least one sub-data stream. Each sub-data stream is then input into a machine learning model of a subsystem for further processing, thereby meeting real-time requirements while ensuring the accuracy of data analysis.

[0071] Step 102: Based on the data characteristics of any sub-data stream, determine the anomaly diagnosis result of any sub-data stream, and call the corresponding machine learning model from the model library according to the anomaly diagnosis result; the model library includes a fault diagnosis model and a lifetime prediction model. The fault diagnosis model is built based on the Ghost network, or the fault diagnosis model is built based on the first autoencoder and the long short-term memory network. The first autoencoder is composed of a Fourier analysis network, and the Fourier analysis network includes at least one FAN layer; the lifetime prediction model is built based on the second autoencoder, the long short-term memory network, and the fully connected layer.

[0072] Step 102, as a key step in data processing and anomaly identification, involves multi-dimensional analysis of the UAV sensor sub-data stream based on feature extraction methods and anomaly mode switching mechanisms.

[0073] In practice, the sub-data stream generated in step 101 is first preprocessed: fixed-length data segments are extracted, and the data within the window is normalized to eliminate dimensional differences. Subsequently, multi-scale feature extraction is performed, including time-domain statistical features (maximum, minimum, mean, and standard deviation) and frequency-domain features (extracting the first 5 dominant frequency amplitudes based on fast Fourier transform).

[0074] During the anomaly pattern recognition phase, the system inputs the extracted feature vectors into a preset threshold judgment module: if the peak factor of three consecutive windows in the time domain features exceeds the safety threshold (e.g., the threshold for the accelerometer is set to 4.2), the anomaly diagnosis result is set to the fault diagnosis type; otherwise, the system maintains the normal monitoring state. For the lifespan prediction task, when the time domain mean of the sensor data shows a monotonically decreasing trend and the deceleration rate exceeds the historical baseline (e.g., the daily average voltage drop rate of the lithium battery > 0.5%), the anomaly diagnosis result is set to the lifespan prediction type.

[0075] Furthermore, based on the identified abnormal diagnostic results, fault diagnosis or life prediction models in the model library are dynamically invoked to address the problem of insufficient adaptability of a single model in existing technologies.

[0076] When a fault diagnosis type is detected, the system provides two optional model architecture implementation methods:

[0077] The first approach is based on the Ghost network architecture, achieving efficient feature extraction through cascaded Ghost Bottleneck modules. The Ghost network (also known as the phantom network) can generate more feature maps through simple linear transformations, reducing computational cost while generating sufficiently flexible feature representations to meet the needs of different tasks. Each module's expansion layer uses a 1×1 convolution to expand the input channels by a factor of 3, followed by a 3×3 depthwise separable convolution to extract spatial features. Finally, a compression layer restores the channel dimensions and connects to the input residuals. This design achieved 94.2% accuracy in motor bearing fault classification, with an inference speed 2.3 times faster than ResNet-50.

[0078] The second implementation adopts a hybrid architecture of Fourier analysis network (FAN layer) and long short-term memory network (LSTM). The FAN layer extracts frequency domain features through a configurable Fourier fundamental frequency filter bank, then models the temporal dependency relationship through the gating mechanism of LSTM, and finally outputs the diagnostic results through a fully connected layer.

[0079] For lifetime prediction, the system calls a dedicated model consisting of a stacked autoencoder and an LSTM. The autoencoder extracts low-dimensional degradation features from the sensor data through the encoder, and the decoder reconstructs and concatenates these features with the temporal state vector of the LSTM. The remaining lifetime (RUL) is then calculated by regression through a fully connected layer.

[0080] Step 103: Input the data features of any sub-data stream into the machine learning model to obtain the model analysis results of any sub-data stream.

[0081] Specifically, in this application, the machine learning model can be a fault diagnosis model or a life prediction model, where the fault diagnosis model includes two model architectures. Therefore, the model analysis results output by these three different architectures are the first diagnostic result, the second diagnostic result, and the life prediction result, respectively. The following embodiments will provide a detailed description of the specific implementation of step 103.

[0082] In one possible implementation, Figure 3 This is the third flowchart of the UAV anomaly analysis method provided by the present invention, as shown below. Figure 3 As shown, step 103 includes steps 301-303:

[0083] Step 301: When the abnormal diagnosis result is a fault diagnosis type, call the fault diagnosis model from the model library.

[0084] Step 302: Extract features from the data using the feature extraction network in the fault diagnosis model to obtain fault diagnosis features; wherein, the feature extraction network includes a feature extraction module, a convolutional layer and a fully connected layer connected in sequence; the feature extraction module includes multiple Ghost Bottleneck modules connected in sequence; the Ghost Bottleneck module includes an extension layer and a compression layer connected in sequence; the input of the extension layer and the output of the compression layer are added together to obtain the output of the Ghost Bottleneck module.

[0085] Step 303: Through the feature classification layer in the fault diagnosis model, the fault diagnosis features are classified to obtain the first diagnosis result.

[0086] Specifically, when an anomaly diagnosis result is identified as a fault diagnosis type, the system calls the fault diagnosis type model from the model library. This design addresses the stringent real-time and computational resource requirements of UAV PHM systems, employing a lightweight model architecture based on Ghost networks. To solve the problems of large parameter count and high inference latency inherent in traditional convolutional neural networks (CNNs) deployed on edge computing devices, steps 301-303 achieve efficient computation through a feature extraction network containing multiple Ghost Bottleneck modules. Each Ghost Bottleneck module includes sequentially connected extension and compression layers.

[0087] Both the extension and compression layers can be built using the Ghost module. The Ghost module can generate phantom features, improving network computational efficiency. For input data features... c, k, and q represent the dimension, length, and width of the input data, respectively. The Ghost module first uses... The convolution kernel for the input data Performing a normal convolution operation, the output feature map is , and To output the length and width of the feature, To determine the dimension of the output features, the width and height of the convolution kernel are both [values ​​missing]. Then, the output feature map... A series of phantom features are generated by performing a linear transformation according to the following formula. As shown below:

[0088]

[0089]

[0090] in, For the output phantom features The data in To output feature map The first in dimension, Represents a linear transformation function. This indicates the types of linear transformations, that is, the existing ones. Each of the feature maps is sequentially transformed by s linear transformations to obtain s phantom features, resulting in a final feature dimension of . Finally, and The output of the Ghost module is obtained after splicing.

[0091] In the Ghost Bottleneck module, the first Ghost module that the input data passes through is an expansion layer, used to increase the number of feature channels. The ratio of the number of input channels to the number of output channels is called the expansion ratio. The second Ghost module is used to reduce the number of feature channels, making the number of output features consistent with the number of input features; this is called the compression layer. Finally, the input and output are added together using a shortcut connection. Batch Normalization and ReLU are used for batch normalization and activation functions, respectively. In other words, the output of the Ghost Bottleneck module is the sum of the input of the expansion layer and the output of the compression layer.

[0092] The fault diagnosis model includes a feature extraction network and a feature classification layer. The feature extraction network consists of a feature extraction module, a convolutional layer, and a fully connected layer connected in sequence. The feature extraction module includes multiple Ghost Bottleneck modules connected in sequence. For example, the feature extraction module could include six Ghost Bottleneck modules. The size of the convolutional layer can be... (Input channels, output channels, convolution window length, convolution window width). The feature classification layer can be a softmax classifier.

[0093] The data features are input into the feature extraction network of the fault diagnosis model to obtain the fault diagnosis features output by the feature extraction network; the fault diagnosis features are deep features. These fault diagnosis features are then input into the feature classification layer of the fault diagnosis model to obtain the first diagnostic result output by the feature classification layer.

[0094] In another possible implementation, Figure 4 This is the fourth flowchart of the UAV anomaly analysis method provided by the present invention, as shown below. Figure 4 As shown, step 103 includes steps 401-405:

[0095] Step 401: When the abnormal diagnosis result is a fault diagnosis type, call the fault diagnosis model from the model library.

[0096] Step 402: Perform periodic feature extraction on the data features through any FAN layer in the first autoencoder.

[0097] Step 403: Stack all FAN layers to obtain periodic features.

[0098] Step 404: Use the gating mechanism of the Long Short-Term Memory network to perform sequence modeling of the periodic features and obtain the first external state at the current moment.

[0099] Step 405: Input the first external state at the current moment into the fully connected layer to obtain the second diagnostic result.

[0100] Steps 401-405 serve as an alternative technical solution for fault diagnosis. When the abnormal diagnosis result is identified as a fault diagnosis type, the system calls the fault diagnosis type model from the model library. This design addresses scenarios where UAV sensor data exhibits significant periodic characteristics but suffers from severe temporal noise interference (such as propeller imbalance faults), thus proposing a hybrid architecture based on Fourier analysis networks (FAN layers) and long short-term memory networks (LSTM).

[0101] Each layer of a Fourier Analysis Network (FAN) can effectively capture periodic failure patterns in data features by extracting features from the data features.

[0102] For any FAN layer, it can be represented by the following formula:

[0103] ;

[0104] In the formula, L is the number of FAN layers. Let x be the l-th FAN layer, and x be the data feature. Let be the first learnable weight matrix of the l-th FAN layer. For the bias term of the l-th FAN layer, Let be the second learnable weight matrix of the l-th FAN layer. Here, || represents the activation function, and || represents concatenation along the first feature dimension. The learnable weight matrix for the last FAN layer. This is the bias vector for the last FAN layer.

[0105] Through the cosine term Sine term The FAN layer simulates the basis functions of a Fourier series, directly capturing the periodic features in the data; through nonlinear terms... The FAN layer enhances the model's non-linear expressive power. For the last FAN layer, its output features are transformed linearly. The calculation simplifies the output calculation process. By stacking multiple FAN layers, multi-level periodic features in the data can be extracted step by step, providing high-quality feature input for subsequent autoencoders and long short-term memory networks.

[0106] For each FAN layer, it is calculated using cosine, sine, and nonlinear terms, respectively simulating the basis functions of the Fourier series and enhancing the nonlinear expressive power. Stacking all FAN layers allows for the synthesis of multi-level periodic features extracted from each layer, generating periodic features rich in information. This stacking operation not only preserves the local periodic patterns extracted by each FAN layer but also enhances the expressive power of global periodic features.

[0107] By stacking all FAN layers, high-quality feature input can be provided for subsequent autoencoders, thereby improving the accuracy and robustness of periodic features. This method significantly improves the performance of periodic feature extraction through multi-level feature extraction.

[0108] Furthermore, the autoencoder performs dimensionality reduction by stacking all FAN layers, which generates periodic feature representations, thereby removing redundant information and retaining key features.

[0109] Furthermore, the first external state at the current moment can be calculated using the following formula:

[0110] ;

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] ;

[0116] In the formula, Output the forget gate for the current moment. The input gate output is the current time. The output of the current output gate. This represents the candidate state at the current moment. The internal state at the current moment. This is the first external state at the current moment. The periodicity of the current moment. This is the first external state at the previous moment. This refers to the internal state at the previous moment. This is the first weight matrix of the forget gate. This is the first weight matrix of the input gate. This is the first weight matrix of the output gate. The first weight matrix for the candidate states. This is the second weight matrix of the forget gate. This is the second weight matrix of the input gate. This is the second weight matrix of the output gate. This is the second weight matrix for the candidate states. Let be the bias vector of the forget gate. This is the bias vector for the input gate. This is the bias vector for the output gate. This is the bias vector for the candidate state. For the Logistic function, This indicates element-wise multiplication.

[0117] The gating mechanism of Long Short-Term Memory (LSTM) networks includes a forget gate, an input gate, and an output gate. First, the forget gate is based on the periodic characteristics of the current input. and the first external state at the previous moment ,calculate Its range is [0,1], and it is used to quantify the historical internal state. The percentage of information retained.

[0118] Next, input gate calculation Simultaneously generate candidate states Subsequently, through Update internal status .

[0119] Finally, output gate calculation and based on Generate the first external state .

[0120] Next, through the fully connected layer... A second diagnostic result is obtained by performing nonlinear transformation and probability mapping, as detailed below:

[0121] First, output the first external state at the current time from the LSTM. (Dimension 64) Input is fed into the fully connected layer, through the weight matrix and bias vector Perform linear projection and calculate the logic value. .

[0122] Subsequently, a Softmax function is applied to the logic value z to generate a probability distribution of the fault categories. ,in , i The values ​​∈{0,1,…,9} correspond to 10 fault types, thus yielding the second diagnostic result. This step transforms the nonlinear combination of high-dimensional features into a probability output through a normalized exponential function, enabling the model to quantify the confidence level of different fault types.

[0123] In another possible implementation, Figure 5 This is the fifth flowchart of the UAV anomaly analysis method provided by the present invention, as shown below. Figure 5 As shown, step 103 includes steps 501-504:

[0124] Step 501: When the abnormal diagnosis result is of the life prediction type, call the life prediction model from the model library.

[0125] Step 502: Perform feature extraction on the data features through the encoder and decoder in the second autoencoder to obtain effective features.

[0126] Step 503: Use the gating mechanism of the Long Short-Term Memory network to perform sequence modeling on the effective features to obtain the second external state at the current time.

[0127] Step 504: Input the second external state at the current moment into the fully connected layer to obtain the lifetime prediction result.

[0128] When an abnormal diagnostic result is identified as a lifespan prediction type, the system calls a lifespan prediction model from the model library. This design solves the problem that traditional single models cannot adapt to complex working conditions. Through a dynamic model calling mechanism, it ensures that the lifespan prediction task can select the optimal model according to actual needs. In actual deployment, the lifespan prediction model adopts a three-level architecture of "second autoencoder + long short-term memory network + fully connected layer". The second autoencoder is responsible for feature dimensionality reduction, the long short-term memory network is responsible for temporal modeling, and the fully connected layer is responsible for regression prediction. This layered design significantly improves the model's computational efficiency and prediction accuracy.

[0129] In the feature extraction stage, the input data features are compressed and represented by the encoder in the second autoencoder. Specifically, the encoder employs a three-layer fully connected network structure. The first layer compresses the original feature dimension from 256 to 128, the second layer further compresses it to 64, and the third layer outputs a 32-dimensional feature representation. This dimensionality reduction process solves the computational complexity problem caused by high-dimensional data, while preserving key feature information through nonlinear transformation. The encoder is trained unsupervised, ensuring the reliability of the feature representation by minimizing the mean square error between the input data and the decoder's reconstructed data.

[0130] Next, the compressed features are reconstructed using the decoder in the second autoencoder. The decoder employs a network structure symmetrical to the encoder, progressively restoring the 32-dimensional features to their original dimensions. This reconstruction process not only verifies the effectiveness of feature extraction but also provides additional regularization constraints for the model, preventing overfitting.

[0131] Furthermore, the specific implementation methods of steps 503 and 504 are the same as those of steps 404 and 405, so they will not be elaborated on here.

[0132] Step 104: Based on the model analysis results of all sub-data streams, obtain the comprehensive analysis results of the UAV.

[0133] This embodiment realizes the fusion processing and multi-terminal distribution of model analysis results from a UAV PHM system. Its core lies in solving the technical challenge of traditional PHM systems being unable to dynamically adjust the result processing method based on anomaly diagnosis results. The specific implementation process is as follows:

[0134] In one possible implementation, Figure 6 This is the sixth flowchart of the UAV anomaly analysis method provided by the present invention, as shown below. Figure 6 As shown, step 104 specifically includes steps 601-602:

[0135] Step 601: When the abnormal diagnosis result is a fault diagnosis type, perform weighted fusion processing on all model analysis results to obtain a comprehensive analysis result.

[0136] Step 602: When the abnormal diagnosis result is of the life prediction type, output all model analysis results directly.

[0137] Specifically, for the multi-model calculation results of fault diagnosis types, this application adopts a fusion strategy based on dynamic adjustment of weights according to confidence level to solve the problem of insufficient adaptability of traditional fixed weight fusion mechanism to real-time model performance.

[0138] In practice, when an abnormal diagnosis result is identified as a fault diagnosis type, the system first obtains the initial diagnosis results output by the fault diagnosis models of each subsystem and their corresponding confidence indices: for fault diagnosis models based on Ghost networks, the confidence is calculated using the probability distribution entropy value output by Softmax; while for fault diagnosis models based on a hybrid model of FAN-LSTM, the prediction variance is calculated using Monte Carlo Dropout sampling as the confidence. The system then dynamically generates weighting coefficients based on the confidence.

[0139] For example, in the scenario of fault diagnosis of the power system of a logistics drone, when the confidence levels of the fault diagnosis models of the four subsystems are 0.92 (fault diagnosis model based on Ghost network), 0.90 (fault diagnosis model based on Ghost network), 0.86 (fault diagnosis model based on FAN-LSTM hybrid model), and 0.85 (fault diagnosis model based on FAN-LSTM hybrid model), the weight allocation can be 0.3, 0.28, 0.22, and 0.2, respectively. The final comprehensive analysis result is as follows: y =0.3× y 1+0.28× y 2+0.22× y 3+0.2× y 4, of which, y 1. y 2. y 3. y Figure 4 shows the initial diagnostic results of the fault diagnosis models for the four subsystems. For cases where the confidence difference is less than a threshold, the system can switch to an average weighted mode to ensure decision robustness.

[0140] This dynamic weight allocation strategy takes into account both the reliability of the model and the importance of the subsystems, ultimately generating comprehensive analysis results.

[0141] When an anomaly diagnosis result is identified as a lifetime prediction type, the system directly outputs the prediction result of the lifetime prediction model. This design simplifies the processing flow of lifetime prediction tasks, avoids unnecessary fusion calculations, and improves system efficiency. In actual deployment, the lifetime prediction result is output in the form of remaining useful life (RUL) with a confidence interval (e.g., "Remaining useful life: 120 hours, confidence level: 95%)", providing users with a more intuitive basis for decision-making.

[0142] In one possible implementation, after step 104, the method further includes the following steps:

[0143] The model analysis results are associated with the corresponding sub-data streams and pushed to at least two types of servers through different message topic channels.

[0144] This embodiment realizes the association and multi-terminal distribution of PHM system result data from a UAV, and its core lies in solving the technical problem that traditional PHM systems cannot effectively manage the association of multi-source data and the reliability of distribution. The specific implementation process is as follows:

[0145] The system associates the model analysis results with corresponding sub-data streams; specifically, it associates the first diagnostic prediction result, the second diagnostic prediction result, and the lifetime prediction result with their respective sub-data streams. These are then pushed to at least two types of servers through different message topic channels. The server types include at least ground user servers and UAV servers. The system first performs spatiotemporal alignment of the first diagnostic prediction result, the second diagnostic prediction result, and the lifetime prediction result with their corresponding sub-data streams to ensure that each calculation result can be traced back to its corresponding original monitoring data. In actual implementation, the system adds a unified timestamp (accurate to milliseconds) and a subsystem identifier to each calculation result and sub-data stream to ensure data consistency in time and space.

[0146] Next, the system distributes the associated data to at least two types of servers through different message topic channels. This design solves the reliability problem of traditional single-point distribution systems. Through a multi-channel distribution mechanism, it ensures that even if one server fails, other servers can still receive data normally. In specific implementation, the system establishes multiple independent message topic channels, including a first message topic channel for real-time monitoring, a second message topic channel for decision analysis, and a third message topic channel as a backup. The real-time monitoring server receives the raw monitoring data stream and calculation results through the first message topic channel, supporting real-time visualization; the decision analysis server receives calculation results and diagnostic reports through the second message topic channel, supporting offline data analysis; and the backup channel temporarily takes over the data distribution task in case of network anomalies, ensuring the continuity and reliability of data distribution.

[0147] Reference Figure 7 , Figure 7 This is a schematic diagram of the UAV anomaly analysis system provided by the present invention. The system includes:

[0148] The first processing module is used to receive the monitoring data stream from the UAV sensor and perform data splitting processing on the monitoring data stream to generate at least one sub-data stream;

[0149] The second processing module is used to determine the anomaly diagnosis result of any sub-data stream based on its data characteristics, and to call the corresponding machine learning model from the model library based on the anomaly diagnosis result. The model library includes a fault diagnosis model and a lifetime prediction model. The fault diagnosis model is built based on a Ghost network, or it is built based on a first autoencoder and a long short-term memory network. The first autoencoder is composed of a Fourier analysis network, which includes at least one FAN layer. The lifetime prediction model is built based on a second autoencoder, a long short-term memory network, and a fully connected layer.

[0150] The third processing module is used to input the data features of any sub-data stream into the machine learning model to obtain the model analysis results of any sub-data stream;

[0151] The fourth processing module is used to obtain the comprehensive analysis results of the UAV based on the model analysis results of all sub-data streams.

[0152] In one possible implementation, the third processing module is further configured to:

[0153] When the abnormal diagnosis result is a fault diagnosis type, the fault diagnosis model is called from the model library;

[0154] When the abnormal diagnosis result is a fault diagnosis type, the fault diagnosis model is called from the model library;

[0155] By using the feature extraction network in the fault diagnosis model, feature extraction is performed on the data features to obtain fault diagnosis features;

[0156] The fault diagnosis features are classified by the feature classification layer in the fault diagnosis model to obtain the first diagnosis result.

[0157] The feature extraction network includes a feature extraction module, a convolutional layer, and a fully connected layer connected in sequence. The feature extraction module includes multiple Ghost Bottleneck modules connected in sequence. The Ghost Bottleneck module includes an extension layer and a compression layer connected in sequence. The input of the extension layer and the output of the compression layer are added together to get the output of the Ghost Bottleneck module.

[0158] In one possible implementation, the third processing module is further configured to:

[0159] When the abnormal diagnosis result is a fault diagnosis type, the fault diagnosis model is called from the model library;

[0160] Periodic feature extraction is performed on the data features through any FAN layer in the first autoencoder;

[0161] Stacking all FAN layers yields periodic features;

[0162] The first external state at the current moment is obtained by using the gating mechanism of the Long Short-Term Memory network to model the periodic features sequentially.

[0163] The first external state at the current moment is input into the fully connected layer to obtain the second diagnostic result.

[0164] In one possible implementation, the third processing module is further configured to:

[0165] When the abnormal diagnosis result is of the life prediction type, call the life prediction model from the model library;

[0166] The encoder and decoder in the second autoencoder perform feature extraction operations on the data features to obtain effective features;

[0167] By using the gating mechanism of a long short-term memory network to perform sequence modeling of effective features, the second external state at the current time is obtained;

[0168] The second external state at the current moment is input into the fully connected layer to obtain the lifetime prediction result.

[0169] In one possible implementation, the fourth processing module is further configured to:

[0170] When the abnormal diagnosis result is a fault diagnosis type, all model analysis results are weighted and fused to obtain a comprehensive analysis result.

[0171] When the abnormal diagnosis result is of the life prediction type, all model analysis results are directly output.

[0172] In one possible implementation, the first processing module is further configured to:

[0173] The stream processing engine, configured based on a preset communication protocol, connects to the drone's sensor data source and receives sensor data streams in real time.

[0174] Perform data cleaning and format standardization on the sensor data stream to generate a standardized monitoring data stream;

[0175] The monitoring data stream is input into a distributed message queue for buffering and storage, and the monitoring data stream is segmented to generate at least one sub-data stream;

[0176] Among them, the distributed message queue is used to persistently store the received data.

[0177] In one possible implementation, the system further includes a fifth processing module, which associates the model analysis results with the corresponding sub-data streams and pushes them to at least two types of servers through different message topic channels.

[0178] It should be noted that the UAV anomaly analysis system provided by the present invention can execute the UAV anomaly analysis method of any of the above embodiments during specific operation, which will not be elaborated in this embodiment.

[0179] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 8As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a UAV anomaly analysis method. This method includes: receiving monitoring data streams from UAV sensors and performing data splitting processing on the monitoring data streams to generate at least one sub-data stream; determining the anomaly diagnosis result of any sub-data stream based on its data characteristics, and calling the corresponding machine learning model from a model library based on the anomaly diagnosis result; the model library includes a fault diagnosis model and a lifetime prediction model. The fault diagnosis model is built based on a Ghost network, or it is built based on a first autoencoder and a long short-term memory network. The first autoencoder consists of a Fourier analysis network, which includes at least one FAN layer. The lifetime prediction model is built based on a second autoencoder, a long short-term memory network, and a fully connected layer; inputting the data characteristics of any sub-data stream into the machine learning model to obtain the model analysis result of any sub-data stream; and obtaining a comprehensive analysis result of the UAV based on the model analysis results of all sub-data streams.

[0180] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0181] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by the computer, the computer can execute the UAV anomaly analysis method provided in the above embodiments. The method includes: receiving monitoring data streams from UAV sensors and performing data splitting processing on the monitoring data streams to generate at least one sub-data stream; determining the anomaly diagnosis result of any sub-data stream based on the data characteristics of any sub-data stream, and calling the corresponding machine learning model from the model library based on the anomaly diagnosis result; the model library includes a fault diagnosis model and a lifetime prediction model. The fault diagnosis model is built based on a Ghost network, or the fault diagnosis model is built based on a first autoencoder and a long short-term memory network. The first autoencoder is composed of a Fourier analysis network, and the Fourier analysis network includes at least one FAN layer; the lifetime prediction model is built based on a second autoencoder, a long short-term memory network, and a fully connected layer; inputting the data characteristics of any sub-data stream into the machine learning model to obtain the model analysis result of any sub-data stream; and obtaining the comprehensive analysis result of the UAV based on the model analysis results of all sub-data streams.

[0182] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the UAV anomaly analysis method provided in the above embodiments. The method includes: receiving monitoring data streams from UAV sensors and performing data splitting processing on the monitoring data streams to generate at least one sub-data stream; determining anomaly diagnosis results for any sub-data stream based on data characteristics of any sub-data stream, and calling a corresponding machine learning model from a model library based on the anomaly diagnosis results; the model library includes a fault diagnosis model and a lifetime prediction model. The fault diagnosis model is built based on a Ghost network, or it is built based on a first autoencoder and a long short-term memory network. The first autoencoder is composed of a Fourier analysis network, which includes at least one FAN layer; the lifetime prediction model is built based on a second autoencoder, a long short-term memory network, and a fully connected layer; inputting data characteristics of any sub-data stream into the machine learning model to obtain the model analysis results of any sub-data stream; and obtaining a comprehensive analysis result of the UAV based on the model analysis results of all sub-data streams.

[0183] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0184] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for anomaly analysis of unmanned aerial vehicles (UAVs), characterized in that, include: The system receives monitoring data streams from UAV sensors and performs data splitting processing on the monitoring data streams to generate at least one sub-data stream; the monitoring data streams include controller status data, environmental monitoring data, and equipment operation data; Based on the data characteristics of any sub-data stream, an anomaly diagnosis result for that sub-data stream is determined, and a corresponding machine learning model is called from the model library based on the anomaly diagnosis result; the model library includes a fault diagnosis model and a lifetime prediction model, wherein the fault diagnosis model is built based on a Ghost network, or the fault diagnosis model is built based on a first autoencoder and a long short-term memory network, wherein the first autoencoder is composed of a Fourier analysis network, and the Fourier analysis network includes at least one FAN layer; the lifetime prediction model is built based on a second autoencoder, a long short-term memory network, and a fully connected layer; The data features of any sub-data stream are input into the machine learning model to obtain the model analysis result of any sub-data stream; Based on the model analysis results of all the sub-data streams, the comprehensive analysis results of the UAV are obtained; The comprehensive analysis result of the UAV obtained based on the model analysis results of all the sub-data streams specifically includes: When the abnormal diagnosis result is a fault diagnosis type, all the model analysis results are weighted and fused to obtain a comprehensive analysis result. When the abnormal diagnosis result is of the life prediction type, all the model analysis results are directly output.

2. The UAV anomaly analysis method according to claim 1, characterized in that, The model analysis results include a first diagnostic result; the step of inputting the data features of any sub-data stream into the machine learning model to obtain the model analysis results of any sub-data stream specifically includes: When the abnormal diagnosis result is a fault diagnosis type, the fault diagnosis model is called from the model library; The fault diagnosis features are obtained by extracting features from the data features through the feature extraction network in the fault diagnosis model. The fault diagnosis features are classified by the feature classification layer in the fault diagnosis model to obtain the first diagnosis result; The feature extraction network includes a feature extraction module, a convolutional layer, and a fully connected layer connected in sequence; the feature extraction module includes multiple Ghost Bottleneck modules connected in sequence; the Ghost Bottleneck module includes an extension layer and a compression layer connected in sequence; the input of the extension layer and the output of the compression layer are added together to obtain the output of the Ghost Bottleneck module.

3. The UAV anomaly analysis method according to claim 1, characterized in that, The model analysis results include a second diagnostic result; the step of inputting the data features of any sub-data stream into the machine learning model to obtain the model analysis results of any sub-data stream specifically includes: When the abnormal diagnosis result is a fault diagnosis type, the fault diagnosis model is called from the model library; Periodic feature extraction is performed on the data features through any one of the FAN layers in the first autoencoder; By stacking all the FAN layers, the periodic feature is obtained; The periodic features are sequentially modeled using the gating mechanism of the Long Short-Term Memory network to obtain the first external state at the current moment; The first external state at the current moment is input into the fully connected layer to obtain the second diagnostic result.

4. The UAV anomaly analysis method according to claim 1, characterized in that, The model analysis results include lifetime prediction results; the step of inputting the data features of any sub-data stream into the machine learning model to obtain the model analysis results of any sub-data stream specifically includes: When the abnormal diagnosis result is of the lifespan prediction type, the lifespan prediction model is called from the model library; The data features are extracted by the encoder and decoder in the second autoencoder to obtain effective features. The effective features are sequentially modeled using the gating mechanism of the Long Short-Term Memory network to obtain the second external state at the current moment; The second external state at the current moment is input into the fully connected layer to obtain the lifetime prediction result.

5. The UAV anomaly analysis method according to claim 1, characterized in that, The process of receiving monitoring data streams from UAV sensors and performing data splitting processing on these data streams to generate at least one sub-data stream specifically includes: The stream processing engine, configured based on a preset communication protocol, connects to the drone's sensor data source and receives sensor data streams in real time. The sensor data stream is cleaned and its format is standardized to generate a standardized monitoring data stream. The monitoring data stream is input into a distributed message queue for buffering and storage, and the monitoring data stream is segmented to generate at least one sub-data stream; The distributed message queue is used to persistently store the received data.

6. The UAV anomaly analysis method according to claim 1, characterized in that, The method further includes: The model analysis results are associated with the corresponding sub-data streams and pushed to at least two types of servers through different message topic channels.

7. An anomaly analysis system for unmanned aerial vehicles (UAVs), characterized in that, include: The first processing module is used to receive the monitoring data stream from the UAV sensor and perform data splitting processing on the monitoring data stream to generate at least one sub-data stream; The monitoring data stream includes controller status data, environmental monitoring data, and equipment operation data; The second processing module is used to determine the anomaly diagnosis result of any sub-data stream based on its data characteristics, and to call the corresponding machine learning model from the model library based on the anomaly diagnosis result. The model library includes a fault diagnosis model and a lifetime prediction model. The fault diagnosis model is built based on a Ghost network, or it is built based on a first autoencoder and a long short-term memory network. The first autoencoder is composed of a Fourier analysis network, and the Fourier analysis network includes at least one FAN layer. The lifetime prediction model is built based on a second autoencoder, a long short-term memory network, and a fully connected layer. The third processing module is used to input the data features of any sub-data stream into the machine learning model to obtain the model analysis result of any sub-data stream; The fourth processing module is used to obtain the comprehensive analysis result of the UAV based on the model analysis results of all the sub-data streams. Specifically, it includes: when the anomaly diagnosis result is a fault diagnosis type, performing weighted fusion processing on all the model analysis results to obtain the comprehensive analysis result. When the abnormal diagnosis result is of the life prediction type, all the model analysis results are directly output.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the UAV anomaly analysis method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the UAV anomaly analysis method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the UAV anomaly analysis method as described in any one of claims 1 to 6.

Images