CN117390350A - Bridge girder erection machine time sequence abnormality detection system and detection method based on ExpertTowerGate network - Google Patents

Bridge girder erection machine time sequence abnormality detection system and detection method based on ExpertTowerGate network Download PDF

Info

Publication number
CN117390350A
CN117390350A CN202311460804.5A CN202311460804A CN117390350A CN 117390350 A CN117390350 A CN 117390350A CN 202311460804 A CN202311460804 A CN 202311460804A CN 117390350 A CN117390350 A CN 117390350A
Authority
CN
China
Prior art keywords
module
input
tower
bridge girder
girder erection
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
Application number
CN202311460804.5A
Other languages
Chinese (zh)
Inventor
赵宏魁
王亮
杨忠允
陈振山
王帅
刘淦
何爱东
王怡智
李英波
李文强
徐少博
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong Expressway Road And Bridge International Engineering Co ltd
Sinohydro Bureau 8 Co Ltd
Original Assignee
Shandong Expressway Road And Bridge International Engineering Co ltd
Sinohydro Bureau 8 Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shandong Expressway Road And Bridge International Engineering Co ltd, Sinohydro Bureau 8 Co Ltd filed Critical Shandong Expressway Road And Bridge International Engineering Co ltd
Priority to CN202311460804.5A priority Critical patent/CN117390350A/en
Publication of CN117390350A publication Critical patent/CN117390350A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/15—Correlation function computation including computation of convolution operations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/251—Fusion techniques of input or preprocessed data
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Mathematical Analysis (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Computational Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Algebra (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a bridge girder erection machine time sequence abnormality detection system and method based on an ExpertTowerGate network. The expert TowerGate network includes an expert module, a first stacking module, a gate computing module, a tower input integration module, a tower computing module, and a second stacking module. In the detection system, a mixed mode of the shared gate and the personalized gate is adopted, and for a large number of indexes, the characteristic space selected from each index is converted into a series of weights to fuse embedded values extracted from different expert networks, so that the relevance of different indexes is reflected. The expert module adopts a convolution unit, so that the correlation between the representation of the time sequence and the index can be enriched; the expert module can perform maximum feature extraction and correlation reservation on the data of the time window, so that the output has time features and context correlation, and for the bridge girder erection machine time series abnormality, abnormality detection can be performed according to the index of the correlation, and the bridge girder erection machine time series abnormality is not limited to a single index.

Description

Bridge girder erection machine time sequence abnormality detection system and detection method based on ExpertTowerGate network
Technical Field
The invention relates to the technical field of bridge engineering construction, in particular to a bridge girder erection machine time sequence anomaly detection system and method based on an ExpertTowerGate network.
Background
In the bridge construction technology of the overhead operation railway line and the narrow space bridge engineering, the bridge construction often spans complex conditions such as canyons, riding operation railway lines and the like, the construction environment is complex, and the conventional equipment cannot be directly adopted to erect the bridge for construction operation. In order to improve the construction quality of the bridge, ensure the construction safety and shorten the construction period, the portal type bridge erecting equipment is needed to assist. The adjustable movable support portal system is used for construction of a bent cap of a bridge in a complex construction environment of a riding operation railway line, does not interrupt railway operation, can be used for moving in narrow spaces under the conditions of dense urban areas of the operation railway, too close road sections of buildings and bridge sites, small-radius curve road sections and the like.
The adjustable movable portal bridge girder erection machine has the advantages that the working face is narrow, the mutual influence of the traffic carriers of the operation railway and the highway and the crowd is more direct, the adjustable portal bridge girder erection machine is used as large-scale construction equipment, the structural logic degree is high, the structure is complex, the precision and the working stability requirements of construction operation are extremely strict, and under the coupling effect of climate (such as ambient wind speed) and equipment operation states (levelness, longitudinal inclination angle and the like), the safety risk and the operation states of the adjustable movable portal bridge girder erection machine are accurately identified, predicted and analyzed, so that the safety production under various complex and severe bridge erection conditions is guaranteed. Therefore, a method for dynamically monitoring the real-time running state of the adjustable movable portal bridge girder erection machine and performing scientific analysis is required to be provided, so that the real working state of the adjustable movable portal bridge girder erection machine can be accurately judged, the state development trend of equipment can be perceived and predicted in advance, the bridge risk can be avoided in advance, and major production accidents can be avoided.
Disclosure of Invention
The invention provides a bridge girder erection machine time sequence abnormality detection system based on an ExpertTowerGate network, which is characterized by comprising an expert module, a first stacking module, a door calculation module, a tower input integration module, a tower calculation module and a second stacking module;
specifically, x_input represents input, and a time window of length l is used for the index Xt at time t, { X t-l ,…,X t-1 Denoted as Wt;
the input is respectively passed through 5 expert modules to obtain output as X 1 ,X 2 ,X 3 ,X 4 ,X 5 These 5 outputs are then used as inputs to the first stacking module, resulting in an expert stack, denoted Xs;
the input X_input is taken as the input of the gate computing module together with the parameters and the task number, and the output of the gate computing module is marked as G K (W t K ) The method comprises the steps of carrying out a first treatment on the surface of the The output of the gate computing module is divided into N tasks, and the output of the first stacking moduleXs is taken as the input of the tower input integration module, and the output of the tower input integration module is denoted as W;
finally, the output of the tower input integration module is processed by the tower calculation module and a plurality of outputs are processed by the second stacking module to obtain an output Ts, namely a vector
Subtracting the threshold value from the first one, and then calculating a mean square error sigma, wherein the result is 1,0, and the result respectively represents whether each index is normal or not, 1 is abnormal, and 0 is normal;
wherein,μ represents a threshold value, and N represents the length of the vector.
Based on the scheme, the calculation method of the first stacking module is as follows:
E (t) (W t )＝[f 1 (W t ),f 2 (W t ),…,f M (W t )]
wherein f i (x) Representing the output of the ith expert module for input x, E (t) (W t ) Is representative of the Xs output.
Based on the scheme, the formula of the tower input integration module is as follows:
B (t，k) ＝G k (W t k )⊙E (t) (W t )
wherein E is (t) (W t ) Representing the calculation result of the first stacking module, G K (W t K ) N task inputs representing the division of the Gate computing result, where "" -represents the multiplication of the vector and W=B (t,k) 。
Based on the scheme, the calculation method of the second stacking module comprises the following steps:
where Tower represents the Tower computation, and the intermediate result of Tower is denoted as T 1 ,…T n N represents the length of the tower calculation vector.
Based on the scheme, G K (W t K ) Is calculated using a shared gate Gs to receive all windows from the input, and a personalized gateOnly the kth index is concerned;
the mixed mode calculation formula of the shared gate and the personalized gate is as follows:
wherein Ws and W pk (k=1, 2, …, K) is a trainable parameter, ε>0.5, is the weight coefficient of the shared gate.
G K (W t K ) Representing the results of the shared and personalized gate mix calculations,representing the kth vector at time t.
On the basis of the scheme, in the expert module, input is firstly subjected to an unsqueeze unit, an input matrix is deformed to obtain M, then the M is input to a Conv2d convolution unit to obtain N, and then the N is sequentially subjected to a Relu activation unit, a Flatten flattening unit, an FC1 full connection 1 unit, a Relu activation unit and an FC2 full connection 2 unit to obtain output.
Based on the scheme, in the tower calculation module, the input is firstly subjected to FC1 full connection 1 unit to obtain U, then is subjected to Relu activation unit to obtain V, and finally is subjected to FC2 full connection 2 unit to obtain output.
The invention also provides a bridge girder erection machine time sequence abnormality detection method based on the ExpertTowerGate network, which specifically uses the detection system, before using, uses the existing sensor data to carry out iterative training on the detection system so as to lead the value of the loss function to reach the minimum, then evaluates the accuracy on a test set or a verification set, and leads the accuracy to reach more than 90 percent; in the training process, the adopted objective function is as follows:
as predicted value, y t Is true value, I.I.I 2 Representing the L2 paradigm calculation. .
In addition, the invention also provides an electronic device, which comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, wherein the processor realizes the steps of the detection method when executing the computer program.
In addition to the above-described electronic device, the present invention also provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the above-described detection method steps.
The bridge girder erection machine time sequence anomaly detection system based on the ExpertTowerGate network adopts a mixed mode of a shared door and a personalized door, and for a large number of indexes, the characteristic space selected from each index is converted into a series of Weights (Weights) to fuse embedded values (embedded) extracted from different Expert networks, so that the relevance of different indexes is reflected; the expert module adopts a convolution unit, so that the mutual relevance between the time sequence representation and the index can be enriched; the expert module can perform maximum feature extraction and correlation reservation on the data of the time window, so that the output has time features and context correlation, and for the bridge girder erection machine time series abnormality, abnormality detection can be performed according to the index of the correlation, and the bridge girder erection machine time series abnormality is not limited to a single index.
Drawings
FIG. 1 is an overall workflow diagram of bridge girder erection machine time sequence anomaly detection based on an ExpertTowerGate network;
FIG. 2 is a schematic diagram of an ExpertTowerGate network model in accordance with the present invention;
FIG. 3 is a diagram of the architecture of an expert module in the ExpertTowerGate network model;
FIG. 4 is an architecture diagram of a tower computation module in the ExpertTowerGate network model.
Detailed Description
The invention will be described in further detail with reference to the drawings and examples, it being noted that the examples described below are intended to facilitate the understanding of the invention and are not intended to limit the invention in any way.
The invention provides a time sequence abnormality detection system of a bridge girder erection machine, which is based on an ExpertTowerGate network model and can detect time sequence data (levelness, ambient wind speed, longitudinal inclination angle and the like) collected by a sensor arranged on the bridge girder erection machine in real time so as to judge whether the bridge girder erection machine is in a normal working state at the moment.
The levelness is measured by a level sensor, the level sensor belongs to one of angle sensors, the function of the level sensor is to measure the levelness of the bridge girder erection machine, which is also called an inclination angle sensor, and a level instrument or an inclination angle instrument is commonly called in engineering; the wind speed of the current working environment of the bridge girder erection machine is measured when the environment wind speed is the wind speed, and the invention adopts a TSI8475-03 wind speed sensor; the longitudinal inclination angle is measured by measuring the longitudinal inclination angle of the bridge girder erection machine to analyze the safety state of the bridge girder erection machine, and an SCL3300 inclination angle sensor is adopted.
If abnormality is detected at a certain moment, an alarm notification is sent out in time, after the notification is received by engineering personnel, the bridge girder erection machine is rechecked and rechecked at the first time, and if abnormality exists in the true, relevant adjustment is made, so that the bridge girder erection machine is ensured to stably run all the time.
The time series data format in construction from bridge girder erection machine sensor is:
X＝{X 1 ,…,X i ，…，X T }
wherein X is i For at a specific moment i.e. [1, T]The index above.
Each data pointThe j-th index value representing the point in time i (e.g., 2023/10/14 12:08:05) is, for example, levelness, ambient wind speed, longitudinal tilt angle, etc.
The bridge girder erection machine for construction of a certain bridge comprises the following time period data:
index X for time t t The anomaly detection system calculates an anomaly score based on historical observations and compares it with a threshold value (the threshold value is 80) to determine and decide X t Whether or not it is abnormal.
The whole work flow of the time sequence abnormality detection of the bridge girder erection machine is shown in figure 1, a sensor system of the bridge girder erection machine is deployed in advance, running data (levelness, ambient wind speed, longitudinal inclination angle and the like) are collected once every 5 seconds, and the collected data are sent to a workstation and stored in a computer hard disk; at the same time, the moment when the bridge girder erection machine is abnormal is manually recorded, for example 2023/10/14/12:08:15 is abnormal, and then the abnormal moment is manually marked. The collected time series data and anomaly data are then saved to a computer storage system as training data. At the same time, a ExpertTowerGate Network model is constructed, and parameters are initial parameters which are not trained. Training data is matched with a ExpertTowerGate Network model, and the model is trained. Data collected in three months by using sensors deployed by a bridge girder erection machine for constructing a certain bridge are taken as data samples, and 800000 pieces of data are used in total during training. Of these, 90% was used for training, 10% retention test, and ExpertTowerGate Network model was constructed. The invention adopts Adam as an optimizer, and the loss function adopts a mean square error loss function. And then carrying out iterative training on the data to enable the value of the loss function to reach the minimum, then evaluating the accuracy on a test set or a verification set, and obtaining an optimal model if the accuracy reaches more than 90%. The optimal ExpertTowerGate Network model is saved as a binary file. In the training process, the adopted objective function is as follows:
as predicted value, y t Is true value, I.I.I 2 Representing the L2 paradigm calculation.
For other running bridge girder erection machines, the sensor is used for collecting running data (levelness, ambient wind speed, longitudinal inclination angle and the like), an optimal ExpertTowerGate Network model is loaded and input into the model, time series data are detected, if an abnormal time point exists, relevant engineering personnel check the machine and make corresponding adjustment, normal and stable running of the machine is ensured, and if no abnormality exists, the detection is continuously carried out.
In the anomaly detection task, context observations, such as observations at nearby points in time, play an extremely important role in understanding current data, as they represent temporal patterns of relevant contexts. So for the index Xt, useA time window of length l, { X t-l ,…,X t-1 Denoted Wt, as input to accurately capture the temporal interdependence.
The architecture of the ExpertTowerGate Network model is shown in FIG. 2, in which X_input represents input, and for the index Xt at time t (e.g. 2023/10/14/12:08:25), a time window with length of l is used, { X t-l ,…,X t-1 Denoted as W t . The inputs are respectively passed through 5 expert modules (experiments) to obtain outputs X 1 ,X 2 ,X 3 ,X 4 ,X 5 These 5 outputs are then used as inputs to the first Stack module (Stack), resulting in an Expert Stack, denoted Xs.
The formula for the first stacked module is written as: e (E) (t) (W t )＝[f 1 (W t ),f 2 (W t ),…,f M (W t )]
Where f i (x) Representing the output of the ith expert module for input x, E (t) (W t ) That is, the output Xs.
X_input is taken as input of a Gate computing module together with parameters (parameters) and task numbers (task numbers), and the computing result is marked as G after Gate computing K (W t K ). The result of the calculation is divided into N Tasks (Tasks), and the output X of the first stack module s Together, as input to the tower input integration module (Tower input consolidation). The calculated output of the tower input integration module is denoted W. The formula is written as:
B (t，k) ＝G k (W t k )⊙E (t) (W t )
wherein E is (t) (W t ) Representing the calculation result of the first stacking module, G K (W t K ) N task inputs representing the division of the Gate computing result, where "" -represents the multiplication of the vector and W=B (t,k) 。
Finally, the output of the Tower input integration module is processed by a Tower calculation module (Tower) and a plurality of outputs are processed by a second stacking module (Stack) to obtainTo the output (denoted as Ts, i.e. vector) The formula is:
where Tower represents the Tower computation, and the intermediate result of Tower is denoted as T 1 … Tn, n represents the length of the tower calculated vector.
The finally obtainedThe predicted value is a vector indicating whether each index at that time is abnormal or not. Opred is the predicted value +.>And (3) subtracting the threshold value from the predicted value according to the abnormal prediction result of the abnormality, and then calculating a mean square error sigma, wherein the result is 1 and 0, and the result is respectively representing whether each index is normal or not, 1 is abnormal, and 0 is normal.
The formula is as followsHere μ represents a threshold, namely 80, n represents the length of the vector.
In the training process, the adopted objective function is as follows:
as predicted value, y t Is true value, I.I.I 2 Representing the L2 paradigm calculation.
In the calculation of the tower input integration module, in particular for G K (W t K ) Adopts a Shared gate (G) s Receiving all windows from the input, and a personalized door (Personalized gate)Only the kth index is concerned, so it receives only the value of the corresponding kth index. The sharing gate with higher weight can learn more stable corresponding relation from expert output fusion and input data.
The mixed mode calculation formula of the shared gate and the personalized gate is as follows:
wherein Ws and W pk (k=1, 2, …, K) is a trainable parameter, ε>0.5, which is the weight coefficient of the shared gate; g K (W t K ) Representing the results of the shared and personalized gate mix calculations,representing the kth vector at time t.
The method adopts a mixed gate computing mode, and for a large number of indexes, the feature space selected from each index is converted into a series of Weights (Weights) to fuse embedded values (embedded) extracted from different Expert networks. The extracted embedded values are then sent to a Tower (Tower) network, respectively. All tower networks jointly determine the final anomaly score value based on the difference between the predicted value and the true value.
The architecture of the Expert module (Expert) is shown in fig. 3, where the input first goes through an unsqueeze unit, transforms the input matrix to obtain M, then inputs the M to a Conv2d convolution unit to obtain N, and then goes through a Relu activation unit, a flat unit, an FC1 fully connected 1 unit, a Relu activation unit, and an FC2 fully connected 2 unit in sequence to obtain Output (Output). The expert module adopts a matrix deformation and convolution unit, can perform maximum feature extraction and correlation reservation on the data of the time window, so that the time features and the context correlation are output, and for the bridge girder erection machine time series abnormality, abnormality detection can be performed according to the correlation index instead of being limited to a single index.
The architecture of the tower calculation module is shown in fig. 4, where the input is first through FC1 fully connected 1 unit to obtain U, then through the Relu activation unit to obtain V, and finally through FC2 fully connected 2 unit to obtain Output (Output). The Power module converts the time series characteristic data into abnormality prediction data through a plurality of simple full connections, so that the detection of the abnormality is possible.
Therefore, for the ExpertTowerGate Network model, a mixed mode of a shared gate and a personalized gate is adopted, and for a large number of indexes, a feature space selected from each index is converted into a series of Weights (Weights) to fuse embedded values (embedded) extracted from different Expert (Expert) networks, so that the relevance of different indexes is reflected; the expert module adopts a convolution unit, so that the mutual relevance between the time sequence representation and the index can be enriched; the expert module can perform maximum feature extraction and correlation reservation on the data of the time window, so that the output has time features and context correlation, and for the bridge girder erection machine time series abnormality, abnormality detection can be performed according to the index of the correlation, and the bridge girder erection machine time series abnormality is not limited to a single index.
The system is applied to a bridge construction site, early warning is carried out on the abnormal operation condition of the bridge girder erection machine, and the accuracy rate is as high as 93% after test, so that the quality and the working efficiency of bridge erection construction are greatly improved, and the potential construction risk in production activities is reduced.

Claims (10)

1. The bridge girder erection machine time sequence abnormality detection system based on the ExpertTowerGate network is characterized in that the ExpertTowerGate network comprises an expert module, a first stacking module, a door calculation module, a tower input integration module, a tower calculation module and a second stacking module;
specifically, X_input represents input, and the index X at time t is t A time window of length l, { X t-l ,…,X t-1 Denoted as Wt;
the input is respectively passed through 5 expert modules to obtain output as X 1 ,X 2 ,X 3 ,X 4 ,X 5 These 5 outputs are then used as inputs to the first stacking module, resulting in an expert stack, denoted Xs;
the input X_input is taken as the input of the gate computing module together with the parameters and the task number, and the output of the gate computing module is marked as G K (W t K ) The method comprises the steps of carrying out a first treatment on the surface of the The output of the gate computation module is divided into N tasks, and the output Xs of the first stacking module is used as the input of the tower input integration module, and the output of the tower input integration module is marked as W;
finally, the output of the tower input integration module is processed by the tower calculation module and a plurality of outputs are processed by the second stacking module to obtain an output Ts, namely a vector
Subtracting the threshold value from the first one, and then calculating a mean square error sigma, wherein the result is 1,0, and the result respectively represents whether each index is normal or not, 1 is abnormal, and 0 is normal;
wherein,μ represents a threshold value, and N represents the length of the vector.
2. The bridge girder erection machine time sequence abnormality detection system based on the ExpertTowerGate network according to claim 1, wherein,
the calculation method of the first stacking module comprises the following steps: e (E) (t) (W t )＝[f 1 (W t ),f 2 (W t ),…,f M (W t )]
Wherein f i (x) Representing the output of the ith expert module for input x, E (t) (W t ) Is representative of the Xs output.
3. The bridge girder erection machine time sequence abnormality detection system based on the ExpertTowerGate network according to claim 1, wherein,
the formula of the tower input integration module is:
B (t，k) ＝G k (W t k )⊙E (t) (W t )
wherein E is (t) (W t ) Representing the calculation result of the first stacking module, G K (W t K ) N task inputs representing the division of the Gate computing result, where "" -represents the multiplication of the vector and W=B (t,k) 。
4. The bridge girder erection machine time sequence abnormality detection system based on the ExpertTowerGate network according to claim 1, wherein,
the calculation method of the second stacking module comprises the following steps:
where Tower represents the Tower computation, and the intermediate result of Tower is denoted as T 1 ,…T n N represents the length of the tower calculation vector.
5. The bridge girder erection machine time sequence abnormality detection system based on the ExpertTowerGate network according to claim 3, wherein,
G K (W t K ) Adopts a shared gate G for calculation s Receiving all windows from input, and a personalized doorOnly the kth index is concerned;
the mixed mode calculation formula of the shared gate and the personalized gate is as follows:
wherein Ws and W pk (k=1, 2, …, K) is a trainable parameter, ε>0.5, which is the weight coefficient of the shared gate; g K (W t K ) Representing the results of the shared and personalized gate mix calculations,representing the kth vector at time t.
6. The bridge girder erection machine time sequence abnormality detection system based on the ExpertTowerGate network according to claim 1, wherein,
in the expert module, input is firstly subjected to an unscqueeze unit, an input matrix is deformed to obtain M, then the M is input into a Conv2d convolution unit to obtain N, and then the N is sequentially subjected to a Relu activation unit, a Flatten flattening unit, an FC1 full connection 1 unit, a Relu activation unit and an FC2 full connection 2 unit to obtain output.
7. The bridge girder erection machine time sequence abnormality detection system based on the ExpertTowerGate network according to claim 1, wherein,
in the tower calculation module, the input is firstly subjected to FC1 full connection 1 unit to obtain U, then is subjected to Relu activation unit to obtain V, and finally is subjected to FC2 full connection 2 unit to obtain output.
8. The bridge girder erection machine time sequence anomaly detection method based on the ExpertTowerGate network is characterized in that the detection system is used, the existing sensor data is used for carrying out iterative training on the detection system before the detection system is used, so that the value of a loss function reaches the minimum, then the accuracy is evaluated on a test set or a verification set, the accuracy reaches more than 90%, and an obtained system model is stored as a binary system;
in the training process, the adopted objective function is as follows:
as predicted value, y t Is true value, I.I.I 2 Representing the L2 paradigm calculation.
9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the method steps of claim 8 when the computer program is executed by the processor.
10. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the method steps of claim 8.
CN202311460804.5A 2023-11-06 2023-11-06 Bridge girder erection machine time sequence abnormality detection system and detection method based on ExpertTowerGate network Pending CN117390350A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311460804.5A CN117390350A (en) 2023-11-06 2023-11-06 Bridge girder erection machine time sequence abnormality detection system and detection method based on ExpertTowerGate network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311460804.5A CN117390350A (en) 2023-11-06 2023-11-06 Bridge girder erection machine time sequence abnormality detection system and detection method based on ExpertTowerGate network

Publications (1)

Publication Number Publication Date
CN117390350A true CN117390350A (en) 2024-01-12

Family

ID=89439037

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311460804.5A Pending CN117390350A (en) 2023-11-06 2023-11-06 Bridge girder erection machine time sequence abnormality detection system and detection method based on ExpertTowerGate network

Country Status (1)

Country Link
CN (1) CN117390350A (en)

Similar Documents

Publication Publication Date Title
CN107941537B (en) A kind of mechanical equipment health state evaluation method
CN105139585A (en) Intelligent early warning and prediction method for soil slope danger
CN102520697B (en) Onsite information preprocessing method of remote cooperative diagnosis
CN111274737A (en) Method and system for predicting remaining service life of mechanical equipment
CN104200687B (en) A kind of driver's speed control behavior monitoring device and monitoring method
CN107862331A (en) It is a kind of based on time series and CNN unsafe acts recognition methods and system
CN105846780A (en) Decision tree model-based photovoltaic assembly fault diagnosis method
CN103514366A (en) Urban air quality concentration monitoring missing data recovering method
CN104675380B (en) Oil drilling drill string on-line monitoring system and method for diagnosing faults
CN104900061A (en) Road section travel time monitoring method and device
CN105243393A (en) Characteristic-based fault forecasting method for complex electromechanical system
CN115022187B (en) Situation awareness method and device for electric-gas comprehensive energy system
CN114169036A (en) Wind vibration response early warning system and method for large-span bridge
CN111122811A (en) Sewage treatment process fault monitoring method of OICA and RNN fusion model
Li et al. Tailings pond risk prediction using long short-term memory networks
CN115129030A (en) Unmanned aerial vehicle fault diagnosis system based on combination of model and neural network
CN202373020U (en) Intelligent coach operation system realizing device based on cloud computing platform
CN109406076A (en) A method of beam bridge structure damage reason location is carried out using the mobile principal component of displacement sensor array output
CN117191147A (en) Flood discharge dam water level monitoring and early warning method and system
CN113538971A (en) Risk ship detection method based on multi-source heterogeneous data fusion
CN112287586A (en) Rapid reliability assessment method based on medium and small bridge damage identification result
CN117390350A (en) Bridge girder erection machine time sequence abnormality detection system and detection method based on ExpertTowerGate network
Łęczycki et al. Extended sensor reliability evaluation method in multi-sensor control systems
CN111882135B (en) Internet of things equipment intrusion detection method and related device
CN111476383A (en) Pump station unit state maintenance dynamic decision method

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