CN116774089A - Convolutional neural network battery state of health estimation method and system based on feature fusion - Google Patents
Convolutional neural network battery state of health estimation method and system based on feature fusion Download PDFInfo
- Publication number
- CN116774089A CN116774089A CN202310762725.3A CN202310762725A CN116774089A CN 116774089 A CN116774089 A CN 116774089A CN 202310762725 A CN202310762725 A CN 202310762725A CN 116774089 A CN116774089 A CN 116774089A
- Authority
- CN
- China
- Prior art keywords
- health
- data
- estimation
- state
- feature
- 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
- 230000036541 health Effects 0.000 title claims abstract description 115
- 230000004927 fusion Effects 0.000 title claims abstract description 70
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 41
- 238000000605 extraction Methods 0.000 claims abstract description 45
- 238000005070 sampling Methods 0.000 claims abstract description 30
- 238000012545 processing Methods 0.000 claims abstract description 25
- 238000007781 pre-processing Methods 0.000 claims abstract description 18
- 238000011176 pooling Methods 0.000 claims description 16
- 238000004422 calculation algorithm Methods 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 6
- 238000013461 design Methods 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- HBBGRARXTFLTSG-UHFFFAOYSA-N Lithium ion Chemical compound [Li+] HBBGRARXTFLTSG-UHFFFAOYSA-N 0.000 abstract description 9
- 229910001416 lithium ion Inorganic materials 0.000 abstract description 9
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 23
- 230000015556 catabolic process Effects 0.000 description 8
- 238000006731 degradation reaction Methods 0.000 description 8
- 230000004913 activation Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 238000012549 training Methods 0.000 description 6
- 238000013136 deep learning model Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000003993 interaction Effects 0.000 description 4
- 230000032683 aging Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 229920002430 Fibre-reinforced plastic Polymers 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 239000002131 composite material Substances 0.000 description 2
- 238000007599 discharging Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000011151 fibre-reinforced plastic Substances 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003990 capacitor Substances 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000007429 general method Methods 0.000 description 1
- 230000003862 health status Effects 0.000 description 1
- 238000001453 impedance spectrum Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000008929 regeneration Effects 0.000 description 1
- 238000011069 regeneration method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/392—Determining battery ageing or deterioration, e.g. state of health
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/367—Software therefor, e.g. for battery testing using modelling or look-up tables
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/27—Regression, e.g. linear or logistic regression
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Secondary Cells (AREA)
Abstract
The application relates to the field of artificial intelligence technology application, in particular to a convolutional neural network battery state of health estimation method and system based on feature fusion, comprising the following steps: acquiring sampling data of a battery to be subjected to state of health estimation; carrying out one-dimensional feature extraction processing on the sampled data to obtain one-dimensional features after preprocessing a plurality of data; the method comprises the steps of constructing a health state estimation model based on a convolutional neural network, wherein the health state estimation model comprises a feature extraction module, a feature fusion module and a regression estimation module; the feature fusion module performs feature fusion on the feature graphs generated by the feature extraction module in a point-by-point addition mode; the regression estimation module is used for extracting information after feature fusion; and inputting the one-dimensional characteristics into an estimation model to estimate the state of health of the battery, and obtaining an estimation result. According to the application, by performing various voltage characteristic expansion on the original voltage data and designing a new convolutional neural network structure with a characteristic fusion layer, the accuracy of estimating the health state of the lithium ion battery is improved.
Description
Technical Field
The application relates to the field of artificial intelligence technology application, in particular to a convolutional neural network battery state of health estimation method and system based on feature fusion.
Background
Lithium ion batteries have the advantages of high energy density, high power density, long cycle life, low self-discharge and the like, have been rapidly developed in recent years, and are widely used in various fields such as electric automobiles, mobile phones, other mobile electronic devices and the like. Performance degradation of lithium ion batteries typically occurs upon aging, primarily as a result of reduced maximum available capacity and power decay of the battery. Generally, a battery is considered to be at the end of its life when its current capacity falls below 80% of its initial capacity. In particular, as the battery approaches the end of life, the available capacity of the battery will decrease rapidly.
In order to ensure the safety, reliability, etc. of the battery system as a whole, the battery management system needs to perform accurate and reliable health status monitoring on the battery. The state of health is generally defined in a battery management system as an indicator of the state of health of the battery and can be calculated by the ratio of the current maximum available capacity of the battery to its initial nominal value. Accurate battery state of health estimation is important to monitor the current state of the battery to ensure safe operation of the battery and that the battery is replaced in time before the end of life is reached.
One common approach to battery state of health estimation is to estimate the battery state of health factor by collecting voltage, current, and temperature data generated during battery operation. This is because it is difficult to directly obtain a battery state of health estimation factor in a real environment, and it is necessary to calculate the current maximum capacity value of the battery after recording data of the battery from a full charge state to a full discharge state at a fixed discharge rate. In practical engineering application, the lithium ion battery has complex use scene, uncertainty in data acquisition and complex nonlinear aging characteristic of the battery, and the method brings great challenges to accurate estimation of the health state of the lithium ion battery.
Current methods of battery state of health estimation are largely classified into model-based methods and data-driven based methods. Among them, the model-based methods can be largely classified into methods based on an equivalent circuit model, an electrochemical model, and an empirical degradation model. In equivalent circuit models, the model is typically composed of voltage sources, resistors, capacitors, and other components to simulate the state of the battery, such as RC equivalent circuit models, fractional equivalent circuit models, and impedance spectrum growth models, and combined with filtering algorithms to estimate the state of health of the battery. The equivalent circuit model is simple in general method principle and small in calculated amount, but the estimation accuracy is limited. Whereas in electrochemical model-based methods, the internal physical and chemical mechanisms of cell degradation are typically described by a series of partial differential equations, such as pseudo-two-dimensional models (P2D) and simplified models thereof; but in general the method has difficult model parameter extraction and complex equation calculation. In the empirical degradation model, an empirical expression is established for battery aging behavior by analyzing the battery historical capacity fade data, which can be used to describe the overall capacity degradation trend of the battery. Although the parameter identification in the empirical degradation model is simple, it is difficult to explain the battery capacity regeneration phenomenon, i.e., the local fluctuation of the battery capacity. In general, since different chemical reactions occur inside the battery under different operating conditions, it is difficult to build an accurate battery model by a model-based method, and simply estimating model parameters results in limited estimation accuracy of the state of health of the battery. (the whole section can be optimized again)
In recent years, students at home and abroad perform a series of work on estimating the health state of lithium ion batteries based on a data-driven method. These methods are based on a large amount of battery operating data without prior knowledge of the chemical mechanism inside the battery to make a battery state of health estimate. These methods can be categorized into two types, traditional machine learning models and deep learning models. In conventional machine learning models, such as support vector machines, gaussian process regression, random forests, etc., it is often necessary to manually design features related to battery health. For example, characteristics related to battery state of health may be designed based on voltage data, including end voltage values, fixed voltage interval charge-discharge times, slope/area/radius of curvature of the voltage curve, and the like. Conventional machine learning methods are generally simple and interpretable, however, only information from a single data point is typically utilized in manually designed extracted features, and feature information from consecutive sequence data is lost, thereby affecting the effectiveness of battery state of health estimation. In the deep learning model, a convolutional neural network, a long-term and short-term memory network and the like are generally used for automatically extracting characteristics of mass data, and training of the model is completed in an end-to-end mode, so that the problem of insufficient utilization of data in manually extracting the characteristics is solved.
The convolutional neural network has the characteristics of sparse connection, weight sharing and pooling operation, is beneficial to reducing the number of model parameters, complexity of a model, relieving the problems of overfitting and the like, and has better model generalization capability. The basic structure of the convolutional neural network comprises a convolutional layer, a pooling layer and a full-connection layer, and the performance of the convolutional neural network with different topological structures formed by different network layers is different.
In the field of lithium ion battery health state estimation, the following problems mainly exist in the application of convolutional neural networks:
(1) Most models of existing documents lack the ability to learn, when a variety of features are used as inputs, the lack of inherent relationships between features by further designing the corresponding structure, which is expressed as: all features directly form a matrix as input to enter the convolutional neural network structure, and then all features are processed in the same structure. From the structural point of view of convolutional neural network, the characteristic of weight sharing is that the link between the features is learned by a model. However, this method of construction is too implicit and there is still room for further optimization.
(2) Most of the research on the neural network input is often lack of enough data preprocessing on the input characteristic data, so that the learning effect of the network structure on the input characteristics of the original data is still insufficient. Although the deep learning model usually adopts an end-to-end model structure, the deep learning model is limited by the learning capability of the existing model structure, and the quality of a learning sample is improved, so that the learning effect of the model on the data characteristics can be effectively improved.
Disclosure of Invention
In order to solve the problems in the prior art, the application provides a convolutional neural network battery state of health estimation method and system based on feature fusion, which improves the accuracy of lithium ion battery state of health estimation by expanding various voltage features of original voltage data and designing a new convolutional neural network structure with a feature fusion layer.
The estimation method of the application adopts the following technical scheme: a convolutional neural network battery state of health estimation method based on feature fusion comprises the following steps:
acquiring sampling data of a battery to be subjected to state of health estimation;
carrying out one-dimensional feature extraction processing on the sampling data to obtain one-dimensional features after preprocessing a plurality of data;
building a health state estimation model based on a convolutional neural network, wherein the health state estimation model comprises a feature extraction module, a feature fusion module and a regression estimation module; the feature fusion module performs feature fusion on the feature graphs generated by the feature extraction module in a point-by-point addition mode; the regression estimation module is used for extracting information after feature fusion;
and inputting the one-dimensional characteristics into the health state estimation model to estimate the health state of the battery, and obtaining a health state estimation result.
Preferably, the one-dimensional feature extraction processing is performed on the sampled data to obtain a plurality of one-dimensional features after data preprocessing, including:
performing feature extraction processing on the sampling data to obtain discharge capacity data and voltage data in a plurality of cycle periods;
constructing and obtaining a first function curve according to the discharge capacity data and the voltage data;
performing differential processing on the voltage data according to the discharge capacity data to obtain the change rate of the discharge capacity data as a second function curve;
and obtaining a third function curve according to the difference value between the discharge capacity data in each cycle period and the discharge capacity data in the first cycle period.
Preferably, the feature extraction module comprises three parallel sub-model structures, which are used for respectively carrying out independent feature extraction on each input one-dimensional feature to obtain a corresponding feature map; in the feature fusion module, feature graphs from three sub-model structures are subjected to point-by-point bitwise addition operation, so that feature fusion among inputs is realized.
The estimation system of the application adopts the following technical scheme: a convolutional neural network battery state of health estimation system based on feature fusion comprises the following modules:
the data sampling module is used for acquiring sampling data of the battery to be subjected to state of health estimation;
the preprocessing module is used for carrying out one-dimensional feature extraction processing on the sampling data to obtain one-dimensional features after preprocessing a plurality of data;
the model construction module is used for constructing a health state estimation model based on the convolutional neural network, and the health state estimation model comprises a feature extraction module, a feature fusion module and a regression estimation module; the feature fusion module performs feature fusion on the feature graphs generated by the feature extraction module in a point-by-point addition mode; the regression estimation module is used for extracting information after feature fusion;
and the state estimation module is used for inputting the one-dimensional characteristics into the health state estimation model to estimate the health state of the battery, so as to obtain a health state estimation result.
Compared with the prior art, the application has the following advantages and effects:
1. firstly, based on the collected voltage data generated by the battery during operation, performing various voltage characteristic expansion as data preprocessing on the original voltage data, so that the original data is more effectively applied; and then, the feature fusion module is designed in the convolutional neural network model, so that the internal relation among the features can be better learned. By expanding various voltage characteristics of the original voltage data and newly designing a convolutional neural network structure with a characteristic fusion layer, the accuracy of estimating the health state of the lithium ion battery is improved.
2. On one hand, if the feature fusion module is placed at a position of a front part of the model, the feature of each input can not be fully extracted during feature fusion; on the other hand, if the feature fusion module is placed at a later position of the model, namely in a deeper network structure, the information learned by the model is often more abstract semantic information, and the feature fusion of the model is more difficult. Based on the method, the feature fusion module is placed in a middle position of the whole model.
3. Because of adopting a plurality of preprocessed one-dimensional voltage characteristics, the information input into the convolutional neural network model is richer than that of the convolutional neural network model which only uses the original voltage characteristics as input, and the model is helped to learn more information.
Drawings
Fig. 1 is a flowchart of a battery state of health estimation method based on feature fusion provided by an embodiment of the present application;
FIG. 2 is a network architecture diagram of a health estimation model provided by an embodiment of the present application;
FIG. 3 is a specific parameter setting of a network architecture diagram of a health estimation model provided by an embodiment of the present application;
fig. 4 is a diagram showing a comparison of a model health state estimation value and a true value according to an embodiment of the present application.
Detailed Description
The present application will be described in further detail with reference to examples and drawings, but embodiments of the present application are not limited thereto.
Example 1
Referring to fig. 1, the embodiment provides a convolutional neural network battery state of health estimation method based on feature fusion, which includes the following steps:
s101, acquiring sampling data of a battery to be subjected to state of health estimation.
In this embodiment, the step performs data acquisition processing during the discharging process of the battery to be estimated of the state of health to obtain discharging data; and carrying out discharge capacity and voltage data sampling processing on the discharge data to obtain sampling data.
In this embodiment, the estimation of the battery state of health is directly and quickly achieved by building a deep learning model using the observed data generated by the battery during operation. The data generated during the operation of the battery may include discharge data such as current, voltage, capacity, internal resistance, temperature, etc., detected by the battery capacity tester, and the discharge capacity and voltage data are sampled from the discharge data to obtain sampling data.
It should be noted that, during actual operation, it is difficult to obtain all the information of the battery from the full charge state to the full discharge state, because in practical application, it is difficult for the battery to reach the full discharge state. At this time, battery operation data can be collected uniformly by setting a fixed voltage interval.
S102, carrying out one-dimensional feature extraction processing on the sampling data to obtain one-dimensional features after preprocessing a plurality of data. In this embodiment, the method specifically includes the following steps:
performing feature extraction processing on the sampling data to obtain discharge capacity data and voltage data in a plurality of cycle periods;
constructing and obtaining a first function curve according to the discharge capacity data and the voltage data;
performing differential processing on the voltage data according to the discharge capacity data to obtain the change rate of the discharge capacity data as a second function curve;
and obtaining a third function curve according to the difference value between the discharge capacity data in each cycle period and the discharge capacity data in the first cycle period.
In the present embodiment, the discharge capacity and voltage data in one cycle period of the battery are extracted from the sampling data of the battery to be estimated of the state of health, respectively using Q d And v represents. Estimation of the state of health to be treatedThe discharge capacity is regarded as a function of the discharge voltage in one cycle period of the battery, and the discharge voltage is taken as the horizontal axis. For convenience of description, use Q 1 (v) Time sequence data representing discharge capacity-discharge voltage curve of the battery in one cycle period is obtained to obtain a first function curve. Next, calculating the first differential of discharge capacity to voltage in one discharge period to obtain discharge capacity Q d The rate of change of (i.e. the second function curve) is denoted as Q 2 (v) A. The application relates to a method for producing a fibre-reinforced plastic composite The calculation formula of the second function curve is as follows:
wherein j represents the j-th sampling point, Q d (j) Discharge capacity data indicating the jth sampling point, and v (j) indicates voltage data of the jth sampling point.
In addition, the variation of the discharge capacity and the voltage data along with the cycle period is considered, the discharge capacity data is subjected to difference processing, the time sequence discharge capacity data of each cycle period and the 1 st cycle period in the service life of the battery to be predicted is subjected to difference processing, a third function curve is obtained, and the result is recorded as Q 3 (v) A. The application relates to a method for producing a fibre-reinforced plastic composite The calculation formula of the third function curve is as follows:
Q 3 ＝Q di -Q d1
wherein Q is di And Q d1 Discharge capacity curve data for the i-th and 1-th cycle periods are shown, respectively.
Since the amount of data collected in the actual sampled data for the battery during different cycle periods may be different, the data generated during operation of the battery within a fixed voltage interval will be taken together for unified training of the subsequent model. Meanwhile, 100 equidistant voltage sampling points are arranged between the upper limit and the lower limit of the battery discharge voltage, and corresponding discharge capacity data of the corresponding voltage sampling points are obtained in an interpolation fitting mode. Finally, three voltage-dependent function curves Q 1 (v)、Q 2 (v) And Q 3 (v) Is three 1 x 100 one-dimensional data.
S103, constructing a health state estimation model based on a convolutional neural network, wherein the health state estimation model comprises a feature extraction module, a feature fusion module and a regression estimation module, and the convolutional neural network adopts an autonomously constructed network architecture. The method comprises the following steps:
constructing a feature extraction module, wherein the feature extraction module comprises a plurality of convolution layers and a pooling layer;
a feature fusion module is constructed, wherein the feature fusion module adopts a method of adding the feature images in a point-by-point mode to perform feature fusion on a plurality of feature images generated by the feature extraction module;
constructing a regression estimation module, wherein the regression estimation module comprises a plurality of full-connection layers and is used for extracting information after feature fusion;
and constructing and obtaining a battery health state estimation model according to the feature extraction module, the feature fusion module and the regression estimation module.
In the embodiment of the application, the model is based on a convolutional neural network structure for feature extraction. The convolutional neural network is a feed-forward multilayer network, and has excellent performance for extracting features in high-dimensional data. The method is characterized by comprising sparse connection, weight sharing and pooling operation, so that the convolutional neural network structure can greatly reduce the training quantity of parameters, the complexity of a model and the risk of overfitting, and better generalization capability is obtained. The convolutional neural network structure basically comprises a convolutional layer, a pooling layer and a full-connection layer.
In this embodiment, the battery state of health estimation is performed by autonomously constructing a battery state of health estimation model based on a convolutional neural network, the model structure is referred to fig. 2, and the corresponding network structure parameters are referred to fig. 3. The input of the model is three one-dimensional data of 1 x 100, and is input into the model through z-score standardization, and finally the output is the estimated value of the battery state of health.
In this embodiment, a battery state of health estimation model is built based on a convolutional neural network, and a feature extraction module, namely the first 6 layers of the model, is first built, and specific parameter settings refer to L1-L6 of fig. 3. The feature extraction module comprises three parallel sub-model structures, which are used for respectively carrying out independent feature extraction on each input one-dimensional feature to obtain a corresponding feature map, thereby reducing the problem that when all the features are processed together to input the model, the attention of the model on certain features is not enough. Each sub-model structure has 6 layers, and comprises a plurality of convolution layers and a maximum pooling layer arranged behind the convolution layers, so that the structure can effectively extract low-level features. Meanwhile, the convolution kernel size of each convolution layer in the sub-model structure is 1*3, the step size is 1, and the design of a plurality of small convolution kernels is considered to increase the depth of a network without increasing the parameter number of the model, so that the model has better performance. In the maximum pooling layer of the submodel, an overlapping pooling strategy is adopted, so that the generalization capability of the model can be improved, the window size is 1*3, and the step length is 2.
In the feature fusion module of the present embodiment, feature graphs from the previous three sub-model structures will be subjected to a point-wise bitwise addition operation at L7 of fig. 3, thereby achieving feature fusion between inputs. The point-by-point bit addition operation is used as a connection mode to connect different layers as fusion of characteristics. In the feature fusion module of the present embodiment, the feature fusion module is derived from Q 1 (v)、Q 2 (v) And Q 3 (v) The feature maps of (2) will be added so that the information of the multiple features is interacted with to improve the performance of the model. In L6 of FIG. 3, Q 1 (v)、Q 2 (v) And Q 3 (v) Each generates a corresponding characteristic diagram X i 、Y i And Z i The feature maps are all a×b in size, which is the basis between which point-by-point bitwise addition can be performed; through the feature fusion of L7, the corresponding Q 1 (v)、Q 2 (v) And Q 3 (v) Will generate a new feature map R i ＝X i +Y i +Z i As an information interaction and supplement, the information expression capability of the feature map is enriched, because the new feature map is directly connected with the previous Q 1 (v)、Q 2 (v) And Q 3 (v) The extracted feature graphs are related, and the information interaction of a plurality of features is realized in a direct and simple mode, so that the performance of the network is improved. Without the point-wise bitwise addition operation of L7 of FIG. 3, the information interaction between them can only check all of the preceding bits by convolutions of the convolutions layerAnd carrying out information interaction when the information of the feature map is processed. Whereas in the latter two convolutions, referring to L8-L9 of fig. 3, it will be used to further extract the information after fusion of the previous three input features.
In this embodiment, a zero-fill strategy is applied in all convolution and pooling layers to counteract the effects of size shrinkage in the feature map computation. After each convolution layer, a rectifying linear unit (RELU) is introduced as an activation function, the nonlinear characteristics of which can be used to saturate the output or limit the generated output.
In this embodiment, a battery state of health estimation model is built based on a convolutional neural network, and specific parameter settings refer to fig. 3. First, a feature extraction module is constructed, which has three identical sub-models, 6 layers in total. The first convolution layer has a convolution kernel size of 1*3, 6 convolution kernels with a step size of 1, and then a RELU activation function is followed to obtain a feature map with a size of 1 x 100 x 6. Next, a second convolution layer is passed, the convolution kernel size is 1*3, there are 6 convolution kernels, the step size is 1, and then a RELU activation function is followed, resulting in a feature map size of 1×100×6. Then the first maximum pooling layer is passed, the pooling window size of the layer is 1*3, the step length is 2, and after pooling operation, the output size is 1×50×6. Then, through the third convolution layer, the convolution kernel size is 1*3, there are 16 convolution kernels, the step size is 1, and then a RELU activation function is connected to obtain the feature map size is 1×50×16. Then, through the fourth convolution layer, the convolution kernel size is 1*3, there are 16 convolution kernels, the step size is 1, and then a RELU activation function is followed to obtain the feature map size is 1×50×16. Then through the second maximum pooling layer, the pooling window size is 1*3, the step size is 2, and the output size is 1×25×6. The feature fusion module is constructed, the feature images generated by the sub-modules with the same size are added point by point and bit by bit according to the corresponding positions, and the output size of the obtained new feature image is 1 x 25 x 6. Next, a fifth convolution layer is passed, with a convolution kernel size of 1*3, 128 convolution kernels, step size of 1, followed by a RELU activation function, resulting in a feature map size of 1 x 25 x 128. Then, through the sixth convolution layer, the convolution kernel size is 1*3, there are 128 convolution kernels, the step size is 1, and then a RELU activation function is followed to obtain the feature map size of 1×25×128. And finally, constructing a regression estimation module, wherein the regression estimation module comprises a first full-connection layer and a second full-connection layer. The feature map obtained after the sixth convolution layer is accessed to the first full connection layer, and the layer contains 10 neuron nodes. This is followed by accessing a second fully connected layer, which has 1 neuron node. The regression estimation module is used for extracting information after feature fusion, converting the two-dimensional feature map output by convolution into a one-dimensional vector through the full connection layer, and finally outputting an estimated health state value. Each node of the full-connection layer is connected with all nodes of the upper layer to play a role of combining all features of the upper layer, so that information after feature fusion can be effectively extracted.
S104, inputting the one-dimensional features into the health state estimation model to estimate the health state of the battery, and obtaining a health state estimation result.
Preferably, before the battery state of health estimation in this step, the method for estimating a state of health of the present embodiment further includes training a battery state of health estimation model in advance, specifically including:
and performing optimization processing on model parameters of the battery state of health estimation model according to an optimization algorithm to obtain a trained battery state of health estimation model.
In this embodiment, the state of health estimation model also needs to be trained and tested before it can estimate the state of health of the battery. The optimization algorithm trained by the health state estimation model selects an Adam algorithm, and compared with a random gradient descent algorithm, the Adam algorithm increases a first moment and a second moment and sets specific self-adaptive learning rate for different parameters. The initial learning rate was set to 0.001, and after 100 training iterations, the learning rate was set to 0.00001, and the number of training iterations was 200 in total. Then, an experimental test model is performed on the battery degradation data set, and experiments show that the battery state of health estimation model has reliable estimation accuracy in battery state of health estimation, and an estimation result is shown in fig. 4. After the battery state of health estimation model is trained and tested, the battery state of health estimation can be directly carried out by utilizing the data generated when the battery is currently operated.
Referring to fig. 1, an estimation procedure according to an embodiment of the present application specifically includes: and performing feature expansion on the sampled data of the battery to be estimated in the state of health, so that single features of the original data are changed into three features containing more information, and the three features are input into a battery state of health estimation model built on the basis of a convolutional neural network to obtain a reliable battery state of health estimation result.
In the embodiment of the application, the battery health state estimation result with excellent effect can be obtained only by collecting less than 40% of complete discharge data in the current cycle period of the battery. On experimental data, the average absolute error and the average absolute percentage error are respectively not more than 0.0028 and 0.32%, so that the accuracy of the battery health state estimation result is improved.
In summary, the embodiment of the application has the following advantages:
the embodiment of the application is applied to the field of battery health state estimation based on the convolutional neural network with excellent performance in the fields of computer vision, natural language processing, semantic segmentation and the like. In order to fully utilize the learning efficiency of the model when a plurality of features are used as model input, the embodiment of the application specially designs a feature fusion module to realize further communication among feature information. Meanwhile, in order to fully utilize the collected original discharge voltage data, the data are expanded into three voltage characteristics, and the input of a model is enriched. The data preprocessing operation emphasized by the embodiment of the application and the method for further learning the internal connection of a plurality of input features enable the accuracy of battery health state estimation to be further improved. To evaluate its technical effect, embodiments of the present application use the proposed evaluation method to conduct experiments on battery degradation data sets that only require less than 40% of the complete discharge data collected during the current cycle of the battery, wherein the average absolute error is no more than 0.0028 on average and the average absolute percentage error is no more than 0.32 on average.
Example 2
Based on the same inventive concept as embodiment 1, this embodiment provides a convolutional neural network battery state of health estimation system based on feature fusion, including the following modules:
the data sampling module is used for acquiring sampling data of the battery to be subjected to state of health estimation;
the preprocessing module is used for carrying out one-dimensional feature extraction processing on the sampling data to obtain one-dimensional features after preprocessing a plurality of data;
the model construction module is used for constructing a health state estimation model based on the convolutional neural network, and the health state estimation model comprises a feature extraction module, a feature fusion module and a regression estimation module; the feature fusion module performs feature fusion on the feature graphs generated by the feature extraction module in a point-by-point addition mode; the regression estimation module is used for extracting information after feature fusion;
and the state estimation module is used for inputting the one-dimensional characteristics into the health state estimation model to estimate the health state of the battery, so as to obtain a health state estimation result.
The preprocessing process of the preprocessing module comprises the following steps:
performing feature extraction processing on the sampling data to obtain discharge capacity data and voltage data in a plurality of cycle periods;
constructing and obtaining a first function curve according to the discharge capacity data and the voltage data;
performing differential processing on the voltage data according to the discharge capacity data to obtain the change rate of the discharge capacity data as a second function curve;
and obtaining a third function curve according to the difference value between the discharge capacity data in each cycle period and the discharge capacity data in the first cycle period.
In this embodiment, the feature extraction module includes three parallel sub-model structures, which are used to perform independent feature extraction on each input one-dimensional feature to obtain a corresponding feature map; in the feature fusion module, feature graphs from three sub-model structures are subjected to point-by-point bitwise addition operation, so that feature fusion among inputs is realized.
The modules of this embodiment are accordingly used to implement the corresponding steps of embodiment 1, see embodiment 1 for a detailed implementation thereof.
The above examples are preferred embodiments of the present application, but the embodiments of the present application are not limited to the above examples, and any other changes, modifications, substitutions, combinations, and simplifications that do not depart from the spirit and principle of the present application should be made in the equivalent manner, and the embodiments are included in the protection scope of the present application.
Claims (10)
1. The convolutional neural network battery state of health estimation method based on feature fusion is characterized by comprising the following steps:
acquiring sampling data of a battery to be subjected to state of health estimation;
carrying out one-dimensional feature extraction processing on the sampling data to obtain one-dimensional features after preprocessing a plurality of data;
building a health state estimation model based on a convolutional neural network, wherein the health state estimation model comprises a feature extraction module, a feature fusion module and a regression estimation module; the feature fusion module performs feature fusion on the feature graphs generated by the feature extraction module in a point-by-point addition mode; the regression estimation module is used for extracting information after feature fusion;
and inputting the one-dimensional characteristics into the health state estimation model to estimate the health state of the battery, and obtaining a health state estimation result.
2. The battery state of health estimation method according to claim 1, wherein performing one-dimensional feature extraction processing on the sampled data to obtain a plurality of data-preprocessed one-dimensional features, comprises:
performing feature extraction processing on the sampling data to obtain discharge capacity data and voltage data in a plurality of cycle periods;
constructing and obtaining a first function curve according to the discharge capacity data and the voltage data;
performing differential processing on the voltage data according to the discharge capacity data to obtain the change rate of the discharge capacity data as a second function curve;
and obtaining a third function curve according to the difference value between the discharge capacity data in each cycle period and the discharge capacity data in the first cycle period.
3. The battery state of health estimation method according to claim 1 or 2, wherein the feature extraction module comprises three parallel sub-model structures for performing separate feature extraction on each input one-dimensional feature to obtain a corresponding feature map; in the feature fusion module, feature graphs from three sub-model structures are subjected to point-by-point bitwise addition operation, so that feature fusion among inputs is realized.
4. The battery state of health estimation method of claim 3, wherein each sub-model structure comprises a plurality of convolution layers and a max pooling layer disposed after the convolution layers to efficiently extract low-level features; each convolution layer in the sub-model structure employs a design of multiple small convolution kernels to increase the depth of the network without increasing the number of parameters of the model.
5. The method according to claim 1, wherein before estimating the state of health of the battery, the state of health estimation method further performs optimization processing on model parameters of the state of health estimation model according to an optimization algorithm to obtain a trained state of health estimation model of the battery.
6. The battery state of health estimation method of claim 5, wherein said optimization algorithm selects Adam algorithm.
7. The convolutional neural network battery state of health estimation system based on feature fusion is characterized by comprising the following modules:
the data sampling module is used for acquiring sampling data of the battery to be subjected to state of health estimation;
the preprocessing module is used for carrying out one-dimensional feature extraction processing on the sampling data to obtain one-dimensional features after preprocessing a plurality of data;
the model construction module is used for constructing a health state estimation model based on the convolutional neural network, and the health state estimation model comprises a feature extraction module, a feature fusion module and a regression estimation module; the feature fusion module performs feature fusion on the feature graphs generated by the feature extraction module in a point-by-point addition mode; the regression estimation module is used for extracting information after feature fusion;
and the state estimation module is used for inputting the one-dimensional characteristics into the health state estimation model to estimate the health state of the battery, so as to obtain a health state estimation result.
8. The battery state of health estimation system of claim 7, wherein the preprocessing process of the preprocessing module comprises:
performing feature extraction processing on the sampling data to obtain discharge capacity data and voltage data in a plurality of cycle periods;
constructing and obtaining a first function curve according to the discharge capacity data and the voltage data;
performing differential processing on the voltage data according to the discharge capacity data to obtain the change rate of the discharge capacity data as a second function curve;
and obtaining a third function curve according to the difference value between the discharge capacity data in each cycle period and the discharge capacity data in the first cycle period.
9. The battery state of health estimation system of claim 7, wherein the feature extraction module comprises three parallel sub-model structures for performing separate feature extraction on each input one-dimensional feature to obtain a corresponding feature map; in the feature fusion module, feature graphs from three sub-model structures are subjected to point-by-point bitwise addition operation, so that feature fusion among inputs is realized.
10. The battery state of health estimation system of claim 8, wherein each sub-model structure comprises a plurality of convolution layers and a max pooling layer disposed after the convolution layers to efficiently extract low-level features; each convolution layer in the sub-model structure employs a design of multiple small convolution kernels to increase the depth of the network without increasing the number of parameters of the model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310762725.3A CN116774089A (en) | 2023-06-27 | 2023-06-27 | Convolutional neural network battery state of health estimation method and system based on feature fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310762725.3A CN116774089A (en) | 2023-06-27 | 2023-06-27 | Convolutional neural network battery state of health estimation method and system based on feature fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116774089A true CN116774089A (en) | 2023-09-19 |
Family
ID=87985610
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310762725.3A Pending CN116774089A (en) | 2023-06-27 | 2023-06-27 | Convolutional neural network battery state of health estimation method and system based on feature fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116774089A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117572250A (en) * | 2024-01-17 | 2024-02-20 | 山东工商学院 | Method for estimating SOH of battery based on multi-feature fusion and XGBoost |
-
2023
- 2023-06-27 CN CN202310762725.3A patent/CN116774089A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117572250A (en) * | 2024-01-17 | 2024-02-20 | 山东工商学院 | Method for estimating SOH of battery based on multi-feature fusion and XGBoost |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113777496B (en) | Lithium ion battery residual life prediction method based on time convolution neural network | |
| CN112381673B (en) | Park electricity utilization information analysis method and device based on digital twin | |
| CN113821875B (en) | Intelligent vehicle fault real-time prediction method and system based on end cloud cooperation | |
| CN111856287A (en) | Lithium battery health state detection method based on stacked residual causal convolutional neural network | |
| CN110232476A (en) | A kind of micro-grid load prediction technique based on deep learning | |
| CN113093014B (en) | Online collaborative estimation method and system for SOH and SOC based on impedance parameters | |
| CN115758290A (en) | Fan gearbox high-speed shaft temperature trend early warning method based on LSTM | |
| CN116774089A (en) | Convolutional neural network battery state of health estimation method and system based on feature fusion | |
| CN114692827A (en) | Electric vehicle lithium battery SOH online prediction method facing edge federal learning | |
| CN112666479A (en) | Battery service life prediction method based on charging cycle fusion | |
| CN115409263A (en) | Method for predicting remaining life of lithium battery based on gating and attention mechanism | |
| CN115481796A (en) | Method for predicting remaining service life of battery based on Bayesian hybrid neural network | |
| CN117031310A (en) | Method for predicting residual service life of power battery of electric automobile | |
| CN115859792A (en) | Medium-term power load prediction method and system based on attention mechanism | |
| CN113449919A (en) | Power consumption prediction method and system based on feature and trend perception | |
| CN115656824A (en) | Lithium battery nuclear power state prediction method based on CNN-LSTM model | |
| CN115542182A (en) | SOH estimation method for series battery pack of mobile energy storage system | |
| CN115236522A (en) | End-to-end capacity estimation method of energy storage battery based on hybrid deep neural network | |
| CN117825965A (en) | State prediction method and system for lithium ion battery | |
| CN116960487B (en) | Sodium ion battery system capacity estimation method and device considering monomer inconsistency | |
| CN117113054A (en) | Multi-element time sequence prediction method based on graph neural network and transducer | |
| CN113158134B (en) | Method, device and storage medium for constructing non-invasive load identification model | |
| Guan et al. | Research on the method of remaining useful life prediction of lithium-ion battery based on LSTM | |
| CN114548762A (en) | Real-time power system cascading failure risk assessment method and system based on space-time diagram neural network | |
| CN114254828A (en) | Power load prediction method based on hybrid convolution feature extractor and GRU |
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 |