CN115510950A - Aircraft telemetry data anomaly detection method and system based on time convolution network - Google Patents
Aircraft telemetry data anomaly detection method and system based on time convolution network Download PDFInfo
- Publication number
- CN115510950A CN115510950A CN202211035316.5A CN202211035316A CN115510950A CN 115510950 A CN115510950 A CN 115510950A CN 202211035316 A CN202211035316 A CN 202211035316A CN 115510950 A CN115510950 A CN 115510950A
- Authority
- CN
- China
- Prior art keywords
- data
- convolution
- network
- aircraft
- anomaly detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention provides an aircraft telemetering data anomaly detection method and system based on a time convolution network, which comprises the following steps: carrying out data preprocessing on the telemetering data; inputting the preprocessed telemetering data into an improved time convolution network, and capturing causality among sequences in a causal convolution and expansion convolution mode; inserting the weight with the value of 0 into the convolution kernel to form a hole convolution kernel as a basic calculation unit of the improved time convolution network; carrying out neural network training based on a basic computing unit, introducing a residual connection mode in the training process, replacing a set convolution layer with a residual block, and repeating the process until the construction of a data model is completed; and inputting the test data into the trained data model, and judging whether the data is abnormal or not through a loss function labeled by the data. By applying the technical scheme of the invention, the technical problem that the anomaly detection of a large amount of high-dimensional telemetering data is difficult in the anomaly detection process in the prior art is solved.
Description
Technical Field
The invention relates to the technical field of intersection of engineering application and information science, in particular to a method and a system for detecting the anomaly of telemetering data of an aircraft based on a time convolution network.
Background
Aircraft faults mainly have the following characteristics:
1) And (4) fault multifarities. In the spacecraft test, faults occur frequently during ground test, and opportunities are given to a machine learning training data set. Because the model requires more data sets to train to obtain a more ideal algorithm outlet. Aiming at the problems that training data are not representative and data with poor quality are collected, a screening evaluation and feature extraction method at a source is adopted for processing. The processing method of screening at the source is that atypical factors are artificially excluded from a database and are not used as a training data set; the feature extraction method is to select the most useful features from a large number of existing features for training, or to fit and reduce dimensions of the features, and the commonly used dimension reduction method is a Principal Component Analysis (PCA), but which dimension reduction method is specifically selected needs to be determined according to actual conditions.
2) Multiple fault concurrency. The aerospace craft system is a giant system which relates to multiple persons, multiple machines and multiple environments and has a very complex structure, and can be divided into a carrying system, a launching system, a measurement and control system and the like from hardware; the functions of the system can be divided into a power system, a communication system, a remote measuring system, a remote control system, various functional systems and the like, and each functional system is composed of a plurality of subsystems. The complexity of the composition therefore determines the multiplicity of accidents. Accident multiplicity is also a significant challenge for machine learning algorithms, which means that more data needs to be binned, screened, and cleaned, thereby providing "motive power" for further algorithm processing.
3) The failure modes are diverse. To date, in the field of international process reliability and safety technology, traditional dynamics have been explored; in fault detection and diagnosis of process control systems and mechanical systems, most focus on three classical failure modes: step failures, progressive failures, etc. However, in the aerospace field, most faults are difficult to classify and process simply, and because the faults form a dynamic process, more factors influence the algorithm. The more influencing factors are, the higher the data dimension is, the higher the requirement on the computing capacity of a hardware processing platform is, so that the principal component analysis method is used for dimension reduction processing during data cleaning. The principal component analysis method is to convert multiple indexes into a few comprehensive indexes (namely principal components), wherein each principal component can reflect most information of an original variable and the contained information is not repeated. The method can lead in various variables and simultaneously reduce the complex factors into a plurality of main components, simplify the problems, simultaneously can obtain more scientific and effective data information, and can also save precious airborne hardware resources.
Common and mature supervised learning algorithms such as a K nearest neighbor regression algorithm, a Bayesian algorithm, a classification and regression tree and the like are widely applied to the field of real-time fault analysis. However, the traditional method is often difficult to perform anomaly detection on a large amount of telemetry data with high dimension in the anomaly detection process.
Disclosure of Invention
The invention provides an aircraft telemetering data anomaly detection method and system based on a time convolution network, which can solve the technical problem that anomaly detection is difficult to be performed on high-dimensional mass telemetering data in the anomaly detection process in the prior art.
According to one aspect of the invention, the aircraft telemetry data anomaly detection method based on the time convolution network comprises the following steps: carrying out data preprocessing on the telemetering data; inputting the preprocessed telemetering data into an improved time convolution network, and capturing causality among sequences in a causal convolution and expansion convolution mode; inserting the weight with the value of 0 into the convolution kernel to form a hole convolution kernel as a basic calculation unit of the improved time convolution network; carrying out neural network training based on a basic computing unit, introducing a residual connection mode in the training process, replacing a set convolution layer with a residual block, and repeating the process until the construction of a data model is completed; and inputting the test data into a trained data model, and judging whether the aircraft telemetering data is abnormal or not through a loss function marked by the data.
Further, inputting the preprocessed telemetry data into an improved time convolution network, capturing causality among sequences in a causal convolution and expansion convolution mode, and inserting a weight with a value of 0 into a convolution kernel to form a hole convolution kernel as a basic calculation unit of the improved time convolution network specifically comprises: putting the preprocessed time sequence data into a time convolution network, processing input data by using a causal convolution network and an expansion convolution network, and calculating and extracting characteristic information of bottom layer data; inserting a weight with a value of 0 into the convolution kernel to form a hole convolution kernel, and taking the hole convolution kernel as a basic computing unit of the improved time convolution network; and carrying out nonlinear mapping on the characteristic information of the bottom data extracted by the convolution kernel by using an activation function after the hole convolution operation.
Further, the data preprocessing of the telemetry data to remove outliers and invalid data specifically comprises: the method comprises the steps of preprocessing telemetering data, eliminating null values and outliers in the data in a box diagram mode, and increasing the weight of a short flight stage by a data enhancement or attention mechanism method.
Further, inputting the test data into the trained data model, and judging whether the whole system is abnormal or not through the loss function labeled by the data specifically includes: setting a loss function for the trained data model, inputting test data into the trained data model, and calculating the abnormal score of the label according to the loss function; and comparing the abnormal score with a set threshold value to realize the abnormal detection of the aircraft system.
Further, processing the input data by using a causal convolution network and an expansion convolution network, wherein the processed data is y t ＝w 1 ·x t-2 +w 2 ·x t-1 +w 1 ·x t Wherein x is t-2 ,x t-1 ,x t As input variables, w 1 As an input variable x t-2 And x t Corresponding weight, w 2 As an input variable x t-1 Corresponding weight, y t To output data.
Further, the activation function is ReLU (x) = max { ax, x } (0 < a < 1).
Further, introducing a residual connection mode, wherein replacing the set convolutional layer with a residual block specifically comprises: and (3) realizing cross-layer connection of convolution by using residual connection, setting two layers of expansion convolution and a ReLU nonlinear function in a residual block, and carrying out weight normalization on the weight of a convolution kernel.
According to another aspect of the invention, the aircraft telemetry data anomaly detection system based on the time convolution network is provided, and the aircraft telemetry data anomaly detection system based on the time convolution network performs telemetry data anomaly detection by using the aircraft telemetry data anomaly detection method based on the time convolution network.
Further, an aircraft telemetry data anomaly detection system comprises: the data preprocessing module is used for preprocessing the telemetering data; the causal convolution and expansion convolution module is used for inputting the preprocessed telemetering data into an improved time convolution network and capturing causality among sequences in a causal convolution and expansion convolution mode; the cavity convolution kernel module is used for inserting a weight with a numerical value of 0 into the convolution kernel to form a cavity convolution kernel which is used as a basic computing unit of the improved time convolution network; the data model building module is used for carrying out neural network training based on the basic computing unit so as to complete the building of a data model; and the anomaly judgment module is used for inputting the test data into the trained data model and judging whether the aircraft telemetering data is abnormal or not through a loss function of data annotation.
The technical scheme of the invention provides an aircraft telemetering data anomaly detection method based on a time convolution network. The characteristic information of the input data is extracted through feedforward calculation, and feedback calculation is used for correcting errors of the network, so that the output is close to a true value as much as possible. The feedforward calculation captures causality of a sequence by using causal convolution and expansion convolution, trains a deep model in a residual connection mode, and inputs test data into the trained model for anomaly detection. Compared with the prior art, the aircraft telemetering data anomaly detection method provided by the invention utilizes the telemetering data of the aircraft to carry out overall evaluation on the aircraft system, and is a data-driven anomaly detection method. The method solves the problems that the traditional method is difficult to carry out overall evaluation on the aircraft system and the high-dimensional data anomaly detection effect is poor. The telemetering data of each parameter of the aircraft system is fully utilized, and the time sequence information is combined for anomaly detection. By the method, the abnormity of the aircraft system can be timely and effectively found, so that relevant measures can be quickly taken for repair, and irreversible major influence is prevented.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention. It is obvious that the drawings in the following description are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
FIG. 1 illustrates a flow diagram of a method for aircraft telemetry data anomaly detection based on a time convolutional network provided in accordance with a specific embodiment of the present invention;
fig. 2 is a schematic diagram illustrating a network structure of time convolution according to an embodiment of the present invention.
Detailed Description
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
The relative arrangement of the components and steps, the numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless it is specifically stated otherwise. Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description. Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate. In all examples shown and discussed herein, any particular value should be construed as merely illustrative, and not limiting. Thus, other examples of the exemplary embodiments may have different values. It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.
As shown in fig. 1 and 2, according to an embodiment of the present invention, there is provided an aircraft telemetry data anomaly detection method based on a time convolution network, including: carrying out data preprocessing on the telemetering data; inputting the preprocessed telemetering data into an improved time convolution network, and capturing causality among sequences in a causal convolution and expansion convolution mode; inserting the weight with the value of 0 into the convolution kernel to form a hole convolution kernel as a basic calculation unit of the improved time convolution network; carrying out neural network training based on a basic computing unit, introducing a residual connection mode in the training process, replacing a set convolution layer with a residual block, and repeating the process until the construction of a data model is completed; and inputting the test data into a trained data model, and judging whether the aircraft telemetering data is abnormal or not through a loss function marked by the data.
By applying the configuration mode, the aircraft telemetry data anomaly detection method based on the time convolution network is provided, and by means of data preprocessing of the telemetry data, outliers and invalid data are removed. The characteristic information of the input data is extracted through feedforward calculation, and feedback calculation is used for correcting errors of the network, so that the output is close to a true value as much as possible. The feedforward calculation uses causality of a causal convolution and an expansion convolution capture sequence, a deep layer model is trained in a residual error connection mode, and test data are input into the trained model for anomaly detection. Compared with the prior art, the aircraft telemetering data anomaly detection method provided by the invention utilizes the telemetering data of the aircraft to carry out overall evaluation on the aircraft system, and is a data-driven anomaly detection method. The method solves the problems that the traditional method is difficult to carry out overall evaluation on the aircraft system and the high-dimensional data anomaly detection effect is poor. The telemetering data of each parameter of the aircraft system is fully utilized, and the time sequence information is combined for anomaly detection. By the method, the abnormity of the aircraft system can be timely and effectively found, so that the relevant measures can be quickly taken for repairing, and the irreversible major influence is prevented.
Specifically, in the invention, in order to realize the aircraft telemetry data anomaly detection based on the time convolution network, data preprocessing needs to be carried out on the telemetry data firstly. In the present invention, the aircraft telemetry data typically includes null values and outliers, and during flight, the duration of each phase may vary, possibly affecting model training, resulting in model overfitting. In response to the above problems, data needs to be preprocessed. As a specific embodiment of the present invention, the data preprocessing of the telemetry data to remove outliers and invalid data specifically includes: the method comprises the steps of preprocessing telemetering data, eliminating null values and outliers in the data in a box diagram mode, and increasing the weight of a short flight stage by a data enhancement or attention mechanism method.
Further, after the telemetry data is preprocessed, the preprocessed telemetry data can be input into a modified time convolution network, and causality among sequences can be captured through a causal convolution and a dilation convolution. And inserting the weight with the value of 0 into the convolution kernel to form a hole convolution kernel as a basic calculation unit of the improved time convolution network.
In the present invention, inputting the preprocessed telemetry data into an improved time convolution network, capturing causality between sequences by a causal convolution and an expansion convolution, and inserting a weight with a value of 0 into a convolution kernel to form a hole convolution kernel as a basic calculation unit of the improved time convolution network specifically comprises: putting the preprocessed time sequence data into a time convolution network, processing the input data by using a causal convolution network and an expansion convolution network, calculating and extracting characteristic information of bottom layer data; inserting a weight with a value of 0 into the convolution kernel to form a hole convolution kernel, and taking the hole convolution kernel as a basic computing unit of the improved time convolution network; and carrying out nonlinear mapping on the characteristic information of the bottom data extracted by the convolution kernel by using an activation function after the hole convolution operation.
Under the configuration mode, the preprocessed time sequence data is put into a time convolution network, causality among sequences is captured in a causal convolution and expansion convolution mode, the concept of hollow convolution in the image processing field is used for reference, a weight with the value of 0 is inserted into a convolution kernel after the causal convolution and the expansion convolution to form a hollow convolution kernel, and the hollow convolution kernel is used as a basic computing unit of the improved time convolution network instead of a conventional convolution kernel, so that the network efficiency can be improved, and the training speed of a network model is accelerated.
As a specific embodiment of the invention, the method utilizes the telemetering data of the aircraft system to carry out anomaly detection, extracts the data characteristics through an improved time convolution network, constructs a data model, sets a corresponding loss function for the trained model, and judges whether the whole system has anomaly or not through the loss function marked by the data, thereby achieving the purpose of detecting the anomaly of the aircraft system. In the invention, aiming at the limitation of the time convolution network to the input layer, the improved time convolution network is provided by combining the work required by the invention. The mode of introducing the cavity convolution is adopted to avoid the stacking of the convolution kernel layer, the network structure is optimized, and the improved time convolution network structure is as follows:
1) The output of the convolution operation is the sum of the products of each input variable and the corresponding convolution weight, taking the convolution kernel size as 3 single convolution operations as an example, and the output of the convolution operation is y at any time t t The method comprises the following steps:
y t ＝w 1 ·x t-1 +w 2 ·x t +w 1 ·x t+1
wherein x is t-1 ,x t ,x t+1 As input variables, w 1 As an input variable x t-1 And x t+1 Corresponding weight, w 2 As an input variable x t Corresponding weight, y t To output data.
In the process of processing time series data, theoretically input data are processed for calculating and extracting characteristic information of bottom layer data. The output y at any time t is actually only related to the current time t and the previous input, and the causal convolution network is used for processing the input data so as to calculate and extract the characteristic information of the underlying data.
y t ＝w 1 ·x t-2 +w 2 ·x t-1 +w 1 ·x t
Wherein x is t-2 ,x t-1 ,x t As input variables, w 1 As an input variable x t-2 And x t Corresponding weight, w 2 As an input variable x t-1 Corresponding weight, y t To output data.
And inserting a weight with a value of 0 into the convolution kernel to form a hole convolution kernel, and replacing the conventional convolution kernel as a basic computing unit of the improved TCN (Temporal Convolutional Network), so that the Network efficiency is improved, the training speed of a Network model is accelerated, and the convergence is accelerated.
After the convolution operation, the features extracted by the convolution kernel need to be non-linearly mapped using an activation function. The method has the main function of carrying out nonlinear mapping on linear data and characteristic information so as to fit the real situation and prevent overfitting. In the present invention, the activation function is ReLU (x) = max { ax, x } (0 < a < 1).
Further, after the basic computing unit is obtained, neural network training can be carried out on the basis of the basic computing unit, in the training process, a residual error connection mode is introduced, the set convolution layer is replaced by the residual error block, and the process is repeated until the construction of the data model is completed.
As a specific embodiment of the present invention, the implementation of cross-layer connection of convolution using residual connection, and introducing a residual connection manner, and the replacing of a residual block with a set convolution layer specifically includes: and (3) realizing cross-layer connection of convolution by using residual connection, setting two layers of expansion convolution and a ReLU nonlinear function in a residual block, and carrying out weight normalization on the weight of a convolution kernel.
Furthermore, after the data model is constructed, the test data can be input into the trained data model, and whether the aircraft telemetering data is abnormal or not is judged through a loss function marked by the data. In the invention, inputting test data into a trained data model, and judging whether the whole system is abnormal or not through a loss function labeled by the data specifically comprises the following steps: setting a loss function for the trained data model, inputting test data into the trained data model, and calculating the abnormal score of the label according to the loss function; and comparing the abnormal score with a set threshold value to realize the abnormal detection of the aircraft system.
According to another aspect of the invention, a time convolution network based aircraft telemetry data anomaly detection system is provided, which uses the time convolution network based aircraft telemetry data anomaly detection method as described above for telemetry data anomaly detection.
By applying the configuration mode, the aircraft telemetry data anomaly detection system based on the time convolution network is provided, and the system carries out data preprocessing on the telemetry data to remove outliers and invalid data. The characteristic information of the input data is extracted through feedforward calculation, and feedback calculation is used for correcting errors of the network, so that the output is close to a true value as much as possible. The feedforward calculation captures causality of a sequence by using causal convolution and expansion convolution, trains a deep model in a residual connection mode, and inputs test data into the trained model for anomaly detection. Compared with the prior art, the aircraft telemetering data anomaly detection system provided by the invention utilizes the telemetering data of the aircraft to carry out overall evaluation on the aircraft system, and is a data-driven anomaly detection method. The system solves the problems that the traditional method is difficult to carry out overall evaluation on the aircraft system and the high-dimensional data anomaly detection effect is poor. The telemetering data of each parameter of the aircraft system is fully utilized, and the time sequence information is combined for carrying out anomaly detection. Through the system, the abnormity of the aircraft system can be timely and effectively found, so that the relevant measures can be quickly taken for repairing, and the irreversible major influence is prevented.
Further, in order to realize the detection of the aircraft telemetry data abnormity, the aircraft telemetry data abnormity detection system can be configured to comprise a data preprocessing module, a causal convolution and expansion convolution module, a cavity convolution kernel module, a data model construction module and an abnormity determination module, wherein the data preprocessing module is used for preprocessing telemetry data, the causal convolution and expansion convolution module is used for inputting preprocessed telemetry data into an improved time convolution network, the causality between sequences is captured through a causal convolution and expansion convolution mode, the cavity convolution kernel module is used for inserting a weight with a value of 0 into the convolution kernel to form a cavity convolution kernel as a basic calculation unit of the improved time convolution network, the data model construction module is used for carrying out neural network training based on the basic calculation unit to complete the construction of a data model, the abnormity determination module is used for inputting test data into the trained data model, and whether the telemetry data is abnormal or not is judged through a loss function of data marking.
For further understanding of the present invention, the method for detecting aircraft telemetry data anomaly based on a time convolution network provided by the present invention is described in detail below with reference to fig. 1 and 2.
As shown in fig. 1 and 2, according to a specific embodiment of the present invention, a method for detecting an anomaly in telemetry data of an aircraft based on a time convolution network is provided, which combines the deep learning field of the latest research focus, compares various deep learning methods, provides a method for detecting an anomaly in telemetry data of an aircraft based on a time convolution network, verifies the method, obtains a better effect, and develops a thought for the deep learning application in the field of telemetry data anomaly detection.
The invention aims to provide a method for constructing an aircraft telemetering data automatic interpretation model based on machine learning. The method utilizes the telemetering data of the subsystem of the aircraft to detect the abnormality of different systems, and is a subsystem multi-parameter abnormality detection method. The method can identify the abnormity which is difficult to be found by the traditional method, realizes the multi-parameter abnormity detection of the subsystem of the space vehicle, and has better popularization and application prospects.
In order to achieve the above purpose, the present invention provides an anomaly detection method based on a time convolution network. A time convolutional network is a network structure that solves the sequence problem. Compared with a traditional cyclic neural network for processing a time sequence model, the time convolution network introduces the concept of convolution, realizes parallel computation, and can capture local information and flexibly adjust the receptive field size. According to the method, the outlier and invalid data are removed by performing data preprocessing on the telemetered data. The characteristic information of the input data is extracted through feedforward calculation, and feedback calculation is used for correcting errors of the network, so that the output is close to a true value as much as possible. The feedforward calculation uses causality of a causal convolution and an expansion convolution capture sequence, a deep layer model is trained in a residual error connection mode, and test data are input into the trained model for anomaly detection. The specific scheme comprises the following steps:
the method comprises the following steps: and performing data preprocessing on the telemetry data. In this embodiment, the aircraft telemetry data typically includes null values and outliers, and during flight, the duration of each phase may vary, possibly affecting model training, resulting in model overfitting. Aiming at the problems, the data needs to be preprocessed, null values and outliers in the data are removed in a box diagram mode, and the weight of a flight stage with short time is increased by a data enhancement or attention mechanism method.
Step two: inputting the preprocessed telemetering data into an improved time convolution network, and capturing causality among sequences in a causal convolution and expansion convolution mode; and inserting the weight with the value of 0 into the convolution kernel to form a hole convolution kernel as a basic calculation unit of the improved time convolution network. And (3) carrying out neural network training based on the basic computing unit, introducing a residual connection mode in the training process, replacing the set convolution layer with the residual block, and repeating the process until the construction of the data model is completed.
In this embodiment, the preprocessed time sequence data is put into a time convolution network, causality between sequences is captured in a causal convolution and expansion convolution manner, a concept of hollow convolution in the image processing field is used for reference, a weight with a value of 0 is inserted into a product kernel after the causal convolution and the expansion convolution to form a hollow convolution kernel, and the hollow convolution kernel is used as a basic computing unit of the improved time convolution network instead of a conventional convolution kernel, so that the network efficiency is improved, and the training speed of a network model is accelerated. And (3) inputting the telemetered data of the aircraft into the depth model in the step two after preprocessing in the step one, and in the training process, replacing a certain convolutional layer with a residual block in a mode of introducing residual connection, so that the network can transmit information across layers, the degradation problem of the depth neural network is well solved, and the convergence rate of the model is improved.
And step three, inputting the test data into the trained data model, and judging whether the aircraft telemetering data is abnormal or not through a loss function marked by the data.
In this embodiment, the aircraft telemetry data typically contains null and outliers, and the data is not equal in length during the various phases of the flight. To address this problem, the data needs to be preprocessed before the data is input into the training model. Null values and outliers contained in the data are processed in a box diagram mode, and then the weights of different stages of the telemetry data are balanced in a data enhancement or attention mechanism mode. And inputting the data after the data preprocessing is finished into a time convolution network model for training, and storing the trained model to the local. When data needs to be subjected to anomaly detection, test data is input into a trained model, and anomaly detection can be realized by calculating an anomaly score and a threshold range.
The aircraft system generates periodic high-dimensional time sequence data, and parameters with different dimensions need to be subjected to outlier elimination and data normalization processing. And aiming at the problem that the time of each stage of the flight is different, the data is balanced in a mode of carrying out attention weighting on the data. And dividing the preprocessed model, inputting the divided model into an improved time convolution network for training, wherein the time convolution network can realize parallel computation, and storing the trained model. And inputting the test data into a trained model, and calculating the abnormal score of the label. And setting a threshold according to the actual situation, and calculating the abnormal score and the threshold to realize the abnormal detection of the aircraft system. In this embodiment, it is assumed that the calculated abnormal score exceeds the threshold range, and it is considered that the aircraft system is abnormal, and it is assumed that the calculated abnormal beam splitting is within the set threshold range, and it is considered that the aircraft system is normal.
In summary, the invention provides an anomaly detection method for aircraft telemetry data from a data driving perspective, and the method is based on a time convolution network, extracts causality of the data through causal convolution and expansion convolution, adds methods such as cavity convolution and residual connection, and improves the efficiency of model training. The algorithm can capture timeliness of the telemetering data, can perform multichannel parallel computation, solves the problem of poor detection effect possibly caused by high dimensionality of the telemetering data, and has high detection accuracy. Ground measurement and control personnel can carry out integral modeling on the aircraft system according to the past flight data, set an evaluation standard, find abnormal conditions possibly existing in the system in time, repair the system in time and avoid the serious influence of aircraft damage. The method can also be used for abnormality detection of other high-dimensional time sequence systems after expansion and improvement. Therefore, compared with the prior art, the aircraft telemetering data anomaly detection method provided by the invention utilizes the telemetering data of the aircraft to carry out overall evaluation on the aircraft system, and is a data-driven anomaly detection method. The method solves the problems that the traditional method is difficult to carry out overall evaluation on the aircraft system and the high-dimensional data anomaly detection effect is poor. The telemetering data of each parameter of the aircraft system is fully utilized, and the time sequence information is combined for anomaly detection. By the method, the abnormity of the aircraft system can be timely and effectively found, so that the relevant measures can be quickly taken for repairing, and the irreversible major influence is prevented.
For ease of description, spatially relative terms such as "over … …", "over … …", "over … …", "over", etc. may be used herein to describe the spatial positional relationship of one device or feature to another device or feature as shown in the figures. It will be understood that the spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures. For example, if a device in the figures is turned over, devices described as "above" or "on" other devices or configurations would then be oriented "below" or "under" the other devices or configurations. Thus, the exemplary term "above … …" may include both orientations of "above … …" and "below … …". The device may be otherwise variously oriented (rotated 90 degrees or at other orientations) and the spatially relative descriptors used herein interpreted accordingly.
It should be noted that the terms "first", "second", and the like are used to define the components, and are only used for convenience of distinguishing the corresponding components, and the terms have no special meanings unless otherwise stated, and therefore, the scope of the present invention should not be construed as being limited.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (9)
1. The aircraft telemetry data anomaly detection method based on the time convolution network is characterized by comprising the following steps:
carrying out data preprocessing on the telemetering data;
inputting the preprocessed telemetering data into an improved time convolution network, and capturing causality among sequences in a causal convolution and expansion convolution mode;
inserting the weight with the value of 0 into the convolution kernel to form a hole convolution kernel as a basic calculation unit of the improved time convolution network;
carrying out neural network training based on a basic computing unit, introducing a residual connection mode in the training process, replacing a set convolution layer with a residual block, and repeating the process until the construction of a data model is completed;
and inputting the test data into a trained data model, and judging whether the aircraft telemetering data is abnormal or not through a loss function marked by the data.
2. The method for detecting aircraft telemetry data anomalies based on a time convolution network as claimed in claim 1, wherein the preprocessing telemetry data is input into the improved time convolution network, causality among sequences is captured through causal convolution and expansion convolution, and a weight with a value of 0 is inserted into a convolution kernel to form a hole convolution kernel as a basic calculation unit of the improved time convolution network, and the method specifically comprises the following steps:
putting the preprocessed time sequence data into a time convolution network, processing input data by using a causal convolution network and an expansion convolution network, and calculating and extracting characteristic information of bottom layer data;
inserting a weight with a value of 0 into the convolution kernel to form a hole convolution kernel, and taking the hole convolution kernel as a basic computing unit of the improved time convolution network;
and carrying out nonlinear mapping on the characteristic information of the bottom data extracted by the convolution kernel by using an activation function after the hole convolution operation.
3. The aircraft telemetry data anomaly detection method based on the time convolutional network as claimed in claim 2, wherein the data preprocessing of the telemetry data to eliminate outliers and invalid data specifically comprises: the method comprises the steps of preprocessing telemetering data, eliminating null values and outliers in the data in a box diagram mode, and increasing the weight of a short flight stage by a data enhancement or attention mechanism method.
4. The aircraft telemetry data anomaly detection method based on the time convolutional network as claimed in claim 3, wherein the step of inputting test data into a trained data model, and the step of judging whether the whole system is abnormal through a loss function labeled by data specifically comprises the steps of: setting a loss function for the trained data model, inputting test data into the trained data model, and calculating the abnormal score of the label according to the loss function; and comparing the abnormal score with a set threshold value to realize the abnormal detection of the aircraft system.
5. The method of claim 4, wherein the input data is processed using a causal convolutional network and an expanded convolutional network, and the processed data is y t ＝w 1 ·x t-2 +w 2 ·x t-1 +w 1 ·x t Wherein x is t-2 ,x t-1 ,x t As input variables, w 1 As an input variable x t-2 And x t Corresponding weight, w 2 As an input variable x t-1 Corresponding weight, y t To output data.
6. The method of claim 5, wherein the activation function is ReLU (x) = max { ax, x } (0 < a < 1).
7. The aircraft telemetry data anomaly detection method based on the time convolutional network as claimed in any one of claims 1 to 6, wherein a residual connection mode is introduced, and replacing the set convolutional layer with a residual block specifically comprises: and (3) realizing cross-layer connection of convolution by using residual connection, setting two layers of expansion convolution and a ReLU nonlinear function in a residual block, and carrying out weight normalization on the weight of a convolution kernel.
8. An aircraft telemetry data anomaly detection system based on a time convolution network, characterized in that the aircraft telemetry data anomaly detection system based on the time convolution network performs telemetry data anomaly detection by using the aircraft telemetry data anomaly detection method based on the time convolution network according to claims 1 to 7.
9. The time convolutional network-based aircraft telemetry data anomaly detection system of claim 8, comprising:
the data preprocessing module is used for preprocessing the telemetering data;
a causal convolution and dilation convolution module for inputting the preprocessed telemetry data into a modified time convolution network to capture causality between sequences by means of causal convolution and dilation convolution;
the cavity convolution kernel module is used for inserting a weight with a numerical value of 0 into the convolution kernel to form a cavity convolution kernel which is used as a basic computing unit of the improved time convolution network;
the data model building module is used for carrying out neural network training based on a basic computing unit so as to complete the building of a data model;
and the abnormity judgment module is used for inputting the test data into the trained data model and judging whether the aircraft telemetering data is abnormal or not through a loss function marked by the data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211035316.5A CN115510950A (en) | 2022-08-26 | 2022-08-26 | Aircraft telemetry data anomaly detection method and system based on time convolution network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211035316.5A CN115510950A (en) | 2022-08-26 | 2022-08-26 | Aircraft telemetry data anomaly detection method and system based on time convolution network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115510950A true CN115510950A (en) | 2022-12-23 |
Family
ID=84501024
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211035316.5A Pending CN115510950A (en) | 2022-08-26 | 2022-08-26 | Aircraft telemetry data anomaly detection method and system based on time convolution network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115510950A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116305531A (en) * | 2023-01-13 | 2023-06-23 | 北京控制工程研究所 | Spacecraft health evolution model modeling method, device, equipment and medium |
| CN116304884A (en) * | 2023-05-11 | 2023-06-23 | 西安衍舆航天科技有限公司 | Spacecraft telemetry data health prediction method, system, equipment and storage medium |
| CN116361728A (en) * | 2023-03-14 | 2023-06-30 | 南京航空航天大学 | Civil aircraft system level abnormal precursor identification method based on real-time flight data |
-
2022
- 2022-08-26 CN CN202211035316.5A patent/CN115510950A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116305531A (en) * | 2023-01-13 | 2023-06-23 | 北京控制工程研究所 | Spacecraft health evolution model modeling method, device, equipment and medium |
| CN116305531B (en) * | 2023-01-13 | 2023-09-15 | 北京控制工程研究所 | Spacecraft health evolution model modeling method, device, equipment and medium |
| CN116361728A (en) * | 2023-03-14 | 2023-06-30 | 南京航空航天大学 | Civil aircraft system level abnormal precursor identification method based on real-time flight data |
| CN116361728B (en) * | 2023-03-14 | 2024-01-23 | 南京航空航天大学 | Civil aircraft system level abnormal precursor identification method based on real-time flight data |
| CN116304884A (en) * | 2023-05-11 | 2023-06-23 | 西安衍舆航天科技有限公司 | Spacecraft telemetry data health prediction method, system, equipment and storage medium |
| CN116304884B (en) * | 2023-05-11 | 2023-07-28 | 西安衍舆航天科技有限公司 | Spacecraft telemetry data health prediction method, system, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP7012871B2 (en) | Devices and methods for controlling the system | |
| Han et al. | An adaptive spatiotemporal feature learning approach for fault diagnosis in complex systems | |
| CN115510950A (en) | Aircraft telemetry data anomaly detection method and system based on time convolution network | |
| US8306931B1 (en) | Detecting, classifying, and tracking abnormal data in a data stream | |
| CN111487991B (en) | Unmanned aerial vehicle cluster-oriented security level prediction method and system | |
| CN110609524B (en) | Industrial equipment residual life prediction model and construction method and application thereof | |
| CN110216680B (en) | Cloud-ground cooperative fault diagnosis system and method for service robot | |
| CN112947385B (en) | Aircraft fault diagnosis method and system based on improved Transformer model | |
| CN111289250A (en) | Method for predicting residual service life of rolling bearing of servo motor | |
| CN110851654A (en) | Industrial equipment fault detection and classification method based on tensor data dimension reduction | |
| CN116610092A (en) | Method and system for vehicle analysis | |
| CN115204302A (en) | Unmanned aerial vehicle small sample fault diagnosis system and method | |
| Titouna et al. | An online anomaly detection approach for unmanned aerial vehicles | |
| Wang et al. | CNN-based hybrid optimization for anomaly detection of rudder system | |
| Hasan et al. | Wasserstein GAN-based Digital Twin Inspired Model for Early Drift Fault Detection in Wireless Sensor Networks | |
| CN116560341A (en) | Industrial robot fault diagnosis model and fault diagnosis method | |
| CN116934304A (en) | Intelligent power distribution room equipment operation maintenance management system and method thereof | |
| CN116842379A (en) | Mechanical bearing residual service life prediction method based on DRSN-CS and BiGRU+MLP models | |
| CN116929815A (en) | Equipment working state monitoring system and method based on Internet of things | |
| Verstraete et al. | A deep adversarial approach based on multi-sensor fusion for remaining useful life prognostics | |
| CN115452376A (en) | Bearing fault diagnosis method based on improved lightweight deep convolution neural network | |
| Chernyshev et al. | Integration of building information modeling and artificial intelligence systems to create a digital twin of the construction site | |
| Wang et al. | Complex equipment diagnostic reasoning based on neural network algorithm | |
| Gritsyuk et al. | CONVOLUTIONAL AND LONG SHORT-TERM MEMORY NEURAL NETWORKS BASED MODELS FOR REMAINING USEFUL LIFE PREDICTION. | |
| Zou et al. | Overview of Bearing Fault Diagnosis Based on Deep Learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |