CN112287602B - Motor train axle temperature fault early warning method based on machine learning and isolated forest - Google Patents
Motor train axle temperature fault early warning method based on machine learning and isolated forest Download PDFInfo
- Publication number
- CN112287602B CN112287602B CN202011176557.2A CN202011176557A CN112287602B CN 112287602 B CN112287602 B CN 112287602B CN 202011176557 A CN202011176557 A CN 202011176557A CN 112287602 B CN112287602 B CN 112287602B
- Authority
- CN
- China
- Prior art keywords
- axle temperature
- motor train
- temperature
- train axle
- lstm
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Geometry (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Electric Propulsion And Braking For Vehicles (AREA)
Abstract
The invention relates to the field of axle temperature early warning, and discloses a motor train axle temperature fault early warning method based on machine learning and isolated forests, which comprises a modeling process and a timing scheduling process; the modeling process comprises the steps of establishing a motor train axle temperature model based on LSTM and an axle temperature anomaly detection model based on iForest; the timed scheduling process comprises the steps of predicting the axle temperature of the bullet train in a future preset time length by using an LSTM-based bullet train axle temperature model; and calling an iForest-based shaft temperature anomaly detection model, and performing anomaly detection on the predicted shaft temperature by using the iForest-based shaft temperature anomaly detection model. The invention provides a novel axle temperature early warning method based on the combination of LSTM and iForest, which can predict and detect the axle temperature state within the future preset time, if abnormity occurs, overhaul is intervened in advance, the safe operation of a vehicle within the future preset time is ensured, and the fault during the real-time operation of the vehicle is avoided.
Description
Technical Field
The invention relates to the field of axle temperature early warning, in particular to a motor train axle temperature fault early warning method based on machine learning and isolated forest.
Background
The existing axle temperature early warning technology is characterized in that an alarm is given based on a set threshold value, and once the alarm occurs, a vehicle operation accident is caused; and secondly, the current value of the shaft temperature is predicted based on a prediction algorithm, but the mode can only achieve the real-time detection of the shaft temperature state and cannot achieve the purpose of prediction.
For example, the national patent publication CN106627659A discloses a novel axle temperature monitoring system and a control method for rail vehicles, wherein an axle temperature alarm host is arranged in each carriage of a motor train unit, each axle temperature alarm host is connected with a temperature sensor, the temperature sensors are arranged at each axle end and used for collecting data of each axle temperature, an a/D conversion circuit is integrated in the axle temperature alarm host of each vehicle and used for monitoring temperatures uploaded by each temperature sensor at each axle end, the axle temperature alarm hosts are communicated by an FSK bus, and the axle temperature alarm hosts of the two end vehicles are connected with an on-vehicle control unit. According to the invention, the shaft temperature is monitored in real time through the temperature sensor, but the prediction and state detection on the shaft temperature at the future time cannot be realized, if the shaft temperature is abnormal, the operation of the whole system is influenced, the early intervention of maintenance cannot be realized, the safe operation of the vehicle at the future time cannot be ensured, and the fault in the real-time operation of the vehicle cannot be avoided.
Disclosure of Invention
The invention provides a motor car axle temperature fault early warning method based on machine learning and isolated forests, and aims to solve the problems in the prior art.
A motor car axle temperature fault early warning method based on machine learning and isolated forest comprises a modeling process and a timing scheduling process;
the modeling process comprises the steps of establishing a motor train axle temperature model based on LSTM and establishing an axle temperature anomaly detection model based on iForest;
the timing scheduling process comprises the following steps:
s1) acquiring historical motor train axle temperature data of train operation within a preset time, executing an LSTM-based motor train axle temperature model at regular time, inputting the historical motor train axle temperature data into the LSTM-based motor train axle temperature model, and predicting the motor train axle temperature within the future preset time by using the LSTM-based motor train axle temperature model to obtain the predicted axle temperature of the motor train within the future preset time;
and S2) calling an iForest-based shaft temperature anomaly detection model, carrying out anomaly detection on the predicted shaft temperature by using the iForest-based shaft temperature anomaly detection model to obtain an anomaly detection result, and carrying out motor train shaft temperature fault early warning according to the anomaly detection result.
Further, establishing a motor train axle temperature model based on the LSTM, which comprises the following steps:
a1 Collecting a historical sample set of the axle temperature of the motor train, establishing a motor train axle temperature model formed by a plurality of LSTM network units, and initializing the motor train axle temperature model formed by the plurality of LSTM network units;
a2 Proportionally dividing a historical motor train axle temperature sample set into a training sample set and a testing sample set, and respectively training and testing the motor train axle temperature model formed by the LSTM network units by using the training sample set and the testing sample set;
a3 Obtain a trained LSTM-based axle temperature model of the railcar.
Further, in the step S1), cross validation is carried out on the motor train axle temperature model formed by the plurality of LSTM network units by utilizing a historical motor train axle temperature sample set, and the motor train axle temperature model formed by the plurality of LSTM network units is continuously optimized through the cross validation, so that the finally trained motor train axle temperature model based on the LSTM achieves the optimal model.
Furthermore, the LSTM network unit comprises a plurality of states, wherein the states comprise a hidden layer state, a forgetting door state, a cell output state, a middle cell state, an input door state and an output door state, and the historical motor train axle temperature sample data input of the LSTM network unit at the t moment is x t The hidden layer state is h t The hidden layer state at the previous moment is h t-1 After historical motor train axle temperature sample data is transmitted forwards through the LSTM network unit, the door state f is forgotten t ＝δ(W f ·[h t-1 ,x t ]+d f ) Input door status i t ＝δ(W i ·[h t-1 ,x t ]+d i ) Cell output status
Intermediate cell status->
Output gate state o t ＝δ(W o ·[h t-1 ,x t ]+d o ) The output value h of the LSTM network unit at time t t ＝o t ·tanh(C t ) (ii) a Wherein, tanh is a hyperbolic tangent function, and delta is a sigmoid function; w is a group of f 、W i 、W C 、W o Are weight matrices of the LSTM network, d f 、d i 、d C 、d o Respectively, the deviation vectors of the LSTM network.
Further, in the step S1), historical motor train axle temperature data are input into the LSTM-based motor train axle temperature model, the LSTM-based motor train axle temperature model is used for predicting the motor train axle temperature in the future preset time, historical motor train axle temperature sample data are collected to be one-dimensional axle temperature data, feature transformation is carried out on the one-dimensional axle temperature data, the one-dimensional axle temperature data are changed to be M-dimensional axle temperature data, M is the change period of the selected motor train axle temperature, the prediction of the motor train axle temperature in the future preset time is completed by the LSTM-based motor train axle temperature model, the number of predicted state quantities of the motor train axle temperature is N, the motor train axle temperature values at Q moments are predicted each time, and N-time M-to-Q-dimensional axle temperature data mapping from M to Q-dimensional is formed.
Further, establishing an iForest-based shaft temperature anomaly detection model, which comprises the following steps:
b1 Obtaining a historical normal sample set of the motor train axle temperature data, wherein the historical normal sample set of the motor train axle temperature data comprises a plurality of normal motor train axle temperature samples;
b2 Constructing an exclusive forest iForest which comprises w isolated trees iTree, wherein each isolated tree iTree is of a binary tree structure and comprises a plurality of nodes, the first node is a root node, and other nodes except the root node are inherited nodes in sequence;
b3 Randomly selecting m samples from a historical motor train axle temperature data normal sample set as sub-samples of an ith isolated tree iTree, and placing the sub-samples of the ith isolated tree iTree into a root node of the ith isolated tree iTree; i =1, 2, ..., w;
b4 Randomly generating a cutting point p in the current node, wherein the value range of the cutting point p is between the maximum value and the minimum value of the normal sample of the historical motor train axle temperature data in the current node;
b5 Generating a hyperplane through a cutting point p, judging whether the value of the jth sample in the subsamples of the ith isolated tree iTree is smaller than the cutting point p, and if so, putting the sample of which the historical motor train axle temperature data is smaller than the cutting point p in the subsamples of the ith isolated tree iTree into the left inheritance of the current node; if not, putting a sample of which the historical motor car axle temperature data is greater than or equal to the cutting point p in the sub-samples of the ith isolated tree iTree into the current node and putting the current node in the right inheritance of the current node;
b6 Setting the limit height of the ith isolated tree iTree, and continuously recursing step S34) and step S35) in the inheritance node until the subsample of the ith isolated tree iTree in the inheritance node is not subdivided or reaches the limit height of the ith isolated tree iTree, and stopping recursion;
b7 W isolated trees iTree are constructed in sequence to form an exclusive forest iForest.
Further, in the step S2), calling an iForest-based shaft temperature anomaly detection model, and performing anomaly detection on the predicted shaft temperature by using the iForest-based shaft temperature anomaly detection model, including the following steps:
s21) recording the predicted shaft temperature at the t-th moment as y t The predicted shaft temperature y at the t-th moment t Traversing each isolated tree iTree to obtain the predicted shaft temperature y at the t moment t Respectively positioned at the height of each isolated tree iTree and used for predicting the shaft temperature y according to the t-th moment t The predicted shaft temperature y at the t-th moment is obtained by the height of each isolated tree iTree t Height average h (y) at each isolated tree iTree t ) Calculating the predicted shaft temperature y at the t-th moment t Is abnormal probability of
m is the number of samples, and the expression of c (m) is->
ζ is the Euler constant;
s39) setting an abnormal threshold value, and judging the predicted shaft temperature y at the t-th moment t If the abnormal probability exceeds the abnormal threshold, the predicted shaft temperature y at the t-th moment is determined t Judging as an abnormal value, if not, judging the predicted shaft temperature y at the t-th time t The value is determined as a normal value.
The invention has the beneficial effects that: the invention is divided into a modeling process and a timing scheduling process. Establishing a motor train axle temperature model based on LSTM by using a historical motor train axle temperature sample set, and establishing an axle temperature anomaly detection model based on iForest by using a historical motor train axle temperature data normal sample set; the timed scheduling process is that after the operation of the train is finished every day, the LSTM-based shaft temperature model of the motor train is executed regularly to predict the shaft temperature in the preset time length in the future for the data of the day, and then the iForest-based shaft temperature abnormity detection model is called to perform abnormity detection on the predicted shaft temperature. The invention provides a novel axle temperature early warning method based on the combination of LSTM and iForest, which is used for predicting and detecting the axle temperature state within the future preset time, and if abnormity occurs, the overhaul is intervened in advance, the safe operation of a vehicle within the future preset time is ensured, and the fault during the real-time operation of the vehicle is avoided.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments are briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
Fig. 1 is a schematic flow chart of a motor train axle temperature fault early warning method based on machine learning and isolated forest according to this embodiment.
FIG. 2 is a comparison graph of the prediction using the LSTM-based axle temperature model of the motor vehicle according to the first embodiment.
Fig. 3 is a graph of the axle temperature trend provided in the first embodiment.
Fig. 4 is an analysis diagram of the prediction result based on the LSTM algorithm and the detection result based on the iForest anomaly detection algorithm provided in this embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings. It should be understood that the detailed description and specific examples, while indicating the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.
In the first embodiment, a motor train axle temperature fault early warning method based on machine learning and isolated forest is shown in fig. 1 and comprises a modeling process and a timing scheduling process;
the modeling process comprises the steps of establishing a motor train axle temperature model based on LSTM and establishing an axle temperature anomaly detection model based on iForest;
the method for establishing the motor car axle temperature model based on the LSTM comprises the following steps:
a1 Collecting a historical motor train axle temperature sample set, establishing a motor train axle temperature model formed by a plurality of LSTM network units, and initializing the motor train axle temperature model formed by the plurality of LSTM network units;
a2 Proportionally dividing a historical motor train axle temperature sample set into a training sample set and a test sample set, and respectively training and testing a motor train axle temperature model consisting of a plurality of LSTM network units by utilizing the training sample set and the test sample set;
a3 Obtain a trained LSTM-based axle temperature model of the railcar.
The method for establishing the shaft temperature anomaly detection model based on the iForest comprises the following steps:
b1 Obtaining a historical motor train axle temperature data normal sample set, wherein the historical motor train axle temperature data normal sample set comprises a plurality of normal motor train axle temperature samples;
b2 Constructing an exclusive forest iForest which comprises w isolated trees iTree, wherein each isolated tree iTree is of a binary tree structure and comprises a plurality of nodes, the first node is a root node, and other nodes except the root node are inherited nodes in sequence;
b3 Randomly selecting m samples from a historical motor train axle temperature data normal sample set as sub-samples of an ith isolated tree iTree, and placing the sub-samples of the ith isolated tree iTree into a root node of the ith isolated tree iTree; i =1, 2, ..., w;
b4 Randomly generating a cutting point p in the current node, wherein the value range of the cutting point p is between the maximum value and the minimum value of the normal sample of the historical motor train axle temperature data in the current node;
b5 Generating a hyperplane through a cutting point p, judging whether the value of the jth sample in the subsamples of the ith isolated tree iTree is smaller than the cutting point p, and if so, putting the sample of which the historical motor train axle temperature data is smaller than the cutting point p in the subsamples of the ith isolated tree iTree into the left inheritance of the current node; if not, putting a sample of which the historical motor car axle temperature data is greater than or equal to the cutting point p in the sub-samples of the ith isolated tree iTree into the current node and putting the current node in the right inheritance of the current node;
b6 Setting the limit height of the ith isolated tree iTree, and continuously recursing step S34) and step S35) in the inheritance node until the subsample of the ith isolated tree iTree in the inheritance node is not subdivided or reaches the limit height of the ith isolated tree iTree, and stopping recursion;
b7 W isolated trees iTree are constructed in sequence to form an exclusive forest iForest.
The iForest (Isolation Forest) isolated Forest is the most effective in the current anomaly detection algorithm, and can realize accurate identification of the anomaly data. Due to the real-time nature of the detection to be achieved, the more closely the features are with data closer to the current value, the more accurate the monitoring results for anomalous samples. Therefore, in this embodiment, the axle temperature data 24 hours of the day is selected as the training sample for the anomaly detection, the training sample needs to be subjected to data screening, and all normal sample sets are selected for modeling, that is, the data needs to be cleaned when the axle temperature anomaly detection model based on iForest is trained. And adjusting the model to be optimal through testing the sample, namely identifying the normal state as normal.
The timing scheduling process comprises the following steps:
s1) acquiring historical motor car axle temperature data of train operation in a preset time, executing a motor car axle temperature model based on LSTM at regular time, inputting the historical motor car axle temperature data into the motor car axle temperature model based on LSTM, and predicting the motor car axle temperature in the preset time in the future by using the motor car axle temperature model based on LSTM to obtain the predicted axle temperature of the motor car in the preset time in the future;
in the step S1), the method further comprises the steps of carrying out cross validation on the motor train axle temperature model formed by the plurality of LSTM network units by utilizing a historical motor train axle temperature sample set, and continuously optimizing the motor train axle temperature model formed by the plurality of LSTM network units through the cross validation, so that the finally trained motor train axle temperature model based on the LSTM achieves the optimal model.
The LSTM network unit comprises a plurality of states including a hidden layer state, a forgetting gate state, a cell output state, an intermediate cell state, an input gate state, an output gate state, and an orderThe historical motor train axle temperature sample data input of the LSTM network unit at the t-th moment is x t The hidden layer state is h t The hidden layer state at the previous moment is h t-1 After the historical sample data of the axle temperature of the motor train car is transmitted forwards through the LSTM network unit, the door state f is forgotten t ＝δ(W f ·[h t-1 ,x t ]+d f ) Input door status i t ＝δ(W i ·[h t-1 ,x t ]+d i ) Cell output status
Intermediate cell state
Output gate state o t ＝δ(W o ·[h t-1 ,x t ]+d o ) The output value h of the LSTM network unit at time t t ＝o t ·tanh(C t ) (ii) a Wherein, tanh is a hyperbolic tangent function, and delta is a sigmoid function; w f 、W i 、W C 、W o Are weight matrices of the LSTM network, d f 、d i 、d C 、d o Respectively, the offset vectors of the LSTM network.
In the step S1), historical motor train axle temperature data are input into a motor train axle temperature model based on LSTM, the motor train axle temperature in the future preset duration is predicted by the aid of the motor train axle temperature model based on LSTM, historical motor train axle temperature sample data are collected into one-dimensional axle temperature data, feature transformation is conducted on the one-dimensional axle temperature data, the one-dimensional axle temperature data are changed into M-dimensional axle temperature data, M is the change period of the selected motor train axle temperature, the motor train axle temperature in the future preset duration is predicted by the aid of the motor train axle temperature model based on LSTM, the number of predicted state quantities of the motor train axle temperature is N, motor train axle temperature values at Q moments are predicted each time, and N-time M-dimensional to Q-dimensional axle temperature data mapping is formed.
The method comprises the steps of detecting the axle temperature running condition according to minutes from the actual running condition of the motor train unit, comparing the detected axle temperature running condition according to the minutes, and conforming to the actual condition, namely if the axle temperature of the motor train unit in the next day is predicted, then N =1440, predicting the axle temperature running condition of the next day by using the data of the previous day, if Q = N, directly predicting the axle temperature state of the next day by using the axle temperature data of the current day, waiting for the motor train unit to run completely in the current day, obtaining the data of the motor train unit in the current day, then performing model prediction at regular time, training a motor train axle temperature model based on LSTM in the whole modeling process, comparing the predicted value of the motor train axle temperature model based on LSTM with the actual axle temperature running condition, calculating a loss function, and continuously optimizing the model. All the historical data (including normal values and abnormal values) detected by the axle temperature are added into the LSTM-based motor car axle temperature model for training.
S2) calling an iForest-based shaft temperature abnormity detection model, carrying out abnormity detection on the predicted shaft temperature by using the iForest-based shaft temperature abnormity detection model to obtain an abnormity detection result, and carrying out bullet train shaft temperature fault early warning according to the abnormity detection result.
In the step S2), an axle temperature abnormity detection model based on iForest is called, and abnormity detection is carried out on the predicted axle temperature by using the axle temperature abnormity detection model based on iForest, and the method comprises the following steps:
s21) recording the predicted shaft temperature at the t-th time as y t The predicted shaft temperature y at the t-th moment t Traversing each isolated tree iTree to obtain the predicted shaft temperature y at the t moment t Respectively positioned at the height of each isolated tree iTree and used for predicting the shaft temperature y according to the t-th moment t The predicted shaft temperature y at the t-th moment is obtained by the height of each isolated tree iTree t Height average h (y) at each isolated tree iTree t ) Calculating the predicted shaft temperature y at the t-th time t Is abnormal probability of
m is the number of samples, and the expression of c (m) is->
ζ is the Euler constant;
s39) setting an abnormal threshold value, and judging the predicted shaft temperature y at the t-th moment t If the abnormal probability exceeds the abnormal threshold, the predicted shaft temperature y at the t-th moment is determined t Judging as an abnormal value, if not, then judging the predicted shaft temperature y at the t-th time t The value was determined to be normal.
In the embodiment, step S1) collects historical motor train axle temperature data of train operation in a preset time period as historical motor train axle temperature data of train operation in the same day, a motor train axle temperature model based on LSTM is used for predicting motor train axle temperature data in the next day to obtain a predicted value of motor train axle temperature in the next day, an axle temperature anomaly detection model (namely an optimal anomaly detection model) based on iForest after training is used for performing anomaly judgment on the predicted value of motor train axle temperature in the next day, and if the predicted value of motor train axle temperature is abnormal, an alarm is given.
The method is mainly characterized in that the abnormal detection is carried out on the state of the bearing shaft temperature of the motor train unit in the next day in advance, and the method mainly comprises the following two steps:
the method comprises the steps that the axle temperature value of the motor train unit is predicted in the future day, on the premise that the motor train unit is required to be parked and put in storage in the same day, collected real-time data are transmitted to a ground PHM system, then the axle temperature data of the motor train unit in the same day are scheduled at regular time according to an optimized LSTM-based motor train axle temperature model to execute tasks, and the axle temperature operation trend of the motor train unit in the future day is predicted.
And the second step is to carry out abnormity detection on all the predicted values of the shaft temperature. After the prediction result obtained by scheduling and executing the prediction model is obtained, a timing calculation task is started for the predicted value of the next day according to the optimized iForest-based abnormity detection model, the state evaluation of the shaft temperature of the next day is completed, early warning is carried out if abnormity occurs, related maintenance personnel are informed to carry out professional evaluation, and whether further maintenance is carried out is judged.
In this embodiment, data from 1/2020 to 9/10/2020 of CRH380AL-XXXX8 carriage 1-bit axle box temperature is sampled and selected as a training sample set, and a model cross validation training set and a test set are divided according to a 0.7 ratio. Samples No. 9/12 in 2020 and 9/8 of CRH380 AL-XXXXXX 8 carriage 1-bit axle box temperature are selected as test sample sets, and the axle temperature actually generates abnormality at 8 points 42. And respectively training and testing the LSTM-based motor train axle temperature model by utilizing the training sample set and the testing sample set to obtain the LSTM-based motor train axle temperature model after training.
In this embodiment, a trained motor train axle temperature model based on LSTM is used to predict the axle temperature of a motor train of No. 8 months and No. 20, and the predicted value and the actual value are, for example, as shown in fig. 2, a is a predicted data curve of the axle temperature of the motor train, and b is an actual data curve of the axle temperature of the motor train. Therefore, the LSTM-based motor train axle temperature model can predict normal samples and abnormal samples accurately, and can realize real-time accurate prediction of the axle temperature.
In this embodiment, the actual failure occurs at 42 minutes 8 o' clock on 12 d 9 month, and the specific axle temperature trend graph is shown in fig. 3. Fig. 4 shows the predicted result of the LSTM-based axle temperature model of the motor train and the detected result of the iForest-based axle temperature abnormality detection model, where e is the predicted result of the LSTM-based axle temperature model of the motor train, f is the detected result of the ifoest-based axle temperature abnormality detection model, and in the detected result of the ifoest-based axle temperature abnormality detection model, the normal value of the axle temperature of the motor train is 0 on 12 days in 9 months, and 8 abnormal values are detected at 42 points on 12 days in 9 months, and are marked as 10 by the ifoest-based axle temperature abnormality detection model. Therefore, the invention can detect the abnormality of the shaft temperature one day in advance and can early warn when abnormality occurs.
The invention realizes the advanced prediction of the shaft temperature in the future preset duration by utilizing the deep learning LSTM algorithm, overcomes the defect that the current prediction can only predict the current value in the prior art, and has practical application value; in addition, the method utilizes the random forest algorithm to realize the dynamic sliding window type updated algorithm every day, realizes the model self-learning technology, and has innovativeness and generalization. The invention integrates the deep learning technology and the anomaly detection technology, so that the new method has higher accuracy; the invention creatively realizes the prediction and maintenance in the true sense, and can predict the index state within the future preset time and carry out the abnormal detection. The novel motor train unit shaft temperature early warning model provided by the invention has general universality for solving the motor train unit equipment failure.
Because the original data of the motor train unit has few data which really have faults, the modeling can be generally carried out only on the shaft temperature data of the motor train unit when the motor train unit normally operates. According to the method, the predicted shaft temperature abnormity detection of the motor train unit bearing is realized by using the advantages of an unsupervised abnormity detection technology (establishing an iForest-based shaft temperature abnormity detection model) and deep learning and mining of big data, the prediction accuracy is high, the safe operation of the vehicle within the preset time in the future can be ensured, and the fault in the real-time operation of the vehicle is avoided.
By adopting the technical scheme disclosed by the invention, the following beneficial effects are obtained:
the invention is divided into a modeling process and a timing scheduling process. Establishing a motor train axle temperature model based on LSTM by using a historical motor train axle temperature sample set, and establishing an axle temperature anomaly detection model based on iForest by using a historical motor train axle temperature data normal sample set; the timed scheduling process is that after the operation of the train is finished every day, the LSTM-based shaft temperature model of the motor train is executed regularly to predict the shaft temperature in the preset time length in the future for the data of the day, and then the iForest-based shaft temperature abnormity detection model is called to perform abnormity detection on the predicted shaft temperature. The invention provides a novel axle temperature early warning method based on the combination of LSTM and iForest, which is used for predicting and detecting the axle temperature state within the future preset time, and if abnormity occurs, the overhaul is intervened in advance, the safe operation of a vehicle within the future preset time is ensured, and the fault during the real-time operation of the vehicle is avoided.
The above is only a preferred embodiment of the present invention, and it should be noted that it is obvious to those skilled in the art that a plurality of modifications and embellishments can be made without departing from the principle of the present invention, and these modifications and embellishments should also be viewed as the protection scope of the present invention.
Claims (4)
1. A motor car axle temperature fault early warning method based on machine learning and isolated forest is characterized by comprising a modeling process and a timing scheduling process;
the modeling process comprises the steps of establishing a motor train axle temperature model based on LSTM and establishing an axle temperature anomaly detection model based on iForest;
the timing scheduling process comprises the following steps:
s1) acquiring historical motor car axle temperature data of train operation in a preset time, executing a motor car axle temperature model based on LSTM at regular time, inputting the historical motor car axle temperature data into the motor car axle temperature model based on LSTM, and predicting the motor car axle temperature in the preset time in the future by using the motor car axle temperature model based on LSTM to obtain the predicted axle temperature of the motor car in the preset time in the future;
s2) calling an iForest-based shaft temperature anomaly detection model, carrying out anomaly detection on the predicted shaft temperature by using the iForest-based shaft temperature anomaly detection model to obtain an anomaly detection result, and carrying out motor train shaft temperature fault early warning according to the anomaly detection result;
step S1), inputting historical motor train axle temperature data into the LSTM-based motor train axle temperature model, predicting the motor train axle temperature in a future preset time length by using the LSTM-based motor train axle temperature model, performing characteristic transformation on the one-dimensional axle temperature data by using a historical motor train axle temperature sample set as one-dimensional axle temperature data, converting the one-dimensional axle temperature data into M-dimensional axle temperature data, wherein M is a change period of the selected motor train axle temperature, completing prediction on the motor train axle temperature in the future preset time length by using the LSTM-based motor train axle temperature model, the number of predicted state quantities of the motor train axle temperature is N, predicting the motor train axle temperature value at Q moments each time, and forming N times of M-dimensional to Q-dimensional axle temperature data mapping;
the method for establishing the shaft temperature anomaly detection model based on the iForest comprises the following steps:
b1 Obtaining a historical motor train axle temperature data normal sample set, wherein the historical motor train axle temperature data normal sample set comprises a plurality of normal motor train axle temperature samples;
b2 Constructing an exclusive forest iForest which comprises w isolated trees iTree, wherein each isolated tree iTree is of a binary tree structure and comprises a plurality of nodes, the first node is a root node, and other nodes except the root node are inherited nodes in sequence;
b3 Randomly selecting m samples from the historical motor train axle temperature data normal sample set as sub-samples of an ith isolated tree iTree, and placing the sub-samples of the ith isolated tree iTree into a root node of the ith isolated tree iTree; i =1, 2, · w;
b4 Randomly generating a cutting point p in the current node, wherein the value range of the cutting point p is between the maximum value and the minimum value of the normal sample of the historical motor train axle temperature data in the current node;
b5 Generating a hyperplane through a cutting point p, judging whether the value of the jth sample in the subsamples of the ith isolated tree iTree is smaller than the cutting point p, and if so, putting the sample of which the historical motor train axle temperature data is smaller than the cutting point p in the subsamples of the ith isolated tree iTree into the left inheritance of the current node; if not, putting a sample of which the historical motor car axle temperature data is greater than or equal to the cutting point p in the sub-samples of the ith isolated tree iTree into the current node and putting the current node in the right inheritance of the current node;
b6 Setting the limit height of the ith isolated tree iTree, and continuously recursing step S34) and step S35) in the inheritance node until the subsample of the ith isolated tree iTree in the inheritance node is not subdivided or reaches the limit height of the ith isolated tree iTree, and stopping recursion;
b7 W isolated trees iTrees are constructed in sequence to form an exclusive forest IForest;
in the step S2), an iForest-based shaft temperature anomaly detection model is called, and the predicted shaft temperature is subjected to anomaly detection by using the iForest-based shaft temperature anomaly detection model, and the method comprises the following steps:
s21) recording the predicted shaft temperature at the t-th time as y t The predicted shaft temperature y at the t-th moment t Traversing each isolated tree iTree to obtain the predicted shaft temperature y at the t moment t Respectively positioned at the height of each isolated tree iTree and predicted shaft temperature y at the t-th moment t The predicted shaft temperature y at the t-th moment is obtained by the height of each isolated tree iTree t Height average h (y) at each orphan tree iTree t ) Calculating the prediction of the t-th timeMeasuring shaft temperature y t Is abnormal probability of
m is the number of samples, and the expression of c (m) is->
ζ is the Euler constant;
s22) setting an abnormal threshold value, and judging the predicted shaft temperature y at the t-th moment t If the abnormal probability exceeds the abnormal threshold, the predicted shaft temperature y at the t-th moment is determined t Judging as an abnormal value, if not, then judging the predicted shaft temperature y at the t-th time t The value is determined as a normal value.
2. The motor train axle temperature fault early warning method based on machine learning and isolated forest as claimed in claim 1, wherein the establishing of the LSTM-based motor train axle temperature model comprises the following steps:
a1 Collecting a historical motor train axle temperature sample set, establishing a motor train axle temperature model formed by a plurality of LSTM network units, and initializing the motor train axle temperature model formed by the plurality of LSTM network units;
a2 Proportionally dividing the historical motor train axle temperature sample set into a training sample set and a test sample set, and respectively training and testing the motor train axle temperature model formed by the plurality of LSTM network units by utilizing the training sample set and the test sample set;
a3 Obtain a trained LSTM-based axle temperature model of the railcar.
3. The machine learning and isolated forest-based bullet train axle temperature fault early warning method according to claim 2, wherein in step S1), the method further comprises cross-verifying the bullet train axle temperature model composed of a plurality of LSTM network units by using the historical bullet train axle temperature sample data set, and optimizing the bullet train axle temperature model composed of a plurality of LSTM network units by cross-verifying, so that the finally trained LSTM-based bullet train axle temperature model is optimal.
4. The motor train axle temperature fault early warning method based on machine learning and isolated forest as claimed in claim 2 or 3, wherein the LSTM network unit comprises a plurality of states, the plurality of states comprise a hidden layer state, a forgetting gate state, a cell output state, an intermediate cell state, an input gate state, an output gate state, and the historical motor train axle temperature sample data input of the LSTM network unit at the t-th moment is x t The hidden layer state is h t The hidden layer state at the previous moment is h t-1 After historical motor car axle temperature sample data is transmitted forwards through the LSTM network unit, a door state ft = delta (W) is forgotten f ·[h t-1 ，x t ]+d f ) Input door status i t ＝δ(W i ·[h t-1 ，x t ]+d i ) Cell output status
Intermediate cell state
Output gate state o t ＝δ(W o ·[h t-1 ，x t ]+d o ) The output value h of the LSTM network unit at time t t ＝o t ·tanh(C t ) (ii) a Wherein, tanh is a hyperbolic tangent function, and delta is a sigmoid function; w is a group of f 、W i 、W C 、W o Are weight matrices of the LSTM network, d f 、d i 、d C 、d o Respectively, the offset vectors of the LSTM network. />
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011176557.2A CN112287602B (en) | 2020-10-28 | 2020-10-28 | Motor train axle temperature fault early warning method based on machine learning and isolated forest |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011176557.2A CN112287602B (en) | 2020-10-28 | 2020-10-28 | Motor train axle temperature fault early warning method based on machine learning and isolated forest |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112287602A CN112287602A (en) | 2021-01-29 |
| CN112287602B true CN112287602B (en) | 2023-04-18 |
Family
ID=74373794
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011176557.2A Active CN112287602B (en) | 2020-10-28 | 2020-10-28 | Motor train axle temperature fault early warning method based on machine learning and isolated forest |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112287602B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113076975A (en) * | 2021-03-17 | 2021-07-06 | 长江水利委员会长江科学院 | Dam safety monitoring data anomaly detection method based on unsupervised learning |
| CN114139591A (en) * | 2021-06-23 | 2022-03-04 | 核动力运行研究所 | Mechanical equipment state early warning method |
| CN114219999B (en) * | 2021-11-05 | 2024-03-19 | 华南理工大学 | Automatic machine vision monitoring method and system for preventing external decorative plate of structure from falling off |
| CN114169237B (en) * | 2021-11-30 | 2024-05-03 | 南昌大学 | Power cable joint temperature abnormality early warning method combining EEMD-LSTM and isolated forest algorithm |
| CN115424368A (en) * | 2022-08-25 | 2022-12-02 | 武汉迪昌科技有限公司 | Unpowered grouping test method and device for motor train unit |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108900546A (en) * | 2018-08-13 | 2018-11-27 | 杭州安恒信息技术股份有限公司 | The method and apparatus of time series Network anomaly detection based on LSTM |
| CN110442600A (en) * | 2019-04-17 | 2019-11-12 | 江苏网谱数据服务有限公司 | A kind of time series method for detecting abnormality |
| CN110210174A (en) * | 2019-06-21 | 2019-09-06 | 西南交通大学 | A kind of EMU axle box bearing temperature predicting method based on multilayer LSTM |
| CN110243599B (en) * | 2019-07-02 | 2020-05-05 | 西南交通大学 | Method for monitoring temperature abnormal state of multi-dimensional outlier train motor train unit axle box bearing |
| CN110619182A (en) * | 2019-09-24 | 2019-12-27 | 长沙理工大学 | Power transmission line parameter identification and power transmission network modeling method based on WAMS big data |
| CN111738520A (en) * | 2020-06-24 | 2020-10-02 | 中国电子科技集团公司第二十八研究所 | System load prediction method fusing isolated forest and long-short term memory network |
-
2020
- 2020-10-28 CN CN202011176557.2A patent/CN112287602B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112287602A (en) | 2021-01-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112287602B (en) | Motor train axle temperature fault early warning method based on machine learning and isolated forest | |
| CN112763929B (en) | Method and device for predicting health of battery monomer of energy storage power station system | |
| CN110414155B (en) | Fan component temperature abnormity detection and alarm method with single measuring point | |
| CN116757534B (en) | Intelligent refrigerator reliability analysis method based on neural training network | |
| US10739736B2 (en) | Apparatus and method for event detection and duration determination | |
| CN109034191B (en) | ELM-based one-dimensional telemetry data abnormal interpretation method | |
| CN111504676A (en) | Equipment fault diagnosis method, device and system based on multi-source monitoring data fusion | |
| CN110570013B (en) | Single-station online wave period data prediction diagnosis method | |
| CN112149877B (en) | Multi-source data driven fault prediction method and system for multi-element complex urban power grid | |
| CN115809405A (en) | Fan main shaft gear box temperature anomaly detection method based on multi-feature fusion | |
| CN115958990A (en) | V2G charging pile information diagnosis and safety protection method | |
| CN114036647A (en) | Power battery safety risk assessment method based on real vehicle data | |
| CN114281054A (en) | LSTM-CNN-based airplane remote fault analysis method and system | |
| CN116907772A (en) | Self-diagnosis and fault source identification method and system of bridge structure monitoring sensor | |
| CN110243598B (en) | Train bearing temperature processing method and device and storage medium | |
| CN116451322A (en) | Bayesian optimization-based LSTM deep learning network mechanical prediction method | |
| CN114492636B (en) | Transformer winding state signal acquisition system | |
| CN114997748A (en) | New energy automobile operation safety risk prediction method and system based on model fusion | |
| Xu et al. | Prediction of Lithium-ion Battery Remaining Useful Life Based on Long Short Term Memory | |
| Bai et al. | Remaining Useful Life Prediction of Wheel of Heavy-duty Railway Train based on Dual Channel Multi-scale Deep convolution Multi-scale Deep Long Short-Term Memory network | |
| CN117668579A (en) | MEST-based multi-rotor unmanned aerial vehicle fault diagnosis method, equipment and storage medium | |
| CN114638039B (en) | Structural health monitoring characteristic data interpretation method based on low-rank matrix recovery | |
| CN118154174A (en) | Intelligent operation and maintenance cloud platform for industrial equipment | |
| CN117454221A (en) | Abnormality identification method for multi-sensor fusion of high-speed rail traction braking system | |
| Huang et al. | A Maintenance Decision Framework based on Multi-task Learning from the Perspective of Application |
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 |