CN111208436A

CN111208436A - Energy storage battery overcharge diagnosis method based on IMF energy moment and genetic algorithm optimization SVM

Info

Publication number: CN111208436A
Application number: CN202010107150.8A
Authority: CN
Inventors: 吴元熙; 劳文洁; 郑茜匀; 王彬; 马宏忠; 韦玉蕾
Original assignee: Hohai University HHU
Current assignee: Hohai University HHU
Priority date: 2020-02-21
Filing date: 2020-02-21
Publication date: 2020-05-29

Abstract

The invention discloses an energy storage battery overcharge diagnosis method based on an IMF energy moment and a genetic algorithm optimization SVM, which comprises the steps of firstly, acquiring shell surface vibration signals of an energy storage battery in a normal charging state and an overcharge state by using a sensor; then, denoising the signal by utilizing wavelet denoising; decomposing the noise-canceling signal into a plurality of eigenmode functions through empirical mode decomposition; then, energy moments of all the eigenmode functions are obtained by utilizing the eigenmode functions and serve as characteristic indexes, and a data set is formed after normalization processing; training a genetic algorithm by using a data set to optimize a support vector machine model and determine optimal parameters; and finally extracting the corresponding characteristic vector of the signal to be diagnosed and diagnosing by using the model. Through example inspection, the accuracy of the invention for inspecting the overcharge fault is up to 100%, and a reliable means is provided for the safe operation of the energy storage power station.

Description

Energy storage battery overcharge diagnosis method based on IMF energy moment and genetic algorithm optimization SVM

Technical Field

The invention relates to an energy storage battery overcharge diagnosis method based on an IMF energy moment and a genetic algorithm optimization SVM, and belongs to the field of energy storage battery fault monitoring.

Background

Under the construction background of a smart power grid, an energy storage battery becomes a key technology for ensuring the safe, stable and efficient operation of the power grid, and plays an important role in peak clipping and valley filling, stabilizing fluctuation and improving power supply quality. The current lithium battery has been widely used in a power grid energy storage system, but the energy storage battery has a safety problem in the operation process, and particularly in the field of large-scale energy storage, a large number of monomers are required to form a battery pack and a battery pack through series-parallel connection, so that the complexity of the system is greatly increased, and the potential safety hazard of operation is increased. Since overcharge is an important cause of temperature rise in the battery cell to cause explosion, it is necessary to diagnose and identify the overcharge state of the battery.

The current commonly used fault diagnosis method mainly comprises the steps of establishing a battery model to diagnose the state of a battery, expressing the chemical reaction in the battery by establishing the battery model and searching the mathematical relationship of various parameters of the battery, and taking whether the various parameters exceed a preset threshold value as the basis for judging the fault. Another common method is to collect voltage, current and temperature data of the battery, and use a neural network for modeling, but in order to make the neural network have good fault diagnosis capability, a large amount of typical data needs to be acquired to train the neural network, and the requirements on sample data density and quality are high. Meanwhile, the voltage, the current, the temperature and the internal fault of the battery do not have a strict corresponding relationship, so that the characteristic quantity which can reflect the battery fault more directly needs to be researched.

Disclosure of Invention

The invention aims to overcome the defects in the prior art, provides an energy storage battery overcharge diagnosis method based on an IMF energy moment and a genetic algorithm optimization SVM, extracts vibration signal data for analysis, and can accurately identify the overcharge fault of the energy storage battery.

In order to solve the technical problems, the invention is realized by the following scheme:

adhering an acceleration sensor to the center of the surface of an energy storage battery monomer, realizing two different states of normal charging and overcharging of the battery by setting different charging cut-off voltages, and collecting vibration signals to form original signals;

step two, wavelet denoising processing is carried out on the original signal, the denoising method can effectively denoise noise and can keep most useful signals, and the specific implementation method is as follows: acquiring a default threshold value of a signal by utilizing a function ddencmp () of Matlab, and denoising an original signal by utilizing a function wdencmp (), wherein parameters are selected as follows: db3 wavelet, 2 layers of decomposition layers;

thirdly, performing empirical mode decomposition on the noise-reduced signals, respectively calculating energy moments of the Intrinsic Mode Function (IMF) components, and then performing normalization processing to obtain a data set consisting of characteristic vectors, wherein the method specifically comprises the following steps:

1. the signal is decomposed by EMD to obtain a plurality of eigenmode functions (IMF) z_i(t),i＝1,2,……；

2. For discrete signals, each IMF energy moment is calculated using the following equation:

in the formula, delta t is a sampling period, n is the total number of sampling points, and j is a sampling point;

3. and (3) carrying out normalization calculation according to the following formula to obtain an energy moment feature vector:

step four, training a support vector machine model by using the data set obtained in the step three, wherein a kernel function selects an RBF kernel, and the optimal solution of kernel function parameters and penalty factors is solved by using a genetic algorithm;

and fifthly, acquiring a characteristic vector of the signal in the state to be diagnosed, diagnosing by using the established model, and judging whether the energy storage battery is in an overcharged state.

Compared with the prior art, the invention has the beneficial effects that:

the vibration quantity is a quantity which effectively reflects the state of the battery, and the actually measured vibration signal on the surface of the battery is a non-stationary signal;

the IMF energy moment is selected as a characteristic index, the IMF energy moment is obtained by IMF components based on time axis integration, the distribution characteristics of the energy of the IMF components on the time axis are reflected, and the characteristics of signals can be effectively reflected;

the genetic algorithm optimization support vector machine model adopted by the invention has advantages in solving the problem of small sample classification, and meanwhile, the genetic algorithm can be used for obtaining the optimal parameters of the model, so that the recognition rate can be effectively improved;

the invention can reach 100% of the over-charge fault diagnosis accuracy of the energy storage battery, is a powerful supplement to the fault diagnosis means of the energy storage battery, and greatly improves the reliability of the safe operation of the energy storage battery.

Drawings

FIG. 1 is an overall flow diagram of the present invention;

FIG. 2 is a test chart of 60 sets of test samples classified by the genetic algorithm-optimized support vector machine model obtained under 60 sets of training samples according to the example of the present invention.

Detailed Description

The following description of the embodiments of the present invention is provided in connection with the accompanying drawings and examples.

Considering that most energy storage power stations adopt lithium iron phosphate batteries, selecting a lithium iron phosphate battery with the capacity of 50AH, the charge cut-off voltage of 3.65V and the discharge cut-off voltage of 2V for an experiment, respectively setting the charge cut-off voltage to be 3.65V and 5V to meet the requirement that the battery is in two states of normal charge and overcharge, placing an acceleration sensor at the center of the surface of a battery monomer, respectively collecting vibration signals in the two states, wherein the sampling frequency is 20kHz, the sampling time of each group is 1s, and obtaining 120 groups of sample data in total, wherein 72 groups of normal state samples and 48 groups of overcharge state samples.

And step two, denoising the original signal by using wavelet denoising, preferably selecting a db3 wavelet, and decomposing the layer number to be 2.

And thirdly, decomposing the processed signal by using an EMD method, extracting energy moments of the decomposed IMF components, normalizing the energy moments to obtain a characteristic vector data set, defining a normal state to be represented by 1, defining an overcharging state to be represented by 2, and finally obtaining a 120 x 10 matrix, wherein each row is a sample, the first 9 columns are attribute values of the characteristic vector, and the 10 th column is a sample label. For example, only the first three components of the feature vectors of the first two sets in the normal state and the overcharged state are listed, as shown in Table 1:

TABLE 1

Group/component	1	2	3
				1	0.0040	0.0135	0.0236
2	0.0048	0.0137	0.0176
				……	……	……	……
73	0.0017	0.0286	0.1931
				74	0.0016	0.0274	0.1382

And step four, randomly selecting 60 groups of samples as training samples, training a support vector machine model, selecting a model kernel function as an RBF kernel, determining a penalty factor and kernel function parameters of the RBF kernel by using a genetic algorithm, and finally obtaining the optimal combination (1.00012, 39.1086) of the parameters in the example, so that the model is established.

And step five, in order to verify the effect of the model, selecting the remaining 60 groups of samples as test samples, inputting the models for inspection, and inspecting the models, wherein the classified test chart of the model inspection is shown as the attached figure 2, and the identification rate in the example reaches 100%.

The undescribed parts of the present invention are the same as or implemented using prior art.

Claims

1. An energy storage battery overcharge diagnosis method based on an IMF energy moment and a genetic algorithm optimization SVM is characterized by comprising the following steps:

s1: acquiring shell surface vibration signals of an energy storage battery in normal charging and overcharging states by using a sensor;

s2: performing wavelet denoising processing on the collected signals;

s3: performing empirical mode decomposition on the processed signals, extracting energy moments of all eigenmode functions as characteristic indexes, and performing normalization processing to form a data set;

s4: training a genetic algorithm optimization support vector machine model by using a data set, and determining optimal parameters of the model;

s5: and extracting the characteristic vector of the signal to be diagnosed, and diagnosing whether the energy storage battery is overcharged by using the model.

2. The energy storage battery overcharge diagnosis method based on IMF energy moment and genetic algorithm optimization SVM of claim 1, characterized by: in step S1, the vibration signal is collected using an acceleration sensor that is adhered to the center of the surface of the stationary energy storage battery.

3. The energy storage battery overcharge diagnosis method based on IMF energy moment and genetic algorithm optimization SVM of claim 1, characterized by: in step S1, the energy storage battery is in two different states of normal charging and overcharging by changing the charging cut-off voltage of the energy storage battery.

4. The energy storage battery overcharge diagnosis method based on IMF energy moment and genetic algorithm optimization SVM of claim 1, characterized by: in step S2, a function ddencmp () of Matlab is used to obtain a default threshold of the signal, wdencmp () is then used to denoise the original signal, db3 wavelet is selected, and the number of decomposition layers is 2.

5. The energy storage battery overcharge diagnosis method based on IMF energy moment and genetic algorithm optimization SVM of claim 1, characterized by: in step S3, the energy moment normalization process calculation step for each eigenmode function is as follows:

s3.1: the signal is decomposed by EMD to obtain a plurality of eigenmode functions (IMF) z_i(t),i＝1,2,……；

S3.2: for discrete signals, each IMF energy moment is calculated using the following equation:

s3.3: and (3) carrying out normalization calculation according to the following formula to obtain an energy moment feature vector:

6. the energy storage battery overcharge diagnosis method based on IMF energy moment and genetic algorithm optimization SVM of claim 1, characterized by: in step S4, the main process of searching the optimal parameters for support vector machine energy storage battery overcharge identification by using the genetic algorithm is as follows:

s4.1: and inputting the normalized energy moment feature vector data set into a support vector machine model as a training sample.