CN117290726A - CAE-BiLSTM-based fault early warning method for mobile equipment - Google Patents
CAE-BiLSTM-based fault early warning method for mobile equipment Download PDFInfo
- Publication number
- CN117290726A CN117290726A CN202311237531.8A CN202311237531A CN117290726A CN 117290726 A CN117290726 A CN 117290726A CN 202311237531 A CN202311237531 A CN 202311237531A CN 117290726 A CN117290726 A CN 117290726A
- Authority
- CN
- China
- Prior art keywords
- data
- cae
- original data
- bilstm
- rmse
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000012544 monitoring process Methods 0.000 claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 46
- 238000012545 processing Methods 0.000 claims abstract description 16
- 238000007781 pre-processing Methods 0.000 claims abstract description 11
- 238000011176 pooling Methods 0.000 claims description 27
- 230000006870 function Effects 0.000 claims description 24
- 239000011159 matrix material Substances 0.000 claims description 23
- 230000008569 process Effects 0.000 claims description 19
- 230000004913 activation Effects 0.000 claims description 12
- 238000013528 artificial neural network Methods 0.000 claims description 10
- 238000012216 screening Methods 0.000 claims description 8
- 230000007246 mechanism Effects 0.000 claims description 7
- 230000009467 reduction Effects 0.000 claims description 7
- 230000002441 reversible effect Effects 0.000 claims description 7
- 230000001413 cellular effect Effects 0.000 claims description 6
- 238000004140 cleaning Methods 0.000 claims description 6
- 238000009826 distribution Methods 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 239000006185 dispersion Substances 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 5
- 210000004027 cell Anatomy 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000013178 mathematical model Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000009795 derivation Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06N3/09—Supervised learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a dynamic equipment fault early warning method based on CAE-BiLSTM, which comprises the following steps: acquiring original data running in a normal state of the mobile equipment and preprocessing the original data; intercepting part of the original data as a training set, and processing to obtain a standardized model; constructing a CAE-BiLSTM prediction model, taking original data in a training set as a target value of the prediction model, and taking other original data which are not taken as the training set as input of the prediction model; establishing a security threshold of the RMSE; operating the monitoring data in a state to be monitored, and obtaining a standardized structure after processing; calculating a residual error between the true value and the predicted value by using the RMSE; and judging the magnitude relation between the calculated RMSE of each monitoring parameter and the safety threshold value. The safety threshold value is compared with the root mean square error, so that the risk of false alarm is reduced; by constructing the CAE-BiLSTM training model, the running reliability of the mobile equipment is improved without expert experience.
Description
Technical Field
The invention relates to a fault early-warning method for mobile equipment, in particular to a fault early-warning method for mobile equipment based on CAE-BiLSTM.
Background
With the development of modern industrial Internet technology and production modernization, the traditional field inspection and off-line monitoring method cannot effectively monitor sudden and sporadic faults of equipment. The advantages and disadvantages of the unit state relate to the safe and effective operation of the production platform, the unplanned shutdown of the movable equipment can cause a certain influence on the continuous production of the platform, and the on-line real-time monitoring of the movable equipment can effectively solve the problem. Firstly, the online real-time monitoring technology can convert the traditional periodic maintenance into the optionally maintenance; secondly, the off-line monitoring mode can not effectively record equipment fault data in real time, the on-line real-time monitoring technology can record equipment operation data in real time, the equipment operation state is evaluated accurately by summarizing and analyzing the long-time monitoring data, the equipment management mode is improved, and the existing moving equipment fault early warning method are insufficient:
1. model based on dynamic device operation mechanism: the method is tightly combined with a control theory, a mathematical model is established through an equipment operation mechanism to predict, and the mathematical model is compared with an actual measured value to obtain a residual error; the residual is analyzed to determine if the process is malfunctioning. The method has the defects that most of mechanism models are simplified linear systems, and in the actual industrial process, nonlinear, high-degree-of-freedom and multivariable coupling systems are often adopted, so that the use effect is not ideal.
2. A method based on expertise: the method is based on experience knowledge of people, fault characteristics are deduced, namely after the fault of the movable equipment occurs, an expert discovers a problem by dissecting the equipment, qualitative or quantitative characteristics of the equipment fault are summarized by combining the change condition of historical monitoring parameters before the fault, and equipment fault early warning and monitoring are completed through the characteristics. The method has the defects that the early warning accuracy has strong dependence on the richness of expert knowledge and the level of expert knowledge, and many experiences are difficult to describe in a reasonable formal expression mode.
Disclosure of Invention
In order to solve the defects of the technology, the invention provides a dynamic equipment fault early warning method based on CAE-BiLSTM.
In order to solve the technical problems, the invention adopts the following technical scheme: a dynamic equipment fault early warning method based on CAE-BiLSTM comprises the following steps:
step S1: acquiring original data of the mobile equipment running in a normal state and preprocessing the original data;
step S2: intercepting the original data after the pretreatment of the part in the S1 as a training set, and carrying out parameter processing on the training set to obtain a standardized model;
step S3: constructing a CAE-BiLSTM prediction model, taking original data in a training set as a target value of the prediction model, taking other original data which are not taken as the training set as input of the prediction model, and performing data dimension reduction on the original data in the training set by using a convolution self-encoder CAE to obtain reconstructed data after reconstruction output; modeling the reconstructed data from the forward and reverse directions in a bi-directional short-term neural network BiLSTM, classifying the new data obtained after modeling in a full-connection layer, obtaining a target value of a prediction model, and storing the trained prediction model;
step S4: establishing a safety threshold of Root Mean Square Error (RMSE) by using the original data running in a normal state and applying a CAE-BiLSTM prediction model;
step S5: operating the monitoring data in a state to be monitored, and preprocessing the monitoring data to obtain a standardized structure;
step S6: calculating residual errors between real monitoring data and predicted values in the CAE-BiLSTM prediction model by using the RMSE;
step S7: and judging the magnitude relation between the RMSE of each monitoring parameter calculated in the step S6 and the safety threshold of the RMSE, and further judging whether the mobile equipment fails or not.
Further, the step S1 of preprocessing the original data specifically includes the following steps:
step S11, collecting original data of running of the mobile equipment in a normal state, and obtaining a data set formed by all running monitoring parameters of the mobile equipment;
step S12, analyzing a mechanism of operation in a data set and a data missing condition;
step S13, selecting original data parameters capable of reflecting the performance of equipment as monitoring parameters;
step S14, saving the names of the parameter items which are selected and used as monitoring parameters.
Further, in step S2, the original data serving as the monitoring parameter is intercepted from the original data of the historical operation of the slave device to form a training set. The parameter processing training set comprises parameter screening, data cleaning and parameter standardization, and a standardized model of the training set is stored; in the cleaning process, the training set only keeps the original data operated in the normal period and eliminates the original data operated in the fault period; in the process of parameter standardization of the training set and the method of Z-score is utilized for obtaining the standardized structure in the step S5, so that the variance of the new X data set is 1, the mean value is 0, and the processed original data and the processed monitoring data both accord with standard normal distribution, and the application formula is as follows:
X＝(x-μ)/σ；
wherein μ is a normal distribution of position parameters, X is an axis of symmetry with μ, and σ is the degree of dispersion.
In step S3, the convolution encoder CAE is used to perform data dimension reduction on the original data, which specifically includes the following steps:
step S31, convolution operation: initializing k convolution kernels W in a convolution layer, wherein each convolution kernel is matched with a bias b, generating k feature graphs h after convolution with an input x, and the activation function is sigma, and the formula is:
h k ＝σ(x*w k +b k )；
wherein h is a feature map; w is a convolution kernel; k is the number of convolution kernels; sigma is an activation function; x is input data; b is the current bias;
step S32, pooling operation: carrying out pooling operation on the feature map generated in the step S31 at a pooling layer, and reserving a matrix of the position relationship during pooling;
step S33, reverse pooling operation: performing reverse pooling operation on the feature map generated in the step S32 at a pooling layer, and restoring the pooled original data to the corresponding position of the matrix with the original size by using the matrix with the position relation when pooling is reserved;
step S34, deconvolution operation: each feature map h performs a convolution operation with its corresponding transpose of the convolution kernel and sums the results, then adds the offset c, with the formula:
y＝σ(∑h k *W k +c)；
wherein σ is the activation function; h is a feature map; w is a convolution kernel; k is the number of convolution kernels; c is the current bias;
step S35, updating the weight value: at the convolution layer, updating the weight, firstly determining a cost function costfunction by using a minimum mean square error MSE, wherein the formula is as follows:
wherein n is the number of samples; x is x i Is the target value, y i Is a true value.
In the step S3, modeling is carried out on the original data after CAE processing by utilizing BiLSTM, namely modeling is carried out on the original data after CAE processing by utilizing LSTM from the front direction and the back direction, and then information is spliced together; the information useful for calculation at the subsequent moment is transmitted by forgetting the information in the cell state and memorizing the new information, and useless information is discarded, and the forward propagation formula is as follows:
f t ＝σ(W f ·[h t-1 ，x t ]+b f )；
i t ＝σ(W i ·[h t-1 ，x t ]+b i )；
o t ＝σ(W o [h t-1 ，x t ]+b o )；
h t ＝o t ·Relu(C t )；
wherein f t Is a forgetful door; i.e t Is a memory gate; o (o) t Is an output door; c (C) t Is in a cellular state;is a temporary cellular state; h is a t 、h t-1 The output of the LSTM and the output of the last unit; x is x t Is the current input; sigma and Relu are sigmoid functions and linear rectification activation functions; w (W) f 、b f A forgetting gate weight matrix and a bias roof; w (W) i 、b i Biasing the top for the weight matrix of the input gate; w (W) c 、b c A weight matrix and a bias roof calculated for the state values; w (W) o 、b o A weight matrix and a bias roof for the output gate;
the back propagation algorithm iteratively updates the original data by using a gradient descent method, derives each parameter by using a loss function, and updates the parameter along the direction of the bias.
Further, in step S4, an exponential weighted moving average method is used to establish a security threshold for RMSE.
Further, in step S5, the monitoring data is preprocessed, which specifically includes the following steps:
step S51, obtaining monitoring data of the current dynamic equipment to be evaluated;
step S52, screening parameter items of the monitoring data according to the names of the saved and selected parameter items;
step S53, the parameter values of the screened parameter items are subjected to standardization processing.
Further, in step S7, it is judged that the RMSE of each calculated monitoring parameter does not exceed the safety threshold of RMSE three times, and if the RMSE is normal, the alarm is not given; and if the calculated RMSE comparison of each monitoring parameter exceeds the safety threshold value of the RMSE for three times, alarming.
The invention discloses a dynamic equipment fault early warning method based on CAE-BiLSTM, which utilizes the sequence structure analysis function of BiLSTM and the feature extraction and transformation function of CAE to construct a deep learning model based on a CAE-BiLSTM network, wherein the model can quantify key parameters of equipment and further early warn the equipment by observing the change trend of residual errors, and has the following beneficial effects:
1. by comparing the safety threshold with the root mean square error, the residual error of the target parameter is statistically analyzed, the running condition of the equipment can be known by observing the variation trend of the residual error, the risk of false alarm is reduced, the residual error is predicted by replacing the mathematical model of most mechanism resume of the equipment, and the fault early warning of the mobile equipment based on CAE-BiLSTM is more suitable for a nonlinear and multivariable coupling system of the mobile equipment in the industrial process.
2. Constructing a CAE-BiLSTM training model, performing data dimension reduction and reconstruction on data in a training set by using a convolution self-encoder CAE, thereby being beneficial to improving the performance of the running equipment and reducing the consumption of a memory; modeling reconstructed data from the front direction and the back direction by utilizing a bi-directional short-term neural network BiLSTM, and learning long-term dependence in the data through a memory unit and a memory gate of the bi-directional short-term neural network BiLSTM, and combining with other deep learning architectures, so that the running performance and reliability of the mobile equipment are improved, the cost can be reduced, and the safe running of the mobile equipment is ensured; the new data after modeling enter the full-connection layer for classification, and due to the structural characteristics of the full-connection layer, when the complexity of the model is increased and the number of full-connection layers is deepened, the learning capacity of the model can be improved, and the learning capacity of the training model is utilized, so that the learning capacity of any expert knowledge and experience is not needed to be used as fault prediction reasoning, and the early warning accuracy is improved.
Drawings
Fig. 1 is a logic diagram of the technical scheme of the present invention.
FIG. 2 is a schematic diagram of the CAE-BiLSTM model structure of the present invention.
FIG. 3 is a schematic diagram of the BiLSTM model structure of the present invention.
Detailed Description
The invention will be described in further detail with reference to the drawings and the detailed description.
Step S1: acquiring original data of the mobile equipment running in a normal state and preprocessing the original data;
step S2: intercepting part of the preprocessed original data in the S1 as a training set, and carrying out parameter processing on the training set to obtain a standardized model;
step S3: constructing a CAE-BiLSTM prediction model, taking original data in a training set as a target value of the prediction model, taking other original data which are not taken as the training set as input of the prediction model, and performing data dimension reduction on the original data in the training set by using a convolution self-encoder CAE to obtain reconstructed data after reconstruction output; modeling the reconstructed data from the forward and reverse directions in a bi-directional short-term neural network BiLSTM, classifying the new data obtained after modeling in a full-connection layer, obtaining a target value of a prediction model, and storing the trained prediction model;
step S4: establishing a safety threshold of Root Mean Square Error (RMSE) by operating original data in a normal state and applying a CAE-BiLSTM prediction model;
step S5: operating the monitoring data in a state to be monitored, and preprocessing the monitoring data to obtain a standardized structure;
step S6: calculating residual errors between real monitoring data and predicted values in the CAE-BiLSTM prediction model by using the RMSE;
step S7: and judging the magnitude relation between the RMSE of each monitoring parameter calculated in the step S6 and the safety threshold of the RMSE, and further judging whether the mobile equipment fails or not.
Specifically, as shown in a logic diagram of a technical scheme of a CAE-BiLSTM-based dynamic equipment fault early warning method shown in fig. 1, in a training process, after original data are operated in a normal state, the original data are preprocessed, after correlation analysis is carried out, input parameters are selected, a prediction model is built, and a threshold value of RMSE (root mean square error) is built through the original data in the normal operation state; in the monitoring process, running data in a state to be monitored, preprocessing the monitored data, selecting input and output parameters determined during training, putting the data to be monitored into a prediction model established during training to obtain a prediction sequence, judging whether the calculated RMSE exceeds a threshold established during training for 3 times, and if the RMSE does not exceed the threshold for 3 times, ensuring that the state is normal; if the RMSE is 3 times, the threshold value is exceeded, then an alarm is given.
The embodiment of the invention provides a mobile equipment fault early warning method based on CAE-BiLSTM, which comprises the following specific steps:
1: raw data of running in a normal state of the mobile equipment are obtained, a data set formed by all running parameters of the mobile equipment is obtained, the running mechanism in the data set is analyzed, the raw data parameters capable of reflecting the performance of the mobile equipment are selected as monitoring parameters, and parameter item names of the monitoring parameters are saved for a training set.
2: performing operations such as parameter screening, data cleaning, parameter standardization and the like on the training set, and storing a standardized model of the training set; in the parameter screening process, the selection of the monitoring parameters is obtained from the data items stored in the step 1; in the data cleaning process, only the original data operated in the normal operation period is reserved, and the original data operated in the fault period is required to be removed; the normalization process adopts a Z-score method, namely, a formula of X= (X-mu)/sigma is applied, wherein mu is a position parameter of normal distribution, X is a symmetry axis with mu, sigma is a degree of dispersion, so that a new X data set variance is 1, and a mean value is 0, and the processed data accords with the normal distribution.
3: the CAE-BiLSTM model shown in figure 2 is constructed, the original data of the dynamic equipment in the training set are respectively used as the targets of the model, other original data which are not used as the training set are used as the input of the model, the supervised training is carried out, and the training results are saved.
31: CAE automatic encoder is a typical unsupervised godA network model, namely a self-encoder; based on a back propagation algorithm and an optimization method, the self-encoder guides the neural network to try to learn a mapping relation by taking input data X as supervision at an input layer to obtain a reconstruction output X R Typically, an algorithm model consists of two parts: an encoder and a decoder, the CAE network structure including an encoding process and a decoding process; in the encoding process, original data input in an input layer are encoded into a pooling layer through a CAE layer, and high-dimensional input X is encoded into a low-dimensional hidden variable h through an encoder, so that the neural network is forced to learn the characteristics of the most information quantity; in the decoding process, the coded variable in the pooling layer of the coding process is decoded to the up-sampling layer through the CAE layer, the decoder restores the hidden variable h of the hidden layer to the initial dimension, and the best state is that the output of the decoder can perfectly or approximately restore the original input, namely X R The pressure is approximately equal to X; the convolution self-encoder replaces a simple self-encoding full-connection layer by using the CAE layer, is matched with a two-dimensional topological structure of the image, avoids the loss of reconstruction errors, changes the encoding process into a convolution process, and changes the decoding process into a deconvolution process; the decoded data enter a BiLSTM layer to model the original data after CAE processing from the positive direction and the negative direction, then the data enter a full-connection layer to be classified and integrated, the full-connection layer is replaced by the CAE layer in the convolution self-encoder, and finally the original data after CAE processing is output to an output layer.
311: convolution layer-convolution operation: initializing k convolution kernels (W), wherein each convolution kernel is matched with one offset b, and generating k feature graphs h after convolution with an input x, and the activation function is sigma, and the formula is:
h k ＝σ(x*w k +b k )；
wherein h is a feature map; w is a convolution kernel; k is the number of convolution kernels; sigma is an activation function; x is input data; b is the current bias;
312: pooling layer-pooling operation (MaxPooling): pooling operation is carried out on the feature map generated in 311, and a matrix of the position relation during pooling is reserved, so that the subsequent anti-pooling operation is facilitated;
313: pooling layer-self-coding (anti-pooling operation): performing inverse pooling operation on the feature map generated in 311, and restoring the pooled original data to the corresponding position of the original size matrix by using the matrix reserved to the position relation during pooling;
314: convolutional layer-self-encoding (deconvolution operation): performing convolution operation on each feature diagram h and the corresponding transpose of the convolution kernel, summing the results, and adding the offset c, wherein the formula is as follows:
y＝σ(∑h k *W k +c)；
wherein σ is the activation function; h is a feature map; w is a convolution kernel; k is the number of convolution kernels; c is the current bias;
315: convolution layer-update weights: to update the weights, firstly, a costfunction cost function is determined, and the MSE (minimum mean square error) function adopted here, that is, the square sum of the target value minus the actual value is averaged, where the formula is:
wherein 2n is the simplification of derivation, x i Is the target value, y i Is an actual value.
32: the LSTM is a special recurrent neural network, the BiLSTM does not change the internal structure of the LSTM, but models the CAE processed original data by using the LSTM from the front direction and the back direction, and then the information is spliced together, but the problem that the importance degree of the information of the original data is changed before and after the LSTM is processed due to the model structure, so that the prediction accuracy is reduced is solved, and meanwhile, gradient disappearance and explosion are avoided. The calculation process of LSTM is summarized as: by forgetting information in the cell state and memorizing new information, information useful for calculation at a later time is transferred, and useless information is discarded, and forward propagation thereof can be expressed as:
f t ＝σ(W f ·[h t-1 ，x t ]+b f )；
i t ＝σ(W i ·[h t-1 ，x t ]+b i )；
o t ＝σ(W o [h t-1 ，x t ]+b o )；
h t ＝o t ·Relu(C t )；
wherein f t Is a forgetful door; i.e t Is a memory gate; o (o) t Is an output door; c (C) t Is in a cellular state;is a temporary cellular state; h is a t 、h t-1 The output of the LSTM and the output of the last unit; x is x t Is the current input; sigma and Relu are sigmoid functions and linear rectification activation functions; w (W) f 、b f A forgetting gate weight matrix and a bias roof; w (W) i 、b i Biasing the top for the weight matrix of the input gate; w (W) c 、b c A weight matrix and a bias roof calculated for the state values; w (W) o 、b o A weight matrix and a bias roof for the output gate;
the back propagation algorithm is to iteratively update the CAE-processed original data by using a gradient descent method, derive each parameter by using a loss function, and update the parameters along the direction of the derivation.
FIG. 3 shows a BiLSTM model structure, wherein W f As a forward weight matrix, W b For the inverse weight matrix, o is the temporary cell state and the cell state, x is the current input, x= [ x ] 1 ，x 2 ，x 3 …x n ]；In order to output the data it is possible,α f for the output of the last neuron when the original data after CAE processing is forward propagated, alpha b And outputting the last neuron when the CAE processed data are back-propagated, and finally combining the result with the result output by the LSTM to obtain an output result. Constructing a CAE-BiLSTM training model, performing original data dimension reduction and reconstruction on data in a training set by using a convolution self-encoder CAE, thereby being beneficial to improving the performance of the running equipment and reducing the consumption of a memory; modeling reconstructed data from the front direction and the back direction by utilizing a bi-directional short-term neural network BiLSTM, and learning long-term dependence in the data through a memory unit and a memory gate of the bi-directional short-term neural network BiLSTM, and combining with other deep learning architectures, so that the running performance and reliability of the mobile equipment are improved, the cost can be reduced, and the safe running of the mobile equipment is ensured; the new data after modeling enter the full-connection layer for classification, and due to the structural characteristics of the full-connection layer, when the complexity of the model is increased, the number of full-connection layers is deepened, the learning capacity of the model can be improved, no expert experience is needed, no manual label establishment is needed, and a large amount of manpower is saved.
4: after the CAE-BiLSTM model is built, before the CAE-BiLSTM model is applied, running monitoring data in a state to be monitored, obtaining the monitoring data of the current dynamic equipment to be evaluated, screening the monitoring parameter items according to the names of the selected parameter items stored in 1, and standardizing a screening result set of the monitoring parameter items.
5: and (3) applying a CAE-BiLSTM training model, putting the data to be monitored into the CAE-BiLSTM model to obtain a predicted sequence, and establishing a Root Mean Square Error (RMSE) safety threshold by using an exponential weighted moving average method.
6: the residual error between the predicted value and the true value is calculated by the Root Mean Square Error (RMSE), and the smaller the root mean square error is, the higher the measurement accuracy is.
7: judging whether the RMSE exceeds a threshold value for three times according to the RMSE of each parameter calculated in the step 6; if the RMSE does not exceed the threshold value for three times, the RMSE is normal; if the RMSE is compared with the threshold value for three times, alarming; the monitoring accuracy can be improved, and the number of false alarms can be reduced.
The above embodiments are not intended to limit the present invention, and the present invention is not limited to the above examples, but is also intended to be limited to the following claims.
Claims (9)
1. The utility model provides a mobile equipment fault early warning method based on CAE-BiLSTM, which is characterized by comprising the following steps:
step S1: acquiring original data of the mobile equipment running in a normal state and preprocessing the original data;
step S2: intercepting the original data after the pretreatment of the part in the S1 as a training set, and carrying out parameter processing on the training set to obtain a standardized model;
step S3: constructing a CAE-BiLSTM prediction model, taking original data in a training set as a target value of the prediction model, taking other original data which are not taken as the training set as input of the prediction model, and performing data dimension reduction on the original data in the training set by using a convolution self-encoder CAE to obtain reconstructed data after reconstruction output; modeling the reconstructed data from the forward and reverse directions in a bi-directional short-term neural network BiLSTM, classifying the new data obtained after modeling in a full-connection layer, obtaining a target value of a prediction model, and storing the trained prediction model;
step S4: establishing a safety threshold of Root Mean Square Error (RMSE) by using the original data running in a normal state and applying a CAE-BiLSTM prediction model;
step S5: operating the monitoring data in a state to be monitored, and preprocessing the monitoring data to obtain a standardized structure;
step S6: calculating residual errors between real monitoring data and predicted values in the CAE-BiLSTM prediction model by using the RMSE;
step S7: and judging the magnitude relation between the RMSE of each monitoring parameter calculated in the step S6 and the safety threshold of the RMSE, and further judging whether the mobile equipment fails or not.
2. The CAE-BiLSTM based dynamic device fault alerting method of claim 1, wherein: the step S1 of preprocessing the original data specifically comprises the following steps:
step S11, collecting original data of running of the mobile equipment in a normal state, and obtaining a data set formed by all running monitoring parameters of the mobile equipment;
step S12, analyzing a mechanism of operation in a data set and a data missing condition;
step S13, selecting original data parameters capable of reflecting the performance of equipment as monitoring parameters;
step S14, saving the names of the parameter items which are selected and used as monitoring parameters.
3. The CAE-BiLSTM based dynamic device fault alerting method of claim 2, wherein: in step S2, the original data serving as the monitoring parameter is intercepted from the original data of the historical operation of the slave device to form a training set.
4. The CAE-BiLSTM based dynamic device fault alerting method of claim 3, wherein: the step S2 of processing the parameters in the training set comprises parameter screening, data cleaning and parameter standardization, and a standardized model of the training set is stored; in the cleaning process, the training set only keeps the original data operated in the normal period and eliminates the original data operated in the fault period; in the process of parameter standardization of the training set and the method of Z-score is utilized for obtaining the standardized structure in the step S5, so that the variance of the new X data set is 1, the mean value is 0, and the processed original data and the processed monitoring data both accord with standard normal distribution, and the application formula is as follows:
X＝(x-μ)/σ；
wherein μ is a normal distribution of position parameters, X is an axis of symmetry with μ, and σ is the degree of dispersion.
5. The CAE-BiLSTM based dynamic device fault alerting method of claim 1, wherein: in the step S3, the convolution encoder CAE is used to perform data dimension reduction on the original data, and specifically includes the following steps:
step S31, convolution operation: initializing k convolution kernels W in a convolution layer, wherein each convolution kernel is matched with a bias b, generating k feature graphs h after convolution with an input x, and the activation function is sigma, and the formula is:
h k =σ(x*W k +b k )；
wherein h is a feature map; w is a convolution kernel; k is the number of convolution kernels; sigma is an activation function; x is input data; b is the current bias;
step S32, pooling operation: carrying out pooling operation on the feature map generated in the step S31 at a pooling layer, and reserving a matrix of the position relationship during pooling;
step S33, reverse pooling operation: performing reverse pooling operation on the feature map generated in the step S32 at a pooling layer, and restoring the pooled original data to the corresponding position of the matrix with the original size by using the matrix with the position relation when pooling is reserved;
step S34, deconvolution operation: each feature map h performs a convolution operation with its corresponding transpose of the convolution kernel and sums the results, then adds the offset c, with the formula:
y=σ(Σh k *W k +c)；
wherein σ is the activation function; h is a feature map; w is a convolution kernel; k is the number of convolution kernels; c is the current bias;
step S35, updating the weight value: at the convolution layer, updating the weight, firstly determining a cost function costfunction by using a minimum mean square error MSE, wherein the formula is as follows:
wherein n is the number of samples; x is x i Is the target value, y i Is a true value.
6. The CAE-BiLSTM based dynamic device fault alerting method of claim 1, wherein: in the step S3, modeling is carried out on the original data after CAE processing by utilizing BiLSTM, namely modeling is carried out on the original data after CAE processing by utilizing LSTM from the front direction and the back direction, and then information is spliced together; the information useful for calculation at the subsequent moment is transmitted by forgetting the information in the cell state and memorizing the new information, and useless information is discarded, and the forward propagation formula is as follows:
f t =σ(W f ·[h t-1 ，x t ]+b f ；
i t =σ(W i ·[h t-1 ，x t ]+b i )；
o t ＝σ(w o [h t-1 ，x t ]+b o )；
h t ＝o t ·Relu(C t )；
wherein f t Is a forgetful door; i.e t Is a memory gate; o (o) t Is an output door; c (C) t Is in a cellular state;is a temporary cellular state; h is a t 、h t-1 The output of the LSTM and the output of the last unit; x is x t Is the current input; sigma and Relu are sigmoid functions and linear rectification activation functions; w (W) f 、b f A forgetting gate weight matrix and a bias roof; w (W) i 、b i Biasing the top for the weight matrix of the input gate; w (W) c 、b c A weight matrix and a bias roof calculated for the state values; w (W) o 、b o A weight matrix and a bias roof for the output gate;
the back propagation algorithm iteratively updates the original data by using a gradient descent method, derives each parameter by using a loss function, and updates the parameter along the direction of the bias.
7. The CAE-BiLSTM based dynamic device fault alerting method of claim 1, wherein: in the step S4, an exponential weighted moving average method is used to establish a security threshold of RMSE.
8. The CAE-BiLSTM based dynamic device fault alerting method of claim 2, wherein: the step S5 of preprocessing the monitoring data specifically includes the following steps:
step S51, obtaining monitoring data of the current dynamic equipment to be evaluated;
step S52, screening parameter items of the monitoring data according to the names of the saved and selected parameter items;
step S53, the parameter values of the screened parameter items are subjected to standardization processing.
9. The CAE-BiLSTM based dynamic device fault alerting method of claim 1, wherein: in the step S7, if the calculated RMSE of each monitoring parameter is judged to be beyond the safety threshold value of the RMSE for three times, the RMSE is normal and does not alarm; and if the calculated RMSE comparison of each monitoring parameter exceeds the safety threshold value of the RMSE for three times, alarming.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311237531.8A CN117290726A (en) | 2023-09-25 | 2023-09-25 | CAE-BiLSTM-based fault early warning method for mobile equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311237531.8A CN117290726A (en) | 2023-09-25 | 2023-09-25 | CAE-BiLSTM-based fault early warning method for mobile equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117290726A true CN117290726A (en) | 2023-12-26 |
Family
ID=89247437
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311237531.8A Pending CN117290726A (en) | 2023-09-25 | 2023-09-25 | CAE-BiLSTM-based fault early warning method for mobile equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117290726A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117911009A (en) * | 2024-03-19 | 2024-04-19 | 江苏金恒信息科技股份有限公司 | XGBoost algorithm-based equipment fault early warning method and system |
| CN118035929A (en) * | 2024-04-12 | 2024-05-14 | 江西江投能源技术研究有限公司 | Power plant safety monitoring and early warning method and system based on thermodynamics and data mining |
-
2023
- 2023-09-25 CN CN202311237531.8A patent/CN117290726A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117911009A (en) * | 2024-03-19 | 2024-04-19 | 江苏金恒信息科技股份有限公司 | XGBoost algorithm-based equipment fault early warning method and system |
| CN117911009B (en) * | 2024-03-19 | 2024-06-11 | 江苏金恒信息科技股份有限公司 | XGBoost algorithm-based equipment fault early warning method and system |
| CN118035929A (en) * | 2024-04-12 | 2024-05-14 | 江西江投能源技术研究有限公司 | Power plant safety monitoring and early warning method and system based on thermodynamics and data mining |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111666982B (en) | Electromechanical equipment fault diagnosis method based on deep neural network | |
| CN117290726A (en) | CAE-BiLSTM-based fault early warning method for mobile equipment | |
| Caceres et al. | A probabilistic Bayesian recurrent neural network for remaining useful life prognostics considering epistemic and aleatory uncertainties | |
| CN111813084B (en) | Mechanical equipment fault diagnosis method based on deep learning | |
| CN113642754B (en) | Complex industrial process fault prediction method based on RF noise reduction self-coding information reconstruction and time convolution network | |
| CN112131212A (en) | Hybrid cloud scene-oriented time sequence data anomaly prediction method based on ensemble learning technology | |
| Chen et al. | Time series data for equipment reliability analysis with deep learning | |
| CN113762329A (en) | Method and system for constructing state prediction model of large rolling mill | |
| CN112785091A (en) | Method for performing fault prediction and health management on oil field electric submersible pump | |
| CN110309537B (en) | Intelligent health prediction method and system for aircraft | |
| Chang et al. | Remaining useful life prediction for rolling bearings using multi-layer grid search and LSTM | |
| CN110736968A (en) | Radar abnormal state diagnosis method based on deep learning | |
| CN114266301A (en) | Intelligent power equipment fault prediction method based on graph convolution neural network | |
| Yong et al. | Bayesian autoencoders for drift detection in industrial environments | |
| CN115526241A (en) | Real-time abnormity detection method, device, equipment and medium for aviation hydraulic pump station | |
| CN114679310A (en) | Network information security detection method | |
| CN115510748A (en) | Landslide displacement prediction method based on variational modal decomposition and CNN-GRU | |
| CN116401516A (en) | Power load abnormal data detection and correction method based on deep learning | |
| KR102419782B1 (en) | Industrial facility failure prediction modeling technique and alarm integrated system and method based on artificial intelligence | |
| CN113779879A (en) | Medium-and-long-term electricity utilization abnormity detection method based on LSTM-seq2seq-attention model | |
| Qin et al. | Remaining useful life prediction using temporal deep degradation network for complex machinery with attention-based feature extraction | |
| CN113762967A (en) | Risk information determination method, model training method, device, and program product | |
| JP2024050510A (en) | Method and system for performing root cause analysis using a trained machine learning model | |
| Zhao et al. | Software reliability growth model based on fuzzy wavelet neural network | |
| Chen et al. | Reliability analysis using deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |