CN114841076A - Power battery production process fluctuation abnormity detection method based on space-time diagram model - Google Patents
Power battery production process fluctuation abnormity detection method based on space-time diagram model Download PDFInfo
- Publication number
- CN114841076A CN114841076A CN202210547979.9A CN202210547979A CN114841076A CN 114841076 A CN114841076 A CN 114841076A CN 202210547979 A CN202210547979 A CN 202210547979A CN 114841076 A CN114841076 A CN 114841076A
- Authority
- CN
- China
- Prior art keywords
- model
- convolution
- time
- production process
- battery production
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06N3/045—Combinations of networks
-
- 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
- G06N3/047—Probabilistic or stochastic networks
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
- H01M10/00—Secondary cells; Manufacture thereof
- H01M10/04—Construction or manufacture in general
-
- H—ELECTRICITY
- H01—ELECTRIC ELEMENTS
- H01M—PROCESSES OR MEANS, e.g. BATTERIES, FOR THE DIRECT CONVERSION OF CHEMICAL ENERGY INTO ELECTRICAL ENERGY
- H01M2220/00—Batteries for particular applications
- H01M2220/20—Batteries in motive systems, e.g. vehicle, ship, plane
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Probability & Statistics with Applications (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Manufacturing & Machinery (AREA)
- Chemical & Material Sciences (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Electrochemistry (AREA)
- General Chemical & Material Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method for detecting fluctuation abnormity of a power battery production process based on a space-time diagram model. The method comprises the following steps: using key parameters of a battery production process for training a model, and obtaining a time model of a sequence by using time convolution networks with different convolution kernel sizes; constructing a tie matrix, and acquiring spatial information of a time sequence by using a graph convolution neural network; constructing a spatio-temporal map rolling block with a gating mechanism to filter the obtained information to obtain effective time and space dependence; aggregating output information of all gated space-time graph convolution networks to perform single-step prediction on an input sliding time window; calculating a prediction error by using the observed value and the predicted value of the data, and calculating a threshold value of the abnormal score by using the prediction error; if the prediction error in the test data is larger than the threshold value, the test data is judged to be abnormal, and an alarm is given out.
Description
Technical Field
The invention relates to the field of battery production process fluctuation analysis, in particular to a battery production process abnormal fluctuation detection method based on a space-time diagram model.
Background
The environmental protection and energy saving characteristics of the new energy automobile promote the rapid popularization and development of the new energy automobile, and the future prospect shows a positive situation. The power battery is used as a core component of the new energy locomotive, and the automatic monitoring of the core process fluctuation in the production process of the battery is beneficial to reducing the cost and improving the efficiency. Throughout the world, currently, a design of a factory battery management system and a design of an intelligent manufacturing plan management module are in a simulation or closed test stage, many core technologies need to be broken through urgently, wherein the core technologies comprise battery production equipment diagnosis and battery production process fluctuation analysis battery production process capability analysis, and the invention focuses on battery production process fluctuation analysis anomaly detection. The battery production process comprises a plurality of core processes, such as mixing, coating, rolling, slitting, laminating, liquid injection, formation and the like, and needs to be monitored timely and accurately. Meanwhile, each process can be timely and effectively recorded by a sensor to obtain a multi-dimensional time sequence. The abnormal model of the battery production process can be hidden in the parameters (current, voltage, temperature and the like), and the abnormal detection in the production process becomes a very important problem. However, the abnormal mode forms are various, the acquisition of all kinds of abnormal labels is very difficult, huge cost is needed, meanwhile, in the running process of battery production equipment, most of time is in a normal state, the abnormal and normal categories are extremely unbalanced, in the deep learning field, many popular classification algorithms for supervising learning cannot work effectively, and the difficulty in detecting the fluctuation abnormality of the production process of the power battery is improved.
The accuracy, robustness and efficiency of abnormality detection of fluctuation analysis of the battery production process are comprehensively considered, and an unsupervised learning method is the most appropriate method. And predicting the key parameters through the potential space-time relationship in the deep learning model learning data, and detecting abnormal values according to the difference between the observed values and the predicted values. However, the stability of the parameters and the linear and nonlinear relationships between the parameters during the cell production process are quite complex and dynamically changing over time. The existing unsupervised anomaly detection algorithm cannot process the time and space dependence of the power battery production process, and a great space is improved.
In summary, in the problem of fluctuation analysis of power battery production process, a new multi-dimensional time series abnormality detection method is found, which can model time information and a spatial mode of battery core production process parameters, and becomes a problem to be solved at present.
Disclosure of Invention
The invention provides an Anomaly Detection algorithm for battery production process fluctuation, which is called a gated space-time map Model (GSTAD). The algorithm considers the time dependence and the spatial mode of the core parameters in the battery production process, and can realize efficient detection of the abnormal process fluctuation in the battery production process.
The technical scheme of the invention comprises the following steps:
1) the key parameters of the battery production process are preprocessed, so that parameters with different attributes and different ranges have the same measurement.
2) Dividing the preprocessed data into three subsets, where N is 1 The subset is used for training the model, and only a normal data set is contained; n is a radical of 2 The subset is used for adjusting the hyper-parameters of the model and selecting the threshold value; n is a radical of 3 Performance for test model, N 2 And N 3 Both contain normal and abnormal data, the input of the model is a sliding time window X W ∈R k×L Where k represents the characteristic dimension of the multidimensional time series and L represents the size of the sliding time window;
3) constructing a gated space-time map model;
firstly, constructing time convolution networks of convolution kernels with different sizes and convolution with convolution kernels of 1 multiplied by 1 to capture the time dependence of key parameters of the power battery production process;
the key parameters of the production process of different power batteries are expressed by the form of a graph,the graph is expressed as G ═ (V, E), where V is the set of nodes and E is the set of edges; the number of nodes in the graph is denoted by k. Let V E V denote a node, E ∈ (V, u) ∈ E denote an edge pointing from u to V; and (3) an adjacency matrix of a directed graph of the constructed battery production process parameter system is expressed as A e R k×k If (v) i ，v j ) E is E, then A ij If 1, if
Then A is ij 0; in the production process of the battery, complex relations exist among different parameters, such as rolling front stretching tension and rolling back stretching tension, splitting speed and working power of a splitting machine, stirring speed and stirring temperature in a homogenate process and the like, the linear and nonlinear relations of different production processes are captured through the structure of an adaptive graph, an adjacent matrix is specifically learned by two methods, and firstly, the adjacent matrix is constructed according to the correlation of the process parameters in the production process of the battery
Second, the adaptive adjacency matrix A is learned by the model through gradient descent apt By randomly initializing the embedded representation of the start and target nodes as a learnable parameter E 1 ,E 2 ∈R k×m Where m represents the embedded dimension of each node, the adaptive connection matrix of the target is represented as equation 3. The linear and non-linear relationship of different production processes is captured by the structure of the adaptive graph, the information of relevant process parameters is aggregated by using the graph convolution neural network, the graph convolution neural network is expressed as formula 4,
wherein A is apt Adaptive representation of initializationAnd (3) a adjacency matrix, wherein a ReLU activation function is used for removing some weak connections and increasing the sparsity of the adjacency matrix, and Softmax is an activation function. During the training of the model, the optimal connection matrix that facilitates prediction is learned. X represents the input of the graph convolution neural network, W i1 And W i2 A parameter matrix representing a graph convolution neural network model.
4) Filtering the obtained spatiotemporal information using a gate structure;
in the gated space-time diagram model, an output gate g shown in formula 5 is adopted i The output of each gate structure contains valid information of the time series. In formula 6, the output results of each gate structure are added, and the spatio-temporal information of the multi-dimensional time sequence is obtained through two fully-connected layers to predict the observed value of the next time step;
g i ＝h i1 (Θ 1 *X+a)⊙h i2 (Θ 2 *X+b) (5)
wherein h is i1 Sum hi2 represents two operations of the series time convolution and graph convolution neural network, Θ 1 ，Θ 2 Respectively, wherein the parameter indicates multiplication of corresponding elements of the matrix, a, b, c, d are offset entries, W 1 And W 2 Tanh and ReLU are activation functions for the parameter matrix of the fully-connected layer.
5) Training the model by using normal data, using an automatic threshold strategy, adjusting the hyper-parameters of the model by using a verification set, and selecting the optimal threshold. And finally, testing the performance of the model intuitively, obtaining a larger prediction error when the test data set contains the abnormality, and judging the observed value of which the prediction error is larger than the threshold value as the abnormality.
Preferably, the making of parameters with different attributes and different ranges have the same metric specifically:
the data for each sensor is normalized using the following formula,
wherein x ∈ R N And (2) a measurement parameter, min (x), max (x) respectively represent the minimum value and the maximum value of x, x' represents the value after x is normalized, and the value of eps is set to be 1e-8, so that the condition of dividing 0 is avoided.
Preferably, the key parameters of the power battery production process comprise rolling speed, slitting length and injection temperature.
Preferably, the sliding time window size is set to 42.
Preferably, the expansion coefficient r is set to 2 or 3.
Preferably, the time convolution network adopts causal convolution and dilation convolution, as shown in formula 2; zero padding is added in the causal convolution, so that the output at the current moment is only related to the historical information and the input at the current moment, and the causality and the autoregressive characteristic of the sequence are ensured; the expansion convolution increases the size of the convolution receptive field, captures time patterns of time sequences under different receptive fields, and expands the size of the receptive field along with the depth index of the model, so that the model obtains more global information. Meanwhile, 1 × 1 convolution is added, so that the linear relation learning capability of the model is improved;
where f represents the convolution kernel, K represents the length of the convolution kernel, r represents the expansion coefficient, x' represents the preprocessed data, and t represents time.
The scheme provided by the invention is as follows: the battery production process fluctuation abnormity detection method based on the space-time diagram convolutional neural network can successfully model a core parameter system of battery production process fluctuation, improves accuracy of online abnormity detection, ensures stability and high efficiency of a battery production line, and reduces production cost and staff operation burden.
Drawings
FIG. 1 is a flow chart of the operation of the present invention;
FIG. 2 is a diagram of a dilated causal convolution;
FIG. 3 is a block diagram of a temporal convolution network;
FIG. 4 is a block diagram of a graph convolution neural network;
FIG. 5 is a block diagram of a gated space-time graph convolutional neural network;
Detailed Description
In order to make the objects, solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the present invention. The anomaly detection method for the fluctuation analysis of the battery production process, disclosed by the invention, comprises the following steps of data preprocessing, model training, threshold selection and online anomaly detection, and is specifically divided into the following steps:
step 1: the data of important parameters of the battery production process, such as rolling speed, slitting length, injection temperature and the like, are collected at equal time intervals, and have different properties and different ranges. The data for each sensor is normalized using the following formula,
wherein x represents a measurement parameter, min (x), max (x) represent the minimum value and the maximum value of x respectively, x' represents the value after x is normalized, and the value of eps is set to be 1e-8, so that the situation of dividing 0 is avoided.
Step 2: divide data into three subsets, where N 1 The subset is used for training the model, and only a normal data set is contained; n is a radical of 2 The subset is used for adjusting the hyper-parameters of the model and selecting the threshold value; n is a radical of 3 For testing the performance of a model whose input is a sliding time window X W ∈R k×L Where k represents the characteristic dimension of the multi-dimensional time series, L represents the size of the sliding time window, the sliding time window size is set to 42; the output of the model is the predicted value of all the characteristics at the next moment;
firstly, constructing a time convolution network with convolution kernel sizes of 1 × 3,1 × 5,1 × 7 and 1 × 9 and a convolution with convolution kernel sizes of 1 × 1 to capture the time dependence of key parameters of the power battery production process; causal convolution and dilation convolution are used in the time convolution network, as shown in equation 2. As shown in FIG. 2, the causal convolution adds zero padding to correlate the output at the current time only with its historical information and the input at the current time, ensuring the causal and autoregressive characteristics of the sequence. The expansion convolution can increase the size of the convolution receptive field, wherein the expansion coefficient r is set to be 2 or 3, time patterns of the time sequence under different receptive fields are captured, and the size of the receptive field is expanded along with the depth index of the model, so that the model obtains more global information. Weight normalization, ReLU activation function and Dropout regularization are also added to the time convolution network, and finally residual connection is added, so that the model can avoid gradient disappearance in a very deep network. As shown in fig. 3, a time convolution block is constructed by using time convolution networks with different convolution kernel sizes and 1 × 1 convolution, and the purpose of adding 1 × 1 convolution is to improve the capability of a model to learn linear relationship;
where f denotes a convolution kernel, K denotes the length of the convolution kernel, and r denotes the expansion coefficient of the expansion convolution.
The key parameters of the production process of different power batteries are expressed by the form of a graph, and the graph is expressed as G ═ V, E, wherein V is a set of nodes and E is a set of edges. The number of nodes in the graph is denoted by k. Let V E V denote a node, and E (V, u) E denote an edge pointing from u to V. And constructing a adjacency matrix of a directed graph of the battery production process parameter system, wherein the adjacency matrix is expressed as A epsilon R k×k If (v) i ，v j ) E is E, then A ij If s, if
Then A is ij 0. In the battery production process, different parameters existThe complex relations such as the rolling front stretching tension and the rolling back stretching tension, the slitting speed and the working power of a slitting machine, the stirring speed and the stirring temperature in the pulp homogenizing process and the like are closely related, the linear and nonlinear relations of different production processes are captured through the structure of the self-adaptive graph, and the prediction of the value of each key parameter at the next moment is of great significance. Two methods are used to learn the adjacency matrix, firstly, the adjacency matrix is constructed by adopting expert knowledge and through the correlation of the process parameters in the battery production process
Second, the adaptive adjacency matrix A is learned by the model through gradient descent apt By randomly initializing the embedded representation of the start and target nodes as a learnable parameter E 1 ,E 2 ∈R k×m Where m represents the embedded dimension of each node, the adaptive connection matrix of the target is represented as equation 3. As shown in fig. 4, each node can aggregate the information of the related process parameters by using the convolution neural network, update its vector expression, the convolution neural network is expressed as formula 4, a space network block is formed by adopting 2 layers of convolution neural networks, learn the space dependence of the multidimensional time sequence,
wherein A is apt And (3) representing an initialized adaptive adjacency matrix, removing some weak connections by using a ReLU activation function, and increasing sparsity of the adjacency matrix, wherein Softmax is an activation function. During the training of the model, the optimal connection matrix that facilitates prediction is learned. X represents the input of the graph convolution neural network, W i1 And W i2 A parameter matrix representing a graph convolution neural network model.
And 4, step 4: as shown in fig. 5, a gated space-time graph convolutional neural network is built using the temporal convolution block in step 3 and the spatial convolution block in step 4. Gating mechanisms are widely used to handle sequential tasks, such as long and short term time networks and gated loop units. In GSTAD, the resulting spatio-temporal information is filtered using a gate structure, and a simple output gate of the gate structure shown in equation 5 is used to select information that is output favorably for prediction. The output of each gate structure contains valid information of the time series. In formula 6, the output results of each gate structure are added, and the observation value prediction of the next time step is performed by obtaining the spatio-temporal information of the multi-dimensional time series through the full connection layer.
g i ＝h i1 (Θ 1 *X+a)⊙h i2 (Θ 2 *X+b) (5)
Wherein h is i1 Sum hi2 represents two operations of the series time convolution and graph convolution neural network, Θ 1 ，Θ 2 Respectively, a parameter therein, a, b, c, d are offset terms, W 1 And W 2 Tanh and ReLU are activation functions for the parameter matrix of the fully-connected layer.
And 5: training the model by using normal data, taking mean square error loss as a loss function, taking Adam as an optimizer, and updating parameters of the model by using a gradient back propagation algorithm. The trained model can make a good prediction result for the normal state, but the prediction result error for the abnormal data is large, and the prediction error is taken as the abnormal score, as shown in formula 7. Using a non-parametric, dynamic and unsupervised approach to residual estimation, threshold selection can be made without any assumptions, identifying extreme values. The method is suitable for data streams with different attributes and different ranges, and solves the problems of diversity, non-stationarity and noise by automatically setting a threshold value. And adjusting the hyper-parameters of the model by using the verification set, and selecting the optimal threshold value.
The threshold is from the set ε, ε can be represented by equation 8, and the optimal threshold is from equation 9.
Wherein:
Δμ(e)＝μ(e)-μ({e∈e|e＜ε})
Δσ(e)＝σ(e)-σ({e∈e|e＜ε})
e a ＝{e∈e|e＞ε)
E seq ＝continoussequences of e a ∈e a (9)
where z represents an ordered set of positive values representing the number of standard deviations above the mean. The value of z depends on the context, but from empirical results it is known that better results are obtained when z is between 3 and 13. This function is also applied to the larger value e of the outliers a And sequence E seq Punishment is carried out, excessive false positive behaviors are avoided, and finally the performance of the model is tested. The performance evaluation indexes of the model are accuracy (Precision, P), Recall (Recall, R) and F1 score (F1 score), as shown in formula 10. Since anomalies often occur in a continuous period of time, if at least one anomaly observation is detected in an anomaly segment, the anomaly segment is considered to be correctly detected.
Wherein TP is true positive, FP is false positive, and FN is false negative. Precision represents the percentage of sequences which are really abnormal in all sequences judged to be abnormal by the model, Recall represents the percentage of all abnormal sequences which are correctly identified as abnormal sequences by the model, and F1 score is the harmonic mean of accuracy and Recall and is a more reasonable performance index. And obtaining the maximum hyperparameter of the F1 fraction adjustment model to obtain the optimal model, thereby realizing the abnormal detection of the fluctuation of the battery production process.
Claims (6)
1. The method for detecting the fluctuation abnormity of the power battery production process based on the space-time diagram model is characterized by comprising the following steps of: the method specifically comprises the following steps:
1) preprocessing key parameters of the battery production process to ensure that parameters with different attributes and different ranges have the same measurement;
2) dividing the preprocessed data into three subsets, where N is 1 The subset is used for training the model, and only a normal data set is contained; n is a radical of 2 The subset is used for adjusting the hyper-parameters of the model and selecting the threshold value; n is a radical of 3 Performance for test model, N 2 And N 3 Both contain normal and abnormal data, the input of the model is a sliding time window X W ∈R k×L Where k represents the characteristic dimension of the multidimensional time series and L represents the size of the sliding time window;
3) constructing a gated space-time map model;
firstly, constructing time convolution networks of convolution kernels with different sizes and convolution with convolution kernels of 1 multiplied by 1 to capture the time dependence of key parameters of the power battery production process;
key parameters of different power battery production processes are expressed by the form of a graph, wherein the graph is expressed as G ═ V, E, wherein V is a set of nodes, and E is a set of edges; k is used for representing the number of nodes in the graph; let V E V denote a node, E ∈ (V, u) ∈ E denote an edge pointing from u to V; and (3) an adjacency matrix of a directed graph of the constructed battery production process parameter system is expressed as A e R k×k If (v) i ，v j ) E is E, then A ij If 1, if
Then A is ij 0; the relationship between linearity and nonlinearity of different production processes is captured by the structure of the adaptive map, and two methods are specifically adopted to learn the adjacency matrix, firstly, according to the battery production processCorrelation of parameters, constructing adjacency matrix
Second, the adaptive adjacency matrix A is learned by the model through gradient descent apt By randomly initializing the embedded representation of the start and target nodes as a learnable parameter E 1 ,E 2 ∈R k×m Wherein m represents the embedded dimension of each node, and the adaptive connection matrix of the target is represented as formula 3; the linear and non-linear relationship of different production processes is captured by the structure of the adaptive graph, the information of relevant process parameters is aggregated by using the graph convolution neural network, the graph convolution neural network is expressed as formula 4,
wherein A is apt Denotes an initialized adaptive adjacency matrix, X denotes the input of the graph convolution neural network, W i1 And W i2 A parameter matrix representing a graph convolution neural network model;
4) filtering the obtained spatiotemporal information using a gate structure;
in the gated space-time diagram model, an output gate g shown in formula 5 is adopted i The output of each gate structure contains the effective information of time series; in formula 6, the output results of each gate structure are added, and the spatio-temporal information of the multi-dimensional time sequence is obtained through two fully-connected layers to predict the observed value of the next time step;
g i ＝h i1 (Θ 1 *X+a)⊙h i2 (Θ 2 *X+b) (5)
wherein h is i1 Sum hi2 represents two operations of the series time convolution and graph convolution neural network, Θ 1 ，Θ 2 Respectively, wherein the parameter indicates multiplication of corresponding elements of the matrix, a, b, c, d are offset entries, W 1 And W 2 Taking Tanh and ReLU as activation functions for a parameter matrix of a full connection layer;
5) training the model by using normal data, adjusting the hyper-parameters of the model by using an automatic threshold strategy and a verification set, and selecting an optimal threshold; and finally, testing the performance of the model intuitively, obtaining a larger prediction error when the test data set contains the abnormality, and judging the observed value of which the prediction error is larger than the threshold value as the abnormality.
2. The method for detecting the fluctuation abnormality of the power battery production process based on the spatio-temporal map model according to claim 1, characterized in that: the making of parameters with different attributes and different ranges have the same measurement specifically includes:
the data for each sensor is normalized using the following formula,
wherein x ∈ R N And (2) a measurement parameter, min (x), max (x) respectively represent the minimum value and the maximum value of x, x' represents the value after x is normalized, and the value of eps is set to be 1e-8, so that the condition of dividing 0 is avoided.
3. The method for detecting the fluctuation abnormality of the power battery production process based on the spatio-temporal map model according to claim 1, characterized in that: the key parameters of the power battery production process comprise rolling speed, cutting length and liquid injection temperature.
4. The method for detecting the fluctuation abnormality of the power battery production process based on the spatio-temporal map model according to claim 1, characterized in that: the sliding time window size is set to 42.
5. The method for detecting the fluctuation abnormality of the power battery production process based on the spatio-temporal map model according to claim 1, characterized in that: wherein the expansion coefficient r is set to 2 or 3.
6. The method for detecting the fluctuation abnormality of the power battery production process based on the spatio-temporal map model according to claim 1, characterized in that: the time convolution network adopts causal convolution and expansion convolution as shown in a formula 2; zero padding is added in the causal convolution, so that the output at the current moment is only related to the historical information and the input at the current moment, and the causality and the autoregressive characteristic of the sequence are ensured; expanding convolution to increase the size of the convolution receptive field, capturing time modes of a time sequence under different receptive fields, and expanding the size of the receptive field along with the depth index of the model to enable the model to obtain more global information; meanwhile, 1 × 1 convolution is added, so that the linear relation learning capability of the model is improved;
where f represents the convolution kernel, K represents the length of the convolution kernel, r represents the expansion coefficient, x' represents the preprocessed data, and t represents time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210547979.9A CN114841076A (en) | 2022-05-18 | 2022-05-18 | Power battery production process fluctuation abnormity detection method based on space-time diagram model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210547979.9A CN114841076A (en) | 2022-05-18 | 2022-05-18 | Power battery production process fluctuation abnormity detection method based on space-time diagram model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114841076A true CN114841076A (en) | 2022-08-02 |
Family
ID=82569801
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210547979.9A Pending CN114841076A (en) | 2022-05-18 | 2022-05-18 | Power battery production process fluctuation abnormity detection method based on space-time diagram model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114841076A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115511013A (en) * | 2022-11-22 | 2022-12-23 | 河海大学 | Large-scale energy storage power station abnormal battery identification method, device and storage medium |
| CN116577671A (en) * | 2023-07-12 | 2023-08-11 | 中国华能集团清洁能源技术研究院有限公司 | Battery system abnormality detection method and device |
| CN117973904A (en) * | 2024-03-29 | 2024-05-03 | 深圳市联特微电脑信息技术开发有限公司 | Intelligent manufacturing capacity analysis method and system |
| CN117973904B (en) * | 2024-03-29 | 2024-06-07 | 深圳市联特微电脑信息技术开发有限公司 | Intelligent manufacturing capacity analysis method and system |
-
2022
- 2022-05-18 CN CN202210547979.9A patent/CN114841076A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115511013A (en) * | 2022-11-22 | 2022-12-23 | 河海大学 | Large-scale energy storage power station abnormal battery identification method, device and storage medium |
| CN116577671A (en) * | 2023-07-12 | 2023-08-11 | 中国华能集团清洁能源技术研究院有限公司 | Battery system abnormality detection method and device |
| CN116577671B (en) * | 2023-07-12 | 2023-09-29 | 中国华能集团清洁能源技术研究院有限公司 | Battery system abnormality detection method and device |
| CN117973904A (en) * | 2024-03-29 | 2024-05-03 | 深圳市联特微电脑信息技术开发有限公司 | Intelligent manufacturing capacity analysis method and system |
| CN117973904B (en) * | 2024-03-29 | 2024-06-07 | 深圳市联特微电脑信息技术开发有限公司 | Intelligent manufacturing capacity analysis method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Capozzoli et al. | Fault detection analysis using data mining techniques for a cluster of smart office buildings | |
| Gao et al. | A neural network-based joint prognostic model for data fusion and remaining useful life prediction | |
| CN114841076A (en) | Power battery production process fluctuation abnormity detection method based on space-time diagram model | |
| CN111340282B (en) | DA-TCN-based method and system for estimating residual service life of equipment | |
| CN110757510B (en) | Method and system for predicting remaining life of robot | |
| Remadna et al. | Leveraging the power of the combination of CNN and bi-directional LSTM networks for aircraft engine RUL estimation | |
| CN114282443B (en) | Residual service life prediction method based on MLP-LSTM supervised joint model | |
| Zhang et al. | A novel fault diagnosis method based on stacked lstm | |
| CN116683588B (en) | Lithium ion battery charge and discharge control method and system | |
| CN113780684A (en) | Intelligent building user energy consumption behavior prediction method based on LSTM neural network | |
| Lim et al. | A novel time series-histogram of features (TS-HoF) method for prognostic applications | |
| CN117273440A (en) | Engineering construction Internet of things monitoring and managing system and method based on deep learning | |
| CN114500004A (en) | Anomaly detection method based on conditional diffusion probability generation model | |
| Huang et al. | Bayesian neural network based method of remaining useful life prediction and uncertainty quantification for aircraft engine | |
| Zhu et al. | Long Short Term Memory Networks Based Anomaly Detection for KPIs. | |
| De Vita et al. | A semi-supervised bayesian anomaly detection technique for diagnosing faults in industrial iot systems | |
| Dang et al. | seq2graph: Discovering dynamic non-linear dependencies from multivariate time series | |
| CN114486262B (en) | CNN-AT-LSTM-based bearing residual service life prediction method | |
| CN115496384A (en) | Monitoring management method and device for industrial equipment and computer equipment | |
| Liu et al. | Prognostics for rotating machinery using variational mode decomposition and long short-term memory network | |
| CN113469013A (en) | Motor fault prediction method and system based on transfer learning and time sequence | |
| CN113569460A (en) | Real vehicle fuel cell system state multi-parameter prediction method and device | |
| CN112232557A (en) | Switch machine health degree short-term prediction method based on long-term and short-term memory network | |
| Yan et al. | Nonlinear quality-relevant process monitoring based on maximizing correlation neural network | |
| Su et al. | ADCMO: an anomaly detection approach based on local outlier factor for continuously monitored object |
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 |