CN115859202B - Abnormality detection method and device under non-stationary time sequence data stream scene - Google Patents
Abnormality detection method and device under non-stationary time sequence data stream scene Download PDFInfo
- Publication number
- CN115859202B CN115859202B CN202211484703.7A CN202211484703A CN115859202B CN 115859202 B CN115859202 B CN 115859202B CN 202211484703 A CN202211484703 A CN 202211484703A CN 115859202 B CN115859202 B CN 115859202B
- Authority
- CN
- China
- Prior art keywords
- data
- time
- data stream
- gate
- 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.)
- Active
Links
Landscapes
- Testing Or Calibration Of Command Recording Devices (AREA)
Abstract
The invention discloses an anomaly detection method and device under a non-stationary time sequence data stream scene. The model optimized by the method improves modeling precision and ensures global optimization achieved by the solution on the premise of ensuring rationality and basic accuracy of the method. The invention can effectively distinguish the non-stationarity and the data abnormality of the data, and greatly reduces the error detection rate and the omission rate of the abnormality detection under the non-stationary time sequence data stream scene.
Description
Technical Field
The invention relates to the field of data mining, in particular to the field of time sequence data anomaly detection, and specifically relates to an anomaly detection method and device under a non-stationary time sequence data stream scene.
Background
The time-series data is data that changes with time, and is recorded in time series by a unified index. Anomaly detection is a critical part of the task of time series data analysis. Time-series data can be divided into a stationary sequence and a non-stationary sequence according to the change rule of the data:
1) A plateau sequence, i.e. a sequence with substantially no trend changes. The observed values of such sequences fluctuate substantially within a fixed interval. Although the fluctuation degree of different time periods is different, the overall trend is unchanged, and the data value does not exceed the extreme value of the interval;
2) A non-stationary sequence, i.e. a sequence containing significant trend changes, seasonal or periodic features. It may comprise only one time series component or may be a combination of several time series components.
The non-stationary nature of time series data is also known in the art of data mining as data concept drift. In a real-world scenario, many time series data are accompanied by conceptual drift, i.e., the data stream is non-stationary and changes regularly over time. For example, in a bridge data acquisition scenario, over time, the ambient temperature may change significantly, which in turn may cause changes in the properties of the bridge structure, making the data acquired by some sensors extremely unstable. These data are not outliers due to bridge structural anomalies or sensor anomalies, but normal changes due to the environment, but existing models are not able to discriminate between such changes and anomalies in bridge data. Therefore, for data containing a concept drift phenomenon, the data value of the current period may be abnormal even if it occurs in the previous period; also, it may be normal even if the data value of the current period is significantly different from the previous data value.
In recent years, due to the influence of deep learning, an abnormality detection task also generates a new framework based on a variational self-encoder, and the framework takes a reconstruction error as a standard, so that the theoretical rationality and the accuracy of detection are ensured, and the application scene is wide. However, the currently existing models for time series anomaly detection tasks cannot distinguish between conceptual drift of data and data anomalies. The normal time series model can confuse abnormal data with normal data of different modes, and a new data mode generated by concept drift is regarded as the abnormal data, so that the performance of the time series abnormal detection model is generally obviously reduced. Therefore, it is necessary to improve the existing time series data modeling method to sense and adapt to the non-stationarity of the data.
Disclosure of Invention
Aiming at the problem that the existing time sequence abnormality detection model does not consider modeling of data non-stationarity, the invention adopts a variation self-encoder as an external frame of abnormality detection, models and optimizes the non-stationarity of the data and the time sequence of the data together by improving the acquisition mode of hidden vectors in a long-short time memory network, and provides an abnormality detection method and device based on the improved long-short time memory network under the scene, which effectively distinguishes the non-stationarity of the time sequence data and the data abnormality, and reduces the error detection rate and the omission rate of abnormality detection under the scene of non-stationary time sequence data flow.
The technical scheme adopted for solving the technical problems is as follows: an anomaly detection method in a non-stationary time series data stream scene, the method comprising the steps of:
(1) Data preprocessing: collecting data from m sensors for bridge structure health monitoring in real time, integrating and averaging the sensor data according to the sampling frequency of the sensors, and transversely splicing the m sensor data to form n-dimensional time sequence data;
(2) Inputting the obtained n-dimensional time series data into an encoder part of a variable self-encoder, wherein the encoder has a composition structure for improving a long-short-time memory network of hidden vectors and simultaneously learning the time sequence and the non-stationarity of the data; the hidden vector acquisition mode is as follows: the hidden state is mapped through two different custom matrixes respectively to obtain the expression of the mean value and the variance, wherein the variance part is also required to be processed through a softplus function, all the custom matrixes are randomly initialized through an Xavier algorithm, and the weight is continuously adjusted in the training process; the average value and the variance are used for obtaining the expression of the Gaussian distribution of the data at the current moment, the KL distance is calculated with the Gaussian distribution at the previous moment, the data distribution quantization difference of the two parts is obtained, and the quantization result and the hidden state are spliced to obtain the hidden vector; the expression is as follows:
z t ＝[h t ,D(N(W μ ·h t ,softplus(W σ ·h t ))||N(W μ ·h t-1 ,softplus(W σ ·h t-1 )))]
wherein h is t For the output value at the current time, W μ 、W σ H is a custom matrix of mean and variance respectively t-1 The softplus is an activation function for the output value of the last moment;
(3) Output hidden vector z of encoder t Decoding as input to a decoder; the structure of the decoder is consistent with that of the encoder, but a layer of fully-connected neural network is added at the end of the decoder, and the decoding result is mapped back to the data space;
(4) And carrying out reconstruction error calculation on the output result of the decoder and the original input, wherein the error quantization result obtained by normalization is abnormal data with the value larger than the self-defined threshold value, and the other data are normal data.
Further, in the step (1), data of each minute is averaged respectively with a minute as a minimum time unit, and the average value of the data sampled every minute is used as the data record of the current time to process the missing value of the sensor sample.
Further, in step (2), the forget gate f in the long-short-term memory network t The expression of (2) is as follows:
f t ＝sigmoid(W f ·[z t-1 ,x t ]+b f )
wherein z is t-1 For the hidden vector obtained in the previous period, x t W is the data of the current moment f Custom matrix for forgetting gate, b f For forgetting the offset of the gate, sigmoid is the activation function.
Further, in step (2), the input gate i in the long-short-time memory network t The acquisition steps are consistent with those of the forgetting gate, but the custom matrix and the offset are different; the expression is as follows:
i t ＝sigmoid(W i ·[z t-1 ,x t ]+b i )
wherein W is i Custom matrix for input gates, b i Is the offset of the input gate.
Further, in step (2), the intermediate state C 'of the cells in the network is memorized for a long time' t The acquisition steps of the method are consistent with the forgetting gate and the input gate, but the custom matrix, the offset and the activation function are different, and the expression is as follows:
C′ t ＝tanh(W c ·[z t-1 ,x t ]+b C )
wherein W is C Custom matrix for cell intermediate state, b C The tan h is the activation function, which is the offset of the cell intermediate state.
Further, in step (2), the current state C of the node in the network is memorized for a long time t The information quantity of the two-part structure is determined by a forgetting gate and an input gate and the expression is as follows:
C t ＝f t *C t-1 +i t *C′ t 。
further, in step (2), the output gate o in the long-short-term memory network t Acquisition step and forget gate andthe input gates are consistent, but the custom matrix and the offset are different; the expression is as follows:
o t ＝sigmoid(W o ·[z t-1 ,x t ]+b o )
wherein W is o Custom matrix for cell intermediate state, b o Is the offset of the cell intermediate state.
Further, in step (2), the hidden state h of the current node in the network is memorized for a long time t The information quantity is controlled by an output gate and is obtained by processing the current node state through an activation function tanh, and the expression is as follows:
h t ＝o t *tanh(C t )。
further, in the step (4), the reconstruction error adopts euclidean distance, and the expression is:
wherein x is i For the original input value, x' i Output values for the decoder.
On the other hand, the invention also provides an abnormality detection device under the non-stationary time sequence data stream scene, which comprises a memory and one or more processors, wherein executable codes are stored in the memory.
The beneficial effects of the invention are as follows: an anomaly detection method in a non-stationary time series data stream scenario is provided. The variable self-encoder is used as an external basic frame, and modeling and optimization are carried out on the non-stationarity of the data and the time sequence of the data by improving the acquisition mode of hidden vectors in a long-short-time memory network. The model optimized by the method improves modeling precision and ensures global optimization achieved by the solution on the premise of ensuring rationality and basic accuracy of the method.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is an overall schematic diagram of a non-stationary time series data flow anomaly detection method framework;
fig. 2 is a detailed diagram of an improved long and short term memory network.
Fig. 3 is a block diagram of an anomaly detection device in a non-stationary time-series data stream scene.
Detailed Description
The invention will be described in more detail in the following, in particular in real time, with reference to the accompanying drawings.
The method selects bridge structure health detection as a task, and the data set is data acquired in real time in a bridge structure health monitoring system sensor. And extracting time sequence characteristics and non-stationary characteristics of bridge structure data, and optimizing a long-short-term memory neural network model. And inputting the real-time collected data in the bridge structure health monitoring system sensor into the optimized model, outputting an error quantification result serving as bridge structure data, and completing the health monitoring of the bridge structure through a self-defined threshold value.
As shown in fig. 1, an overall schematic diagram of an anomaly detection method framework in a non-stationary timing scene is provided, and the method encodes processed data, extracts timing characteristics and non-stationary characteristics of the data, and maps the data to a hidden space. The hidden vector corresponding to the current data is obtained through sampling the hidden space, the hidden vector is input to a decoder for decoding operation, and the hidden vector is mapped back to the data space through key information contained in the hidden vector, so that the aim of reconstruction is achieved.
For normal data, because the change trend of the data is learned by a model, the model can reconstruct the normal data well in theory, namely, the reconstruction error is smaller; for abnormal data, the model cannot learn the data mode of the abnormal data because the abnormal data does not accord with the conventional change rule, so that the reconstruction error is larger. By the above judgment criteria, anomalies in data can be accurately captured.
The method proposed by the invention needs to satisfy the following assumptions:
1) The data contains time sequence property, namely each data point in the data has a front-back association relation;
2) The data contains non-stationary properties, i.e., the data varies or fades periodically over time, rather than over a fixed interval.
Aiming at a real bridge data set, the method provided by the invention comprises the following specific steps:
(1) Data preprocessing: the sensor for acquiring the bridge data comprises a cable force acceleration sensor, a mechanical anemometer, a structural strain sensor, an external temperature and humidity sensor, a structural temperature sensor and a water level sensor. Firstly, data of 6 sensors are respectively processed, the sampling frequency of each sensor is very high, generally 10Hz or 50Hz, the data volume is far greater than the model training requirement, most of the data has almost no change in a period of time, and the data volume needs to be reduced because the data is redundant for the training process. In addition, the sampling frequency and the sampling time period of different sensors are inconsistent, and the sampling of each sensor has a certain defect, but the missing data is also completely inconsistent, so that great difficulty is caused to data alignment. The data of a certain month is selected as the preprocessed data, and all sensors in the month are guaranteed to have sampling records, so that the problem of inconsistent sampling time periods of the sensors can be solved. However, even one month of data has millions of records, and it is difficult to align the data records of all sensors by time stamp due to missing problems. Therefore, we average the data of each minute separately in the smallest time unit of one minute as the data record of the current time. The data of the sensor is not continuously lost for one minute, so that the problem of data loss can be solved, and redundant data is obviously reduced. After the processing of the steps is finished, respectively carrying out normalization operation on the data of each sensor, and finally splicing to obtain multi-characteristic time sequence data X epsilon l X d.
(2) The preprocessed bridge data is input into an encoder part of a variable self-encoder as shown in fig. 2, the basic composition structure of the encoder is an improved long-short-time memory network, and the time sequence and the non-stationarity of the data are simultaneously learned: for data point x at the current time t Performing linear transformation through a plurality of different custom matrixes, adding different bias items, and finally obtaining a forgetting gate f after processing by an activation function sigmoid or tanh t Input gate i t Output gate o t Cell intermediate state C 'at current moment' t . The specific process is as follows:
(2.1) first calculate the forgetting door f t . The hidden vector z obtained in the previous period t-1 Data x from the current time t After combination, with custom matrix W f Multiply and add offset b f And obtaining the forgetting gate after the obtained result is subjected to sigmoid processing of the activation function. The formula is expressed as follows: f (f) t ＝sigmoid(W f ·[z t-1 ,x t ]+b f )。z t-1 For the hidden vector obtained in the previous period, x t W is the data of the current moment f Custom matrix for forgetting gate, b f For forgetting the offset of the gate, sigmoid is the activation function. Wherein the manner of acquisition of the hidden vector is described in the following steps;
(2.2) calculating an input gate. The acquisition steps of the input gates are consistent with the forgetting gates, but the custom matrix and offset are different. The formula is expressed as follows: i.e t ＝sigmoid(W i ·[z t-1 ,x t ]+b i )；W i Custom matrix for input gates, b i Is the offset of the input gate.
(2.3) calculating the intermediate state and the current state of the cell. The cell intermediate state obtaining step is consistent with the two gating units, but the custom matrix, the offset and the activation function are different, and the formula is expressed as follows: c'. t ＝tanh(W C ·[z t-1 ,x t ]+b C )。W C Custom matrix for cell intermediate state, b C The tan h is the activation function, which is the offset of the cell intermediate state.
The information quantity of the cell state at the previous moment and the cell intermediate state at the current moment which is input to the cell state at the current moment is determined through the weighting processing of the forgetting gate and the input gate, and the cell state at the current moment, namely C, is obtained t ＝f t *C t-1 +i t *C′ t 。
(2.4) calculating an output gate. The acquisition steps of the input gates are identical to the gating cells described above, but the custom matrix and offset are different. The formula is expressed as follows: o (o) t ＝sigmoid(W o ·[z t-1 ,x t ]+b o )；W o Custom matrix for cell intermediate state, b o Is the offset of the cell intermediate state.
(2.5) calculating the hidden state and hidden vector of the current node. After the cell state at the current moment is processed by an activation function tanh, the output gate determines the output information quantity to obtain the hidden state of the current node, namely h t ＝o t *tanh(C t )。
And obtaining the distribution expression of the hidden space through two different linear transformations according to the hidden state at the current moment. Obtaining a quantization result of the non-stationarity of the current time data according to the difference between the hidden space distribution of the previous time and the hidden space distribution of the current time data, and splicing the result with the hidden state of the current node to obtain a hidden vector z of the current time data t . The above steps can be expressed as: z t ＝[h t ,D(N(W μ ·h t ,softplus(W σ ·h t ))||N(W μ ·h t-1 ,softplus(W σ ·h t-1 )))]. Wherein h is t For the output value at the current time, W μ 、W σ H is a custom matrix of mean and variance respectively t-1 The softplus is an activation function for the output value of the last moment;
(3) The hidden vector z obtained by the output of the encoder t Input to decoder, the decoder is divided into two parts, the first part has an internal junctionThe structure is consistent with the encoder structure, and the intermediate output z is obtained t ′ . In addition, the decoder needs to map the hidden space data back to the original data space, i.e. first perform a linear transformation W on the intermediate output z · t ′ And adding the bias term b z Then obtaining the reconstruction output x through the sigmoid processing of the activation function t ′ 。
(4) Calculating Euclidean distance of reconstructed output and original input, i.ex i As the original input value, x i ′ Output values for the decoder. In the model training process, the model training process is used as a reconstruction error of bridge structure data at the current moment and is used as one of loss functions for optimization; in the using process of the model, the result is an abnormal score, and whether the bridge structure data at the current moment is abnormal or not is judged according to the relation between the score and the self-defined threshold value.
The invention also provides an embodiment of an abnormality detection device in the non-stationary time sequence data stream scene corresponding to the embodiment of the abnormality detection method in the non-stationary time sequence data stream scene.
Referring to fig. 3, an abnormality detection device in a non-stationary time-series data stream scene provided by an embodiment of the invention includes a memory and one or more processors, where the memory stores executable codes, and the processors are configured to implement the abnormality detection method in the non-stationary time-series data stream scene in the above embodiment when executing the executable codes.
The embodiment of the abnormality detection device in the non-stationary time sequence data flow scene can be applied to any device with data processing capability, and the device with data processing capability can be a device or a device such as a computer. The apparatus embodiments may be implemented by software, or may be implemented by hardware or a combination of hardware and software. Taking software implementation as an example, the device in a logic sense is formed by reading corresponding computer program instructions in a nonvolatile memory into a memory by a processor of any device with data processing capability. In terms of hardware, as shown in fig. 3, a hardware structure diagram of an apparatus with any data processing capability where an anomaly detection device in a non-stationary time-series data flow scenario of the present invention is located is shown in fig. 3, and in addition to a processor, a memory, a network interface, and a nonvolatile memory shown in fig. 3, the apparatus with any data processing capability where an embodiment is located generally includes other hardware according to an actual function of the apparatus with any data processing capability, which is not described herein.
The implementation process of the functions and roles of each unit in the above device is specifically shown in the implementation process of the corresponding steps in the above method, and will not be described herein again.
For the device embodiments, reference is made to the description of the method embodiments for the relevant points, since they essentially correspond to the method embodiments. The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purposes of the present invention. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
The embodiment of the present invention also provides a computer readable storage medium having a program stored thereon, which when executed by a processor, implements the anomaly detection method in the non-stationary time-series data stream scenario in the above embodiment.
The computer readable storage medium may be an internal storage unit, such as a hard disk or a memory, of any of the data processing enabled devices described in any of the previous embodiments. The computer readable storage medium may be any external storage device that has data processing capability, such as a plug-in hard disk, a Smart Media Card (SMC), an SD Card, a Flash memory Card (Flash Card), or the like, which are provided on the device. Further, the computer readable storage medium may include both internal storage units and external storage devices of any data processing device. The computer readable storage medium is used for storing the computer program and other programs and data required by the arbitrary data processing apparatus, and may also be used for temporarily storing data that has been output or is to be output.
The above-described embodiments are intended to illustrate the present invention, not to limit it, and any modifications and variations made thereto are within the spirit of the invention and the scope of the appended claims.
Claims (10)
1. An anomaly detection method in a non-stationary time-series data stream scene is characterized by comprising the following steps:
(1) Data preprocessing: the method comprises the steps of collecting data in real time from m sensors for monitoring the health of a bridge structure, wherein the sensors for collecting bridge data comprise a cable force acceleration sensor, a mechanical anemometer, a structural strain sensor, an external temperature and humidity sensor, a structural temperature sensor and a water level sensor; integrating and averaging the sensor data according to the sampling frequency of the sensor, and transversely splicing m sensor data to form n-dimensional time sequence data;
(2) Inputting the obtained n-dimensional time series data into an encoder part of a variable self-encoder, wherein the encoder has a composition structure for improving a long-short-time memory network of hidden vectors and simultaneously learning the time sequence and the non-stationarity of the data; the hidden vector acquisition mode is as follows: the hidden state is mapped through two different custom matrixes respectively to obtain the expression of the mean value and the variance, wherein the variance part is also required to be processed through a softplus function, all the custom matrixes are randomly initialized through an Xavier algorithm, and the weight is continuously adjusted in the training process; the average value and the variance are used for obtaining the expression of the Gaussian distribution of the data at the current moment, the KL distance is calculated with the Gaussian distribution at the previous moment, the data distribution quantization difference of the two parts is obtained, and the quantization result and the hidden state are spliced to obtain the hidden vector; the expression is as follows:
z t ＝[h t ，D(N(W μ ·h t ，softplus(W σ ·h t ))||N(W μ ·h t-1 ，softplus(W σ ·h t-1 )))]
wherein h is t For the output value at the current time, W μ 、W σ H is a custom matrix of mean and variance respectively t-1 The softplus is an activation function for the output value of the last moment;
(3) Output hidden vector z of encoder t Decoding as input to a decoder; the structure of the decoder is consistent with that of the encoder, but a layer of fully-connected neural network is added at the end of the decoder, and the decoding result is mapped back to the data space;
(4) And carrying out reconstruction error calculation on the output result of the decoder and the original input, wherein the error quantization result obtained by normalization is abnormal data with the value larger than the self-defined threshold value, and the other data are normal data.
2. The anomaly detection method in a non-stationary time-series data stream scenario of claim 1, wherein in step (1), data of each minute is averaged with a minimum time unit of one minute, and the average value of data sampled per minute is used as a data record of the current time to process the missing value of the sensor sample.
3. The anomaly detection method in a non-stationary time-series data stream scenario of claim 1, wherein in step (2), a forgetting gate f in a long-short-term memory network t The expression of (2) is as follows:
f t ＝sigmoid(W f ·[z t-1 ，x t ]+b f )
wherein z is t-1 For the hidden vector obtained in the previous period, x t W is the data of the current moment f Custom matrix for forgetting gate, b f For forgetting the offset of the gate, sigmoid is the activation function.
4. The anomaly detection method in a non-stationary time-series data stream scenario of claim 3, wherein in step (2), the input gate i in the long-short-time memory network t The acquisition steps are consistent with those of the forgetting gate, but the custom matrix and the offset are different; the expression is as follows:
i t ＝sigmoid(W i ·[z t-1 ，x t ]+b i )
wherein W is i Custom matrix for input gates, b i Is the offset of the input gate.
5. The method for anomaly detection in a non-stationary time-series data stream scenario according to claim 3 or 4, wherein in step (2), the intermediate state C 'of cells in the long-short-term memory network' t The acquisition steps of the method are consistent with the forgetting gate and the input gate, but the custom matrix, the offset and the activation function are different, and the expression is as follows:
C′ t ＝tanh(W C ·[z t-1 ，x t ]+b C )
wherein W is C Custom matrix for cell intermediate state, b C The tan h is the activation function, which is the offset of the cell intermediate state.
6. The anomaly detection method in a non-stationary time-series data stream scenario of claim 5, wherein in step (2), the current state C of a node in the network is memorized for a long time t The information quantity of the two-part structure is determined by a forgetting gate and an input gate and the expression is as follows:
C t ＝f t *C t-1 +i t *C′ t 。
7. the anomaly detection method in a non-stationary time-series data stream scenario according to claim 3 or 4, wherein in step (2), the output gate o in the long-short-time memory network t Acquisition step of (a)The forget gate is consistent with the input gate, but the custom matrix and the offset are different; the expression is as follows:
o t ＝sigmoid(W o ·[z t-1 ，x t ]+b o )
wherein W is o Custom matrix for cell intermediate state, b o Is the offset of the cell intermediate state.
8. The anomaly detection method in a non-stationary time-series data stream scenario of claim 7, wherein in step (2), the hidden state h of the current node in the network is memorized for a long time t The information quantity is controlled by an output gate and is obtained by processing the current node state through an activation function tanh, and the expression is as follows:
h t ＝o t *tanh(C t )。
9. the method for anomaly detection in a non-stationary time-series data stream scenario of claim 1, wherein in step (4), the reconstruction error uses euclidean distance, and the expression is:
wherein x is i For the original input value, x' i Output values for the decoder.
10. An anomaly detection apparatus in a non-stationary time series data stream scenario, comprising a memory and one or more processors, the memory having executable code stored therein, wherein the processor, when executing the executable code, is configured to implement the steps of the anomaly detection method in a non-stationary time series data stream scenario as claimed in any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211484703.7A CN115859202B (en) | 2022-11-24 | 2022-11-24 | Abnormality detection method and device under non-stationary time sequence data stream scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211484703.7A CN115859202B (en) | 2022-11-24 | 2022-11-24 | Abnormality detection method and device under non-stationary time sequence data stream scene |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115859202A CN115859202A (en) | 2023-03-28 |
| CN115859202B true CN115859202B (en) | 2023-10-10 |
Family
ID=85666071
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211484703.7A Active CN115859202B (en) | 2022-11-24 | 2022-11-24 | Abnormality detection method and device under non-stationary time sequence data stream scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115859202B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583570A (en) * | 2018-11-30 | 2019-04-05 | 重庆大学 | The method for determining bridge health monitoring system abnormal data source based on deep learning |
| CN109948117A (en) * | 2019-03-13 | 2019-06-28 | 南京航空航天大学 | A kind of satellite method for detecting abnormality fighting network self-encoding encoder |
| CN112488235A (en) * | 2020-12-11 | 2021-03-12 | 江苏省特种设备安全监督检验研究院 | Elevator time sequence data abnormity diagnosis method based on deep learning |
| US11294756B1 (en) * | 2019-09-19 | 2022-04-05 | Amazon Technologies, Inc. | Anomaly detection in a network |
| CN114626479A (en) * | 2022-03-23 | 2022-06-14 | 南京大学 | Time sequence anomaly detection method and device based on variational self-coding time sequence decomposition |
| CN114676778A (en) * | 2022-03-28 | 2022-06-28 | 北京科技大学 | Abnormal data detection method and device for cooperative and mutual feedback of shield equipment data |
| KR102419782B1 (en) * | 2021-09-29 | 2022-07-12 | 가온플랫폼 주식회사 | Industrial facility failure prediction modeling technique and alarm integrated system and method based on artificial intelligence |
| WO2022160902A1 (en) * | 2021-01-28 | 2022-08-04 | 广西大学 | Anomaly detection method for large-scale multivariate time series data in cloud environment |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3620983B1 (en) * | 2018-09-05 | 2023-10-25 | Sartorius Stedim Data Analytics AB | Computer-implemented method, computer program product and system for data analysis |
| EP3767541A1 (en) * | 2019-07-17 | 2021-01-20 | Robert Bosch GmbH | A machine learnable system with conditional normalizing flow |
| KR20220019560A (en) * | 2020-08-10 | 2022-02-17 | 삼성전자주식회사 | Apparatus and method for monitoring network |
| US20220084371A1 (en) * | 2020-09-11 | 2022-03-17 | Capital One Services, Llc | Systems and methods for unsupervised detection of anomalous customer interactions to secure and authenticate a customer session |
-
2022
- 2022-11-24 CN CN202211484703.7A patent/CN115859202B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583570A (en) * | 2018-11-30 | 2019-04-05 | 重庆大学 | The method for determining bridge health monitoring system abnormal data source based on deep learning |
| CN109948117A (en) * | 2019-03-13 | 2019-06-28 | 南京航空航天大学 | A kind of satellite method for detecting abnormality fighting network self-encoding encoder |
| US11294756B1 (en) * | 2019-09-19 | 2022-04-05 | Amazon Technologies, Inc. | Anomaly detection in a network |
| CN112488235A (en) * | 2020-12-11 | 2021-03-12 | 江苏省特种设备安全监督检验研究院 | Elevator time sequence data abnormity diagnosis method based on deep learning |
| WO2022160902A1 (en) * | 2021-01-28 | 2022-08-04 | 广西大学 | Anomaly detection method for large-scale multivariate time series data in cloud environment |
| KR102419782B1 (en) * | 2021-09-29 | 2022-07-12 | 가온플랫폼 주식회사 | Industrial facility failure prediction modeling technique and alarm integrated system and method based on artificial intelligence |
| CN114626479A (en) * | 2022-03-23 | 2022-06-14 | 南京大学 | Time sequence anomaly detection method and device based on variational self-coding time sequence decomposition |
| CN114676778A (en) * | 2022-03-28 | 2022-06-28 | 北京科技大学 | Abnormal data detection method and device for cooperative and mutual feedback of shield equipment data |
Non-Patent Citations (8)
| Title |
|---|
| Chun Su等.Remaining useful life prediction via a variational autoencoder and a time-window-based sequence neural network.Quality and Reliability Engineering International.2020,第36卷(第05期),第1639-1656页. * |
| SANDEEP A. THORAT等.Improving Conversation Modelling using Attention Based Variational Hierarchical RNN.ISSN.2021,第20卷(第01期),第39-45页. * |
| 周壮等.基于 E2E Deep VAE-LSTM 的轴承退化预测应用研究.计算机应用研究.2022,第39卷(第07期),第2091-2097页. * |
| 李宇.基于神经网络与注意力机制的图像修复算法.中国优秀硕士学位论文全文数据库 信息科技辑.2021,(第02期),第I138-2372页. * |
| 李晓菊 ; 顾君忠 ; 程洁 ; .基于变分循环自动编码器的协同推荐方法.计算机应用与软件.2018,第39卷(第09期),第258-280页. * |
| 李晓菊等.基于变分循环自动编码器的协同推荐方法.计算机应用与软件.2018,第39卷(第09期),第258-280 页. * |
| 杨英 ; 唐平 ; .VAE＿LSTM算法在时间序列预测模型中的研究.湖南科技大学学报(自然科学版).2020,第35卷(第03期),第93-101页. * |
| 杨英等.VAE＿LSTM算法在时间序列预测模型中的研究.湖南科技大学学报(自然科学版).2020,第35卷(第03期),第93-101 页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115859202A (en) | 2023-03-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107688871B (en) | Water quality prediction method and device | |
| CN114386521B (en) | Abnormality detection method, abnormality detection system, abnormality detection device, abnormality detection program, and recording medium | |
| Jin et al. | Position encoding based convolutional neural networks for machine remaining useful life prediction | |
| CN110895705B (en) | Abnormal sample detection device, training device and training method thereof | |
| CN112101554A (en) | Anomaly detection method and device, equipment and computer-readable storage medium | |
| CN113569243A (en) | Deep semi-supervised learning network intrusion detection method based on self-supervised variation LSTM | |
| CN112948743B (en) | Coal mine gas concentration deficiency value filling method based on space-time fusion | |
| CN116522265A (en) | Industrial Internet time sequence data anomaly detection method and device | |
| Zhu et al. | A coupled model for dam foundation seepage behavior monitoring and forecasting based on variational mode decomposition and improved temporal convolutional network | |
| CN115062272A (en) | Water quality monitoring data abnormity identification and early warning method | |
| CN116993537A (en) | Power load abnormality detection method and system based on serial GRU (generic routing unit) self-encoder | |
| CN115759461A (en) | Internet of things-oriented multivariate time sequence prediction method and system | |
| CN116364203A (en) | Water quality prediction method, system and device based on deep learning | |
| CN116975645A (en) | Industrial process soft measurement modeling method based on VAE-MRCNN | |
| CN116665798A (en) | Air pollution trend early warning method and related device | |
| CN113609766A (en) | Soft measurement method based on depth probability latent model | |
| CN117874673A (en) | Abnormal data detection and interpretation method | |
| CN115859202B (en) | Abnormality detection method and device under non-stationary time sequence data stream scene | |
| CN112949947A (en) | Power failure early warning method and system based on convolution long-term and short-term memory network | |
| CN110866672B (en) | Data processing method, device, terminal and medium | |
| CN116992380A (en) | Satellite multidimensional telemetry sequence anomaly detection model construction method and device, anomaly detection method and device | |
| CN116070076A (en) | High-precision high-robustness radar missing signal completion system | |
| Xu et al. | Residual autoencoder-LSTM for city region vehicle emission pollution prediction | |
| CN116628612A (en) | Unsupervised anomaly detection method, device, medium and equipment | |
| CN116610973A (en) | Sensor fault monitoring and failure information reconstruction method and system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |