CN111611747B - Online state estimation method and device for hybrid energy storage system - Google Patents

Abstract

The invention belongs to the technical field of clean energy storage, and discloses an online state estimation method for a hybrid energy storage system, which comprises the steps of obtaining the instant voltage and the instant current of a storage battery and a super capacitor which are processed by a sampling circuit; when the increase value of the instant current is detected to exceed a preset threshold value, performing online capacity estimation on the instant voltages of the storage battery and the super capacitor; performing feature extraction on the instant voltage, and estimating the residual electric quantity of the storage battery and the super capacitor through a trained fuzzy brain emotion learning model; sending output data of the fuzzy brain emotion learning model; the device is also disclosed, the voltage and the current of the hybrid energy storage system after the load is impacted are utilized, and a neural network model designed by combining a fuzzy inference system and a brain emotion learning model is used as a state classifier, so that the state of the hybrid energy storage system is quickly, accurately and timely estimated on line, the device is suitable for being applied to a micro-grid and an uninterruptible power system, and the reliability of a power system is improved.

Description

An online state estimation method and device for a hybrid energy storage system

technical field

The invention belongs to the technical field of clean energy energy storage, and in particular relates to an online state estimation method and device for a hybrid energy storage system.

Background technique

With the wide application of clean energy power generation technologies represented by solar energy and wind energy, and at the same time to overcome the volatility of distributed generation and the impact of external shock loads on the safety and reliability of microgrids, energy storage technology is an important supporting technology. received extensive attention and participation. The hybrid energy storage system can effectively exert the characteristics of the battery and the super capacitor, so that the power supply system can better supply power to the shock load, and reduce the internal loss of the battery itself, thereby extending its operating life. Among them, supercapacitors are emerging energy storage components, and the current research mainly focuses on their internal parameter analysis and design and manufacture.

However, in combination with traditional energy storage detection methods, most of them are carried out in the form of periodic offline detection, which consumes a lot of manpower and time and may increase the risk of grid operation. The on-line monitoring of such signals has been widely carried out, but its practical application is still difficult due to the complex electrochemical effects inside the energy storage element.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide an online state estimation method and device for a hybrid energy storage system, in order to solve the problem that the above-mentioned traditional energy storage detection methods are mostly carried out in the form of periodic offline detection, which consumes a lot of manpower and time. And it may increase the risk of grid operation; at the same time, in recent years, parameter modeling and analysis of energy storage modules, and online monitoring of them based on voltage, current and other signals have been widely carried out. However, due to the complex electrochemical effects inside the energy storage elements It is still difficult to make it practical.

A technical solution adopted by the present invention to solve the technical problem is to propose an online state estimation method for a hybrid energy storage system, comprising the following steps:

Obtain the instant voltage and instant current processed by the sampling circuit for the battery and supercapacitor;

When it is detected that the increase value of the instant current exceeds a preset threshold, perform online capacity estimation for the instant voltage of the battery and supercapacitor;

Perform feature extraction on the instantaneous voltage and estimate the remaining capacity of batteries and supercapacitors through the trained fuzzy brain emotion learning model;

Send the output data of the fuzzy brain emotional learning model.

Further preferably, the "performing feature extraction on the instantaneous voltage" specifically includes: extracting effective data features by using a preferred wavelet packet decomposition method, wherein the wavelet basis function family selects the Daubechies wavelet family; calculating the jth-order Daubechies basis function in the fourth layer. For the band energy entropy after wavelet packet decomposition, the order with the smallest band energy entropy is selected as the optimal wavelet basis function, and the signal energy corresponding to the frequency band of the fourth layer under the optimal wavelet basis function is calculated, and the energy is obtained by normalization. Eigenvector; the energy coefficient of the low-frequency part obtained by the decomposition of the wavelet packet and the high-frequency energy coefficient with a total contribution rate of more than 95% after the principal component analysis are used as the input feature data.

Further preferably, the "fuzzy brain emotion learning model" is a fuzzy brain emotion learning neural network comprising a feature input layer, a sensory cortex layer, a sensory weight layer, an emotion weight layer, thalamus, amygdala, prefrontal cortex and output layer. .

Further preferably, the fuzzy brain emotion learning neural network specifically includes: the first layer is the fault feature input layer, and the voltage state characteristics of the storage battery and the super capacitor are introduced into the neural network; the second layer is the sensory cortex layer, with a Gaussian function as the The excitation function is used to quantify the input feature vector; the third layer is the sensory weight layer, the emotional weight layer and the thalamus, which is the weight space, and the fuzzy output of each block represents part of the results of the fuzzy inference rules; the fourth layer is the almond The nuclear and prefrontal cortex layers calculate the algebraic sum of the sensory cortex outputs with sensory weights and emotional weights and normalize them; the fifth layer is the output layer, which subtracts the outputs of the amygdala and prefrontal cortex, and then outputs Status label, judge the status of the battery and super capacitor through the label.

Further preferably, "performing feature extraction on the instantaneous voltage, and estimating the remaining power of the battery and the supercapacitor through the trained fuzzy brain emotional learning model" specifically includes: dividing the feature data into training samples and test samples; using the training samples Train the fuzzy brain emotion learning neural network state estimator and adjust the parameters; when it is detected that the training error meets the expected error value, go to the next step, otherwise continue to update the parameters of the sensory cortex, sensory weight layer and emotion weight layer , and return to the previous step; judge whether the correct rate of the test sample and the obtained weight test set meets the requirements, and if so, end.

Another technical solution adopted by the present invention to solve the technical problem is to propose an online state estimation device for a hybrid energy storage system, including the following:

The sampling module is used to obtain the instant voltage and instant current processed by the sampling circuit for the battery and supercapacitor;

The detection module is used to perform online capacity estimation on the instantaneous voltage of the battery and the super capacitor when it is detected that the increase value of the instantaneous current exceeds a preset threshold;

A data processing module for performing feature extraction on the instantaneous voltage and estimating the remaining power of the battery and supercapacitor through the trained fuzzy brain emotional learning model;

The communication module is used to send the output data of the fuzzy brain emotional learning model to the maintenance personnel.

Further preferably, in the data processing module, "performing feature extraction on the instantaneous voltage" specifically includes: extracting effective data features by using the optimal wavelet packet decomposition method, wherein the wavelet basis function family selects the Daubechies wavelet family; calculating the jth order Daubechies basis function. In the band energy entropy after wavelet packet decomposition in the fourth layer, the order with the smallest band energy entropy is selected as the optimal wavelet basis function, and the signal energy corresponding to the frequency band of the fourth layer under the optimal wavelet basis function is calculated. The energy eigenvectors are obtained by transforming them into energy eigenvectors; the energy coefficients of the low-frequency part obtained by the wavelet packet decomposition and the high-frequency energy coefficients that account for more than 95% of the total contribution rate after principal component analysis are used as the input feature data.

Further preferably, the "fuzzy brain emotional learning model" in the data processing module is a fuzzy brain comprising a feature input layer, a sensory cortex layer, a sensory weight layer, an emotion weight layer, thalamus, amygdala, prefrontal cortex and output layer. Emotional Learning Neural Networks.

Further preferably, the fuzzy brain emotion learning neural network specifically includes: the first layer is the fault feature input layer, and the voltage state characteristics of the storage battery and the super capacitor are introduced into the neural network; the second layer is the sensory cortex layer, with a Gaussian function as the The excitation function is used to quantify the input feature vector; the third layer is the sensory weight layer, the emotional weight layer and the thalamus, which is the weight space, and the fuzzy output of each block represents part of the results of the fuzzy inference rules; the fourth layer is the almond The nuclear and prefrontal cortex layers calculate the algebraic sum of the sensory cortex outputs with sensory weights and emotional weights and normalize them; the fifth layer is the output layer, which subtracts the outputs of the amygdala and prefrontal cortex, and then outputs Status label, judge the status of the battery and super capacitor through the label.

Further preferably, the "performing feature extraction on the instantaneous voltage, and estimating the remaining power of the storage battery and the supercapacitor through the trained fuzzy brain emotional learning model" specifically includes: dividing the feature data into training samples and test samples; using The training samples are used to train the fuzzy brain emotion learning neural network state estimator and adjust the parameters; when it is detected that the training error meets the expected error value, go to the next step, otherwise continue to adjust the parameters of the sensory cortex, sensory weight layer and emotion weight layer Update, and return to the previous step; judge whether the correct rate of the test sample and the obtained weight test set meets the requirements, and if so, end.

Beneficial effects of the present invention:

1. The present invention does not require complex electrochemical analysis of batteries and supercapacitors, nor does it need to detect high-frequency harmonic signals injected into batteries and supercapacitors; it can directly use the voltage and If the current is corresponding, the health state of the hybrid energy storage system can be estimated quickly, in real time and accurately, which is suitable for the uninterruptible power supply system and the energy storage system of the substation, reducing the maintenance cost of the system and improving the reliability of the system;

2. Fuzzy brain emotional learning neural network combines fuzzy system theory with neural network with emotional intelligence, so that it has the characteristics of fast convergence speed and low computational complexity, which can meet the needs of engineering.

Description of drawings

FIG. 1 is a flowchart of an online state estimation method for a hybrid energy storage system according to an embodiment of the present invention;

2 is a schematic diagram of a hybrid energy storage system structure and an online state estimator in an online state estimation method for a hybrid energy storage system according to an embodiment of the present invention;

3 is a flowchart of a feature extraction algorithm in an online state estimation method for a hybrid energy storage system according to an embodiment of the present invention;

4 is a flowchart of offline training and online estimation of FBELNN in an online state estimation method for a hybrid energy storage system according to an embodiment of the present invention;

5 is a schematic diagram of a network structure of FBELNN in an online state estimation method for a hybrid energy storage system according to an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of an online state estimation device for a hybrid energy storage system according to an embodiment of the present invention.

Detailed ways

In order to more clearly illustrate the embodiments of the present invention and/or the technical solutions in the prior art, specific embodiments of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative efforts, and obtain other implementations. In addition, the term referring to the orientation only indicates the relative positional relationship between the components, not the absolute positional relationship.

Please refer to FIG. 1 , FIG. 2 , FIG. 3 , FIG. 4 and FIG. 5 , the online state estimation method for a hybrid energy storage system in this embodiment includes the following steps:

S1 obtains the instant voltage and instant current processed by the sampling circuit for the battery and supercapacitor.

As shown in Figure 2, super capacitors and batteries can be obtained from commercial power and other renewable energy sources through bidirectional DC circuits and bidirectional rectifier circuits. %-170%, the impact power is 150KW-200KW.

S2 When it is detected that the increase value of the instant current exceeds a preset threshold, online capacity estimation is performed on the instant voltage of the battery and the super capacitor.

Here, it is judged whether there is an impact load according to whether the instantaneous current increase value exceeds the preset threshold. If it exceeds, the online real-time estimation will be made according to the voltage data of the battery and super capacitor 1 second before and after the impact occurs.

S3 performs feature extraction on the instantaneous voltage and estimates the remaining capacity of batteries and supercapacitors through a trained fuzzy brain affective learning model.

Among them, the "feature extraction for instantaneous voltage" can be shown in Figure 3, which specifically includes: extracting effective data features by using the optimal wavelet packet decomposition method, the wavelet packet is decomposed into four layers, and the Daubechies wavelet family is selected from the wavelet basis function family; calculation The band energy entropy of the jth-order Daubechies basis function after the decomposition of the fourth-layer wavelet packet, according to the information entropy theory, the basis function with the smallest band energy entropy is selected as the optimal wavelet basis function, and the fourth wavelet basis function under the optimal wavelet basis function is calculated. The signal energy corresponding to the frequency band of the layer is normalized to obtain the energy eigenvector; among them, the calculation formula of the band energy entropy of the nth order is:

为第四层小波包分解后频带i的能量概率密度；In the above formula,

is the energy probability density of frequency band i after the fourth-layer wavelet packet decomposition;

Then select the minimum value of the energy entropy of the frequency band, and the calculation formula is as follows:

where m is the optimal vanishing moment order, and the basis function of the optimal wavelet packet can be determined.

Under the basis function of the optimal wavelet packet, the energy of the i-th frequency band in the j-th layer is:

为小波包的每个节点重构系数。in,

Reconstruct the coefficients for each node of the wavelet packet.

Then the total energy of the signal can be calculated as follows:

Information energy and normalization can be calculated as follows:

The energy eigenvectors can be calculated as follows:

The energy eigenvectors of the low-frequency part obtained by wavelet packet decomposition and the high-frequency energy eigenvectors with a total contribution rate of more than 95% after principal component analysis are used as input feature data.

It should be noted that "fuzzy brain affective learning model" is a fuzzy brain affective learning neural network including feature input layer, sensory cortex layer, sensory weight layer, emotion weight layer, thalamus, amygdala, prefrontal cortex and output layer; this See Figure 5 for the neural network of fuzzy brain emotion learning at . The details include: the first layer is the fault feature input layer, which introduces the voltage state features of batteries and supercapacitors into the neural network; the second layer is the sensory cortex layer, which uses a Gaussian function. As an excitation function, the input feature vector is quantized; the third layer is the sensory weight layer, the emotional weight layer and the thalamus, which is the weight space, and the fuzzy output of each block represents part of the results of the fuzzy inference rules; the fourth layer is The amygdala and prefrontal cortex layers calculate the algebraic sum of the sensory cortex outputs with sensory weights and emotional weights and normalize them; the fifth layer is the output layer, which subtracts the outputs of the amygdala and prefrontal cortex layers, and then Output the status label, and judge the status of the battery and super capacitor through the label.

作为输入送入模糊大脑情感学习神经网络进行前向计算；Among them, the first layer is the fault feature input layer: the voltage state features of batteries and supercapacitors are introduced into the neural network. Specifically: the sample feature vector

As input, it is sent to the fuzzy brain emotional learning neural network for forward calculation;

The second layer is the sensory cortex layer: the Gaussian function is used as the excitation function to quantify the input feature vector and send it to the weight space; due to the need to improve the generalization ability and running speed of the model, each block blurs the input features excitation. Here, the Gaussian function is used as the activation function, which can be expressed as

和

分别表示为高斯函数的均值和方差。In the above formula, it is expressed as the output of the i-th input feature on the j-th block,

and

are expressed as the mean and variance of the Gaussian function, respectively.

为

所对应的感觉权重，如下式所示：The third layer is the sensory weight layer, the emotional weight layer and the thalamus: that is, the weight space, the fuzzy output of each block represents part of the results of the fuzzy inference rules, and the multiplication value of the sensory cortex output value and the sensory weight will be sent to the amygdala,

for

The corresponding sensory weight is as follows:

为

所对应的感觉权重，如下式所示：The multiplication of the sensory cortex output value and the emotional weight will be sent to the prefrontal cortex,

for

The corresponding sensory weight is as follows:

The thalamus acts as an inhibitory effect on the amygdala, and the Vth parameter is manually set:

The fourth layer is the amygdala and prefrontal cortex: the sensory cortex output with sensory weights and affective weights is calculated algebraically and normalized. The corresponding output of the amygdala is:

The corresponding output of the prefrontal cortex layer is:

The fifth layer is the output layer: the outputs of the amygdala and the prefrontal cortex are subtracted, and then the output is compressed between (0, 1) by the sigmoid function, and the status of the battery and supercapacitor is judged by the label. Among them, the output Ui is:

和情感权重

是根据神经生理学原理，杏仁核对特定的情绪暗示做出反应，并且调节前额皮质层的输出使杏仁核与情绪提示的差异最小化。所以，可以得到

的更新方式如下：In this implementation example, for the sensory weight

and emotional weight

According to neurophysiological principles, the amygdala responds to specific emotional cues and modulates the output of the prefrontal cortex to minimize the difference between the amygdala and emotional cues. So, one can get

is updated as follows:

是

的学习率，并且奖励信号

是

由神经网络的输出误差信号和输出信号组成。

的表达式如下所示：in,

Yes

, and the reward signal

Yes

It consists of the output error signal and output signal of the neural network.

The expression looks like this:

和

由人工进行调整，

为输出误差，

为期望输出；in,

and

adjusted manually,

is the output error,

is the expected output;

的更新方式如下：In the prefrontal cortex, in order to inhibit or modulate the signal of the amygdala to speed up the learning process,

is updated as follows:

和

参数，通过链式求导法则可得：In general, the sensory cortex of the traditional brain emotion learning neural network does not have any learning process, but the update of the mean and variance of the Gaussian function needs to be considered in FBELNN, which makes the sensory cortex have learning rules. Here we use gradient descent to adjust

and

parameters, which can be obtained by the chain derivation rule:

in,

In particular, "S3 performs feature extraction on the instantaneous voltage, and estimates the remaining power of the battery and supercapacitor through the trained fuzzy brain emotional learning model" can refer to Figure 4, which includes: S31 divides the feature data into training samples and Test sample; S32 use the training sample to train the fuzzy brain emotion learning neural network state estimator and adjust the parameters; S33 When it is detected that the training error meets the expected error value, go to step S34, otherwise continue to the sensory cortex layer, sensory weight layer and The parameters of the emotional weight layer are updated, and return to step S32; S34 judges whether the correct rate of the test sample and the weight test set obtained in step S33 meets the requirements, if so, end the current step, otherwise return to step S32.

The optimal fuzzy brain emotional learning neural network after offline training and testing can be written into the DSP with the feature processing algorithm and fuzzy brain emotional learning model, that is, the status of the battery and super capacitor can be analyzed in real time.

S4 sends the output data of the fuzzy brain emotional learning model; here, the output results of the fuzzy brain emotional learning model can be transmitted to the maintenance personnel through the communication module, which is convenient for the maintenance personnel to judge the situation and subsequent maintenance processing.

This embodiment selects the shock load response signal of the battery and the super capacitor as the diagnosis basis, and the feature extraction method used is more intelligent, eliminating the need to select the order of the wavelet function and the number of obtained energy features through artificial experience. As a neural network with emotion module, it has the characteristics of fast convergence speed and low computational complexity, which can meet the needs of engineering.

Referring to FIG. 6, the present invention also proposes an online state estimation device for a hybrid energy storage system, including:

The sampling module is used to obtain the instant voltage and instant current processed by the sampling circuit for the battery and supercapacitor;

The detection module is used to perform online estimation of the instantaneous voltage of the battery and the super capacitor when it is detected that the instantaneous current exceeds the preset threshold;

A data processing module for performing feature extraction on the instantaneous voltage and estimating the remaining power of the battery and supercapacitor through the trained fuzzy brain emotional learning model;

The communication module is used to send the output data of the fuzzy brain emotional learning model to the maintenance personnel.

In particular, in the data processing module, "performing feature extraction on the instantaneous voltage" specifically includes: extracting effective data features by using the optimal wavelet packet decomposition method, wherein the wavelet basis function family selects the Daubechies wavelet family; calculating the jth-order Daubechies basis function in the fourth The frequency band energy entropy after the decomposition of the layer wavelet packet, select the basis function with the smallest frequency band energy entropy as the optimal wavelet basis function, and calculate the signal energy corresponding to the frequency band of the fourth layer under the optimal wavelet basis function. Energy feature vector; the energy coefficient of the low-frequency part obtained by the wavelet packet decomposition and the high-frequency energy coefficient with a total contribution rate of more than 95% after principal component analysis are used as input feature data.

In particular, the "fuzzy brain affective learning model" in the data processing module is a fuzzy brain affective learning neural network including feature input layer, sensory cortex layer, sensory weight layer, affective weight layer, thalamus, amygdala, prefrontal cortex and output layer network.

In particular, the fuzzy brain emotion learning neural network specifically includes: the first layer is the fault feature input layer, which introduces the voltage state characteristics of batteries and supercapacitors into the neural network; the second layer is the sensory cortex layer, which uses the Gaussian function as the excitation function, The input feature vector is quantized; the third layer is the sensory weight layer, the emotional weight layer and the thalamus, which is the weight space, and the fuzzy output of each block represents part of the results of the fuzzy inference rules; the fourth layer is the amygdala and forehead The cortical layer calculates the algebraic sum of the sensory cortex outputs with sensory weights and emotional weights and normalizes them; the fifth layer is the output layer, which subtracts the outputs of the amygdala and prefrontal cortex, and then outputs the state label, Judging the status of batteries and supercapacitors through labels.

In particular, "Performing feature extraction on instantaneous voltage and estimating the remaining power of batteries and supercapacitors through the trained fuzzy brain emotional learning model" specifically includes: dividing the feature data into training samples and test samples; using the training samples to The fuzzy brain emotion learning neural network state estimator is trained and the parameters are adjusted; when it is detected that the training error meets the expected error value, go to the next step, otherwise continue to update the parameters of the sensory cortex layer, the sensory weight layer and the emotion weight layer, And return to the previous step; judge whether the correct rate of the test sample and the obtained weight test set meets the requirements, and if so, end.

The above content is a further detailed description of the present invention in combination with specific preferred embodiments, and it cannot be considered that the specific implementation of the present invention is limited to these descriptions. For those of ordinary skill in the technical field of the present invention, without departing from the concept of the present invention, some simple deductions or substitutions can be made, which should be regarded as belonging to the protection scope of the present invention.

Claims (4)

1. an online state estimation method for a hybrid energy storage system, characterized in that, comprising the following steps: S1 obtains the instant voltage and instant current processed by the sampling circuit for the battery and supercapacitor; S2, when it is detected that the increase value of the instant current exceeds the preset threshold, perform online capacity estimation through the instant voltage of the battery and the supercapacitor; S3 performs feature extraction on the instantaneous voltage, and estimates the remaining capacity of the battery and supercapacitor through the trained fuzzy brain emotional learning model; S4 sends the output data of the fuzzy brain emotional learning model; Wherein, the "fuzzy brain emotional learning model" in the step S3 is a fuzzy brain emotional learning neural model including a feature input layer, a sensory cortex layer, a sensory weight layer, an emotional weight layer, thalamus, amygdala, prefrontal cortex and output layer. network; The fuzzy brain emotion learning neural network specifically includes: the first layer is the fault feature input layer, which introduces the voltage state characteristics of the battery and the super capacitor into the neural network; the second layer is the sensory cortex layer, which uses the Gaussian function as the excitation function , quantify the input feature vector; the third layer is the sensory weight layer, the emotional weight layer and the thalamus, which is the weight space, and the fuzzy output of each block represents part of the results of the fuzzy inference rules; the fourth layer is the amygdala and the thalamus. The prefrontal cortex layer calculates the algebraic sum of the sensory cortex output with sensory weights and emotional weights and normalizes it; the fifth layer is the output layer, which subtracts the outputs of the amygdala and prefrontal cortex, and then outputs the state Label, judge the status of the battery and super capacitor through the label; Among them, the "S3 performs feature extraction on the instantaneous voltage, and estimates the remaining power of the battery and the supercapacitor through the fuzzy brain emotional learning model after training" specifically includes: S31 The feature data is divided into training samples and test samples; S32 Use the training samples to train the fuzzy brain emotion learning neural network state estimator and adjust the parameters; S33, when it is detected that the training error conforms to the expected error value, go to step S34, otherwise the sensory cortex layer, the sensory weight layer and the emotion weight layer continue to be evaluated. Update the parameters, and return to step S32; S34 judges whether the correct rate of the test sample and the weight test set obtained in step S33 meets the requirements, if so, end the current step, otherwise return to step S32. 2. An online state estimation method for a hybrid energy storage system according to claim 1, characterized in that, in the step S3, "performing feature extraction on the instantaneous voltage" specifically comprises: using a wavelet packet decomposition method to extract effective The data characteristics of , where the wavelet basis function family selects the Daubechies wavelet family; calculate the band energy entropy of the jth-order Daubechies basis function after the fourth layer of wavelet packet decomposition, and select the order with the smallest band energy entropy as the optimal wavelet basis function, and Calculate the signal energy corresponding to the frequency band of the fourth layer under the optimal wavelet basis function, and obtain the energy feature vector through normalization; More than 95% of the high-frequency energy coefficients are used as input feature data. 3. An online state estimation device for a hybrid energy storage system, comprising: The sampling module is used to obtain the instant voltage and instant current processed by the sampling circuit for the battery and supercapacitor; The detection module is used to perform online estimation of the instantaneous voltage of the battery and the super capacitor when it is detected that the increase value of the instantaneous current exceeds a preset threshold; A data processing module for performing feature extraction on the instantaneous voltage and estimating the remaining capacity of the battery and supercapacitor through the trained fuzzy brain emotion learning neural network; The communication module is used to send the output data of the fuzzy brain emotional learning neural network to the maintenance personnel; Wherein, the "fuzzy brain emotional learning model" in the data processing module is a fuzzy brain emotional learning model including a feature input layer, a sensory cortex layer, a sensory weight layer, an emotional weight layer, thalamus, amygdala, prefrontal cortex and output layer Neural Networks; The fuzzy brain emotion learning neural network specifically includes: the first layer is the fault feature input layer, which introduces the voltage state characteristics of the battery and the super capacitor into the neural network; the second layer is the sensory cortex layer, which uses the Gaussian function as the excitation function , quantify the input feature vector; the third layer is the sensory weight layer, the emotional weight layer and the thalamus, which is the weight space, and the fuzzy output of each block represents part of the results of the fuzzy inference rules; the fourth layer is the amygdala and the thalamus. The prefrontal cortex layer calculates the algebraic sum of the sensory cortex output with sensory weights and emotional weights and normalizes it; the fifth layer is the output layer, which subtracts the outputs of the amygdala and prefrontal cortex, and then outputs the state Labels to judge the health status of batteries and supercapacitors through labels; Among them, the "perform feature extraction on the instantaneous voltage, and use the trained fuzzy brain emotional learning model to estimate the remaining power of batteries and supercapacitors" specifically includes: dividing the feature data into training samples and test samples; using training samples Train the fuzzy brain emotion learning neural network state estimator and adjust the parameters; when it is detected that the training error meets the expected error value, go to the next step, otherwise continue to update the parameters of the sensory cortex, sensory weight layer and emotion weight layer , and return to the previous step; judge whether the correct rate of the test sample and the obtained weight test set meets the requirements, and if so, end. 4 . The online state estimation device for a hybrid energy storage system according to claim 3 , wherein, in the data processing module, “performing feature extraction on the instantaneous voltage” specifically comprises: extracting by using a wavelet packet decomposition method. 5 . Effective data features, where the wavelet basis function family selects the Daubechies wavelet family; calculate the band energy entropy of the jth-order Daubechies basis function after the fourth layer of wavelet packet decomposition, and select the order with the smallest band energy entropy as the optimal wavelet basis function, And calculate the signal energy corresponding to the frequency band of the fourth layer under the optimal wavelet basis function, and obtain the energy eigenvector through normalization; The high-frequency energy coefficients above 95% are used as input feature data.

Images