A lithium battery energy storage system monitoring method and system based on noise reduction autoencoding

Abstract

The present invention provides a method and system for monitoring a lithium battery energy storage system based on denoising autoencoder, the method comprising: calculating the mutual information value between the lithium battery terminal voltage and each parameter in the measurement parameter group and / or physical parameter group; constructing a data sample based on the mutual information value; constructing a stack denoising autoencoder network based on the data sample; establishing a regression prediction model of the lithium battery terminal voltage according to the stack denoising autoencoder network based on a regression method; calculating the predicted value of the lithium battery terminal voltage in the battery energy storage system according to the stack denoising autoencoder network and the regression prediction model; judging whether to alarm according to the predicted value of the lithium battery terminal voltage in the battery energy storage system. The present invention uses the self-supervised learning process of the autoencoder to construct a stack denoising autoencoder network, explore the deep connection between characteristic variables, and perform dimensionality reduction processing to remove noise and redundant characteristic variables.

Description

Technical Field

[0001] The present invention relates to the technical field of battery energy storage system monitoring, and in particular to a lithium battery energy storage system monitoring method and system based on noise reduction autoencoding. Background Art

[0002] Battery energy storage system is a kind of electrochemical energy storage method. The technical application of lithium batteries in energy storage mainly focuses on the fields of grid energy storage, base station backup power supply, and electric vehicle photovoltaic charging stations.

[0003] The working principle of lithium-ion battery energy storage is: when the battery is charged, lithium ions are generated on the positive electrode of the battery, and the generated lithium ions will move to the negative electrode through the electrolyte. The carbon in the negative electrode is a layered structure with many micropores. The lithium ions that reach the negative electrode are embedded in the micropores of the carbon layer. The more lithium ions are embedded, the higher the charging capacity. Similarly, when the battery is discharged, the lithium ions embedded in the carbon layer of the negative electrode are released and move back to the positive electrode. The more lithium ions that return to the positive electrode, the higher the discharge capacity.

[0004] Lithium batteries have many advantages in the field of energy storage, such as long cycle life, relatively high energy density, low self-discharge rate, and green environmental protection. In recent years, with the increase in demand for electricity, the scale of the power system has been expanding, and the complexity of the power grid has also increased. At the same time, the country is now vigorously developing smart grids, renewable energy, and distributed energy, which has increased the demand for energy storage technology. In addition, the country strongly supports the development of energy storage and proposes to accelerate the research and development and application of technologies such as large-scale energy storage. Lithium battery energy storage systems are also playing an increasingly important role. However, due to overcharging or over-discharging, the battery will be damaged. The poor health of lithium batteries and uneven charging of single lithium batteries will cause safety problems. The monitoring system can detect these problems and apply remote control to improve the safety and reliability of the energy storage system. However, the existing technical solutions have not disclosed how to monitor lithium batteries in lithium battery energy storage systems. Therefore, there is an urgent need in this field to provide a method for monitoring lithium batteries in lithium battery energy storage systems. Summary of the invention

[0005] Based on this, the purpose of the present invention is to provide a lithium battery energy storage system monitoring method and system based on noise reduction autoencoding to achieve real-time monitoring of the lithium battery energy storage system.

[0006] To achieve the above object, the present invention provides a lithium battery energy storage system monitoring method based on noise reduction autoencoding, the method comprising:

[0007] Step S1: Calculate the mutual information value between the lithium battery terminal voltage and each parameter in the measurement parameter group and / or the physical parameter group;

[0008] Step S2: constructing a data sample based on the mutual information value;

[0009] Step S3: constructing a stacked denoising autoencoder network based on the data sample;

[0010] Step S4: Based on a regression method, a regression prediction model of the lithium battery terminal voltage is established according to the stacked denoising autoencoder network;

[0011] Step S5: calculating a predicted value of a lithium battery terminal voltage in a battery energy storage system according to the stacked denoising autoencoder network and the regression prediction model;

[0012] Step S6: Determine whether to issue an alarm based on the predicted value of the lithium battery terminal voltage in the battery energy storage system.

[0013] Optionally, constructing a stacked denoising autoencoder network based on data samples specifically includes:

[0014] Step S31: training a first denoising autoencoder based on the data sample;

[0015] Step S32: using the output of the hidden layer of the trained first denoising autoencoder as the input of the input layer of the second denoising autoencoder to train the second denoising autoencoder;

[0016] Step S33: stacking the trained first denoising autoencoder and the trained second denoising autoencoder to form a stacked denoising autoencoder network.

[0017] Optionally, the training of the first denoising autoencoder based on the data sample specifically includes:

[0018] Step S311: giving the initialization weights, number of neurons and activation function of the first denoising autoencoder input layer;

[0019] Step S312: selecting a plurality of characteristic variables from the data sample;

[0020] Step S313: training the first denoising autoencoder based on the initialization weights of the first denoising autoencoder input layer and the plurality of feature variables;

[0021] Step S314: using the error function and BP algorithm to reversely calculate two weight changes of the first denoising autoencoder;

[0022] Step S315: using the gradient descent method to calculate the weight required for the next training according to the two weight changes;

[0023] Step S316: Determine whether the number of iterations is greater than or equal to the maximum number of iterations; if the number of iterations is greater than or equal to the maximum number of iterations, execute "Step S32"; if the number of iterations is less than the maximum number of iterations, use the weights required for the next training as the initialization weights of the first denoising autoencoder input layer, and return to "Step S312".

[0024] Optionally, the regression method is based on the stacked denoising autoencoder network to establish a regression prediction model for the terminal voltage of the lithium battery, specifically including:

[0025] Step S41: Based on a regression method, the data sample is input into the stacked denoising autoencoder network to obtain a vector of an output layer of the stacked denoising autoencoder network;

[0026] Step S42: extracting the vector of the output layer of the stacked denoising autoencoder network through matrix transformation to obtain parameters that are strongly correlated with the terminal voltage of the lithium battery; the mutual information value is strongly correlated in the range of 0.6-1;

[0027] Step S43: establishing a regression prediction model for the lithium battery terminal voltage according to the lithium battery terminal voltage and parameters strongly correlated with the lithium battery terminal voltage.

[0028] Optionally, judging whether to issue an alarm according to the predicted value of the lithium battery terminal voltage in the battery energy storage system specifically includes:

[0029] Step S61: Compare the predicted value of the lithium battery terminal voltage with the lithium battery terminal voltage to obtain an error;

[0030] Step S62: Determine whether the error is within a set range; if the error is within the set range, it indicates that the lithium battery in the battery energy storage system is working normally; if the error is not within the set range, the monitoring system issues an alarm.

[0031] The present invention also provides a lithium battery energy storage system monitoring system based on noise reduction autoencoding, the system comprising:

[0032] A mutual information value calculation module, used to calculate the mutual information value between the lithium battery terminal voltage and each parameter in the measurement parameter group and / or the physical parameter group;

[0033] A data sample construction module, constructing a data sample based on the mutual information value;

[0034] The stack denoising autoencoder network building module builds a stack denoising autoencoder network based on data samples;

[0035] A regression prediction model determination module, based on a regression method, establishes a regression prediction model of the lithium battery terminal voltage according to the stack denoising autoencoder network;

[0036] A lithium battery terminal voltage prediction value determination module, used to calculate the lithium battery terminal voltage prediction value in the battery energy storage system according to the stack denoising autoencoder network and the regression prediction model;

[0037] The alarm determination module is used to determine whether to give an alarm based on the predicted value of the lithium battery terminal voltage in the battery energy storage system.

[0038] Optionally, the stack denoising autoencoder network construction module specifically includes:

[0039] A first training unit, training a first denoising autoencoder based on the data sample;

[0040] A second training unit is used to use the output of the hidden layer of the trained first denoising autoencoder as the input of the input layer of the second denoising autoencoder to train the second denoising autoencoder;

[0041] The stacking unit is used to stack the trained first denoising autoencoder and the trained second denoising autoencoder to form a stacked denoising autoencoder network.

[0042] Optionally, the first training unit specifically includes:

[0043] A first given subunit, used to give the initialization weights, number of neurons and activation function of the first denoising autoencoder input layer;

[0044] A selection subunit, used for selecting a plurality of characteristic variables from the data sample;

[0045] A first training subunit trains a first denoising autoencoder based on the initialization weights of the first denoising autoencoder input layer and the plurality of feature variables;

[0046] A first weight change determination subunit is used to reversely calculate two weight changes of the first denoising autoencoder using an error function and a BP algorithm;

[0047] A first weight determination subunit is used to calculate the weight required for the next training according to the two weight changes by using a gradient descent method;

[0048] The first judgment subunit is used to judge whether the number of iterations is greater than or equal to the maximum number of iterations; if the number of iterations is greater than or equal to the maximum number of iterations, the "second training unit" is executed; if the number of iterations is less than the maximum number of iterations, the weights required for the next training are used as the initialization weights of the first denoising autoencoder input layer, and the "selection subunit" is returned.

[0049] Optionally, the regression prediction model determination module specifically includes:

[0050] A vector determination unit, based on a regression method, inputs the data sample into the stacked denoising autoencoder network to obtain a vector of an output layer of the stacked denoising autoencoder network;

[0051] An extraction unit, used to extract the vector of the output layer of the stack denoising autoencoder network through matrix transformation to obtain a parameter strongly correlated with the terminal voltage of the lithium battery; the mutual information value is strongly correlated in the range of 0.6-1;

[0052] The regression prediction model determination unit is used to establish a regression prediction model of the lithium battery terminal voltage according to the lithium battery terminal voltage and a parameter strongly correlated with the lithium battery terminal voltage.

[0053] Optionally, the alarm determination module specifically includes:

[0054] A comparison unit, used for comparing the predicted value of the lithium battery terminal voltage with the lithium battery terminal voltage to obtain an error;

[0055] The judgment unit is used to judge whether the error is within a set range; if the error is within the set range, it means that the lithium battery in the battery energy storage system is working normally; if the error is not within the set range, the monitoring system will issue an alarm.

[0056] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0057] The present invention provides a method and system for monitoring a lithium battery energy storage system based on denoising autoencoder, the method comprising: calculating the mutual information value between the lithium battery terminal voltage and each parameter in the measurement parameter group and / or physical parameter group; constructing a data sample based on the mutual information value; constructing a stack denoising autoencoder network based on the data sample; establishing a regression prediction model of the lithium battery terminal voltage according to the stack denoising autoencoder network based on a regression method; calculating the predicted value of the lithium battery terminal voltage in the battery energy storage system according to the stack denoising autoencoder network and the regression prediction model; judging whether to alarm according to the predicted value of the lithium battery terminal voltage in the battery energy storage system. The present invention uses the self-supervised learning process of the autoencoder to construct a stack denoising autoencoder network, explore the deep connection between characteristic variables, and perform dimensionality reduction processing to remove noise and redundant characteristic variables. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0059] Figure 1This is a flow chart of a method for monitoring a lithium battery energy storage system based on noise reduction autoencoding according to an embodiment of the present invention;

[0060] Figure 2 This is a structural diagram of a lithium battery energy storage system monitoring system based on noise reduction autoencoding according to an embodiment of the present invention;

[0061] Figure 3 This is a data diagram of the mutual information value of the lithium battery energy storage system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] The purpose of the present invention is to provide a lithium battery energy storage system monitoring method and system based on noise reduction autoencoding to achieve real-time monitoring of the lithium battery energy storage system.

[0064] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0065] like Figure 1 As shown, the present invention provides a lithium battery energy storage system monitoring method based on noise reduction autoencoding, the method comprising:

[0066] Step S1: Calculate the mutual information value between the lithium battery terminal voltage and each parameter in the measurement parameter group and / or the physical parameter group.

[0067] Step S2: constructing a data sample based on the mutual information value.

[0068] Step S3: construct a stacked denoising autoencoder network based on the data samples.

[0069] Step S4: Based on the regression method, a regression prediction model of the lithium battery terminal voltage is established according to the stacked denoising autoencoder network.

[0070] Step S5: Calculate the predicted value of the lithium battery terminal voltage in the battery energy storage system according to the stacked denoising autoencoder network and the regression prediction model.

[0071] Step S6: Determine whether to issue an alarm based on the predicted value of the lithium battery terminal voltage in the battery energy storage system.

[0072] The following is a detailed discussion of each step:

[0073] Step S1: Calculating the mutual information value between the lithium battery terminal voltage and each parameter in the measurement parameter group and / or the physical parameter group, specifically including:

[0074] Step S11: Obtain the terminal voltage, measurement parameter group and physical parameter group of the lithium battery in the lithium battery energy storage system. The measurement parameter group includes charge and discharge efficiency, charge rate, discharge rate, state of charge and battery health status; the physical parameter group includes battery capacity, rated voltage, maximum discharge rate, discharge cut-off voltage, charge and discharge current, standard operating temperature, battery temperature, cycle life, consistency, maximum charge current and maximum discharge current.

[0075] Step S12: Calculate the mutual information value between the lithium battery terminal voltage and the measurement parameter group and / or the physical parameter group respectively by using the mutual information variable selection method.

[0076] Assume that the battery terminal voltage is X, and the measurement parameter group or the physical parameter group is Y. If X and Y are two discrete random variables, the mutual information value between X and Y is defined according to the following formula:

[0077]

[0078] Wherein, I(X,Y) represents the mutual information value between the battery terminal voltage X and the measurement parameter group or the physical parameter group Y, β represents a parameter in the measurement parameter group or the physical parameter group Y, α represents a parameter in the battery terminal voltage X, P(α,β) represents the joint probability distribution function of α and β, P(α) represents the marginal probability distribution function of α, and P(β) represents the marginal probability distribution function of β.

[0079] The above formula is applicable to the case where the learner is a classification model. When the learner is a regression model, that is, when X and Y are continuous random variables, the summation sign in the above formula needs to be replaced with an integral sign:

[0080]

[0081] Step S2: constructing a data sample based on the mutual information value, specifically comprising:

[0082] Step S21: determine whether the mutual information value is greater than a set threshold; if the mutual information value is greater than the set threshold, the lithium battery terminal voltage whose mutual information value is greater than the set threshold and each parameter in the measurement parameter group and / or the physical parameter group are used as characteristic variables.

[0083] Step S22: taking the plurality of characteristic variables obtained multiple times as data samples.

[0084] Step S3: constructing a stacked denoising autoencoder network based on the data sample, specifically including:

[0085] Step S31: training a first denoising autoencoder based on the data sample, specifically comprising:

[0086] Step S311: giving the initialization weights of the first denoising autoencoder input layer.

[0087] Step S312: Select multiple feature variables from the data sample.

[0088] Step S313: training the first denoising autoencoder based on the initialization weights of the first denoising autoencoder input layer and the plurality of feature variables.

[0089] Step S314: using the error function and BP algorithm to reversely calculate the two weight changes of the first denoising autoencoder.

[0090] Step S315: using the gradient descent method, the weights required for the next training are calculated according to the two weight changes.

[0091] Step S316: Determine whether the number of iterations is greater than or equal to the maximum number of iterations; if the number of iterations is greater than or equal to the maximum number of iterations, execute "Step S32"; if the number of iterations is less than the maximum number of iterations, use the weights required for the next training as the initialization weights of the first denoising autoencoder input layer, and return to "Step S312".

[0092] The output formula of the output layer of the first denoising autoencoder after training is:

[0093]

[0094] Where z represents the output of the output layer of the first denoising autoencoder after training, f(·) represents the sigmoid function, as 1 t 1 represents the initialization weight of the first denoising autoencoder input layer, b 1 represents the bias vector of the first denoising autoencoder, as 2 t 2 represents the hidden layer weights of the first denoising autoencoder after training, b 2 represents the bias vector of the first denoising autoencoder after training, Indicates that the characteristic variables are obtained by random noise processing.

[0095] The error function is:

[0096]

[0097] Wherein, L represents the error of the first denoising autoencoder, i represents the feature variable number, sj represents multiple input feature variables, and z represents the output of the output layer of the first denoising autoencoder after training.

[0098] Step S32: using the output of the hidden layer of the trained first denoising autoencoder as the input of the input layer of the second denoising autoencoder to train the second denoising autoencoder.

[0099] Step S321: The weights required for the next training obtained from the last training of the first denoising autoencoder are used as the initialization weights of the input layer of the second denoising autoencoder, and the number of neurons and the activation function of the second denoising autoencoder are given.

[0100] Step S322: using the trained output of the hidden layer of the first denoising autoencoder as the input of the input layer of the second denoising autoencoder.

[0101] Step S323: training the second denoising autoencoder based on the initialization weights of the second denoising autoencoder and the output of the hidden layer of the trained first denoising autoencoder.

[0102] Step S324: using the error function and BP algorithm to reversely obtain two weight changes of the second denoising autoencoder;

[0103] Step S325: using the gradient descent method to calculate the weights required for the next training according to the two weight changes;

[0104] Step S326: Determine whether the number of iterations is greater than or equal to the maximum number of iterations; if the number of iterations is greater than or equal to the maximum number of iterations, execute "Step S33"; if the number of iterations is less than the maximum number of iterations, use the weights required for the next training as the initialization weights of the second denoising autoencoder input layer, and return to "Step S323".

[0105] The output formula of the output layer of the second denoising autoencoder after training is the same as the solution method of the output formula of the output layer of the first denoising autoencoder after training, and the error function of the second denoising autoencoder is the same as the solution method of the error function of the first denoising autoencoder, which will not be repeated here.

[0106] In the above embodiment, the maximum number of iterations is 50, that is, the training stops at the 50th iteration.

[0107] Step S33: stacking the trained first denoising autoencoder and the trained second denoising autoencoder to form a stacked denoising autoencoder network. Specifically, retaining the input layer and hidden layer of the trained first denoising autoencoder, and using the hidden layer of the trained second denoising autoencoder as the output layer to form a stacked denoising autoencoder network.

[0108] Step S4: Based on the regression method, a regression prediction model of the lithium battery terminal voltage is established according to the stacked denoising autoencoder network, which specifically includes:

[0109] Step S41: Based on the regression method, the data sample is input into the stacked denoising autoencoder network to obtain the vector of the output layer of the stacked denoising autoencoder network.

[0110] Step S42: extract the vector of the output layer of the stack denoising autoencoder network through matrix transformation to obtain parameters that are strongly correlated with the terminal voltage of the lithium battery; the mutual information value ranges from 0 to 1. The present invention divides the mutual information value into three levels, and considers that from 0 to 0.3 is weakly correlated, from 0.3 to 0.6 is moderately correlated, and from 0.6 to 1 is strongly correlated.

[0111] Step S43: establishing a regression prediction model for the lithium battery terminal voltage according to the lithium battery terminal voltage and parameters strongly correlated with the lithium battery terminal voltage.

[0112] The regression method described in the present invention is support vector or least square method.

[0113] Step S5: Calculate the predicted value of the terminal voltage of the lithium battery in the battery energy storage system according to the stack denoising autoencoder network and the regression prediction model. Specifically, calculate the predicted value of the terminal voltage of the lithium battery in the battery energy storage system according to the characteristic variables, the stack denoising autoencoder network and the regression prediction model.

[0114] Step S6: judging whether to give an alarm according to the predicted value of the lithium battery terminal voltage in the battery energy storage system, specifically comprising:

[0115] Step S61: Compare the predicted value of the lithium battery terminal voltage with the lithium battery terminal voltage to obtain an error; the above-mentioned lithium battery terminal voltages are all actual values ​​of the lithium battery terminal voltages.

[0116] Step S62: Determine whether the error is within a set range; if the error is within the set range, it indicates that the lithium battery in the battery energy storage system is working normally; if the error is not within the set range, the monitoring system issues an alarm to achieve real-time monitoring of the lithium battery in the battery energy storage system.

[0117] like Figure 2As shown, the present invention also provides a lithium battery energy storage system monitoring system based on noise reduction autoencoding, the system comprising:

[0118] The mutual information value calculation module 1 is used to calculate the mutual information value between the lithium battery terminal voltage and each parameter in the measurement parameter group and / or the physical parameter group.

[0119] The data sample construction module 2 constructs a data sample based on the mutual information value.

[0120] The stack denoising autoencoder network construction module 3 constructs a stack denoising autoencoder network based on data samples.

[0121] The regression prediction model determination module 4 establishes a regression prediction model of the lithium battery terminal voltage based on the stack denoising autoencoder network based on a regression method.

[0122] The lithium battery terminal voltage prediction value determination module 5 is used to calculate the lithium battery terminal voltage prediction value in the battery energy storage system according to the stacked denoising autoencoder network and the regression prediction model.

[0123] The alarm judging module 6 is used to judge whether to give an alarm according to the predicted value of the lithium battery terminal voltage in the battery energy storage system.

[0124] As an implementation mode, the mutual information value calculation module 1 of the present invention specifically includes:

[0125] An acquisition unit is used to acquire the lithium battery terminal voltage, measurement parameter group and physical parameter group in the lithium battery energy storage system; the measurement parameter group includes charge and discharge efficiency, charge rate, discharge rate, state of charge and battery health status; the physical parameter group includes battery capacity, rated voltage, maximum discharge rate, discharge cut-off voltage, charge and discharge current, standard operating temperature, battery temperature, cycle life, consistency, maximum charge current and maximum discharge current.

[0126] The mutual information value calculation unit is used to calculate the mutual information value between the lithium battery terminal voltage and the measurement parameter group and / or the physical parameter group respectively by using the variable selection method of mutual information.

[0127] As an implementation mode, the data sample construction module 2 of the present invention specifically includes:

[0128] A screening unit is used to determine whether the mutual information value is greater than a set threshold; if the mutual information value is greater than the set threshold, the lithium battery terminal voltage whose mutual information value is greater than the set threshold and each parameter in the measurement parameter group and / or the physical parameter group are used as characteristic variables.

[0129] The data sample determination unit is used to use the plurality of characteristic variables obtained multiple times as data samples.

[0130] As an implementation mode, the stacked denoising autoencoder network construction module 3 of the present invention specifically includes:

[0131] The first training unit trains a first denoising autoencoder based on the data sample.

[0132] The second training unit is used to use the output of the hidden layer of the trained first denoising autoencoder as the input of the input layer of the second denoising autoencoder to train the second denoising autoencoder.

[0133] The stacking unit is used to stack the trained first denoising autoencoder and the trained second denoising autoencoder to form a stacked denoising autoencoder network.

[0134] As an implementation mode, the first training unit of the present invention specifically includes:

[0135] The first given subunit is used to give the initialization weights, the number of neurons and the activation function of the first denoising autoencoder input layer.

[0136] The selection subunit is used to select multiple feature variables from the data sample.

[0137] The first training subunit trains the first denoising autoencoder based on the initialization weights of the first denoising autoencoder input layer and the plurality of feature variables.

[0138] The first weight change determination subunit is used to reversely calculate two weight changes of the first denoising autoencoder using the error function and the BP algorithm.

[0139] The first weight determination subunit is used to calculate the weight required for the next training according to the two weight changes by using a gradient descent method.

[0140] The first judgment subunit is used to judge whether the number of iterations is greater than or equal to the maximum number of iterations; if the number of iterations is greater than or equal to the maximum number of iterations, the "second training unit" is executed; if the number of iterations is less than the maximum number of iterations, the weights required for the next training are used as the initialization weights of the first denoising autoencoder input layer, and the "selection subunit" is returned.

[0141] As an implementation manner, the second training unit of the present invention specifically includes:

[0142] The second given subunit is used to use the weights required for the next training obtained from the last training of the first denoising autoencoder as the initialization weights of the input layer of the second denoising autoencoder, and to give the number of neurons and the activation function of the second denoising autoencoder.

[0143] The assignment subunit is used to use the output of the hidden layer of the first denoising autoencoder after training as the input of the input layer of the second denoising autoencoder.

[0144] The second training subunit trains a second denoising autoencoder based on the initialization weights of the second denoising autoencoder and the output of the hidden layer of the trained first denoising autoencoder.

[0145] The second weight change determination subunit is used to reversely calculate the two weight changes of the second denoising autoencoder using the error function and the BP algorithm.

[0146] The second weight determination subunit is used to calculate the weight required for the next training according to the two weight changes by using the gradient descent method.

[0147] The second judgment subunit is used to judge whether the number of iterations is greater than or equal to the maximum number of iterations; if the number of iterations is greater than or equal to the maximum number of iterations, the "stacking unit" is executed; if the number of iterations is less than the maximum number of iterations, the weights required for the next training are used as the initialization weights of the second denoising autoencoder input layer, and the "second training subunit" is returned.

[0148] As an implementation mode, the regression prediction model determination module 4 of the present invention specifically includes:

[0149] The vector determination unit inputs the data sample into the stacked denoising autoencoder network based on a regression method to obtain a vector of an output layer of the stacked denoising autoencoder network.

[0150] The extraction unit is used to extract the vector of the output layer of the stack denoising autoencoder network through matrix transformation to obtain parameters that are strongly correlated with the terminal voltage of the lithium battery; the mutual information value is strongly correlated in the range of 0.6-1.

[0151] The regression prediction model determination unit is used to establish a regression prediction model of the lithium battery terminal voltage according to the lithium battery terminal voltage and a parameter strongly correlated with the lithium battery terminal voltage.

[0152] As an implementation mode, the alarm determination module 6 of the present invention specifically includes:

[0153] The comparison unit is used to compare the predicted value of the lithium battery terminal voltage with the lithium battery terminal voltage to obtain an error.

[0154] The judgment unit is used to judge whether the error is within a set range; if the error is within the set range, it means that the lithium battery in the battery energy storage system is working normally; if the error is not within the set range, the monitoring system will issue an alarm.

[0155] Specific examples:

[0156] like Figure 3 As shown in the figure, it can be seen that the mutual information values ​​of several features such as capacity and state of charge are relatively large. Here, the threshold value is set to 0.3, and a few features with mutual information values ​​higher than 0.3 are screened out, which are considered to be features that contribute more to the amount of label information. The feature variables finally screened out are: capacity, standard operating temperature, discharge rate, state of charge, battery health status and cycle life. SCADA was used to test 100 feature variables to form a data sample.

[0157] The present invention adopts a stacked denoising autoencoder network structure composed of two denoising autoencoders. The denoising autoencoder is actually based on the autoencoder. In order to prevent the overfitting problem, noise is added to the input data of the input layer, so that the autoencoder can improve the ability to eliminate noise during learning and has stronger survivability under abnormal conditions. The denoising autoencoder is regarded as an encoder composed of an input layer and a hidden layer and a decoder composed of a hidden layer and an output layer.

[0158] The autoencoder network is a shallow neural network. Its purpose is to make the input and output consistent as much as possible. The stacked denoising autoencoder network parameters need to be initialized. The input layer is 6-dimensional data, the hidden layer is set to 4-dimensional data, and the output layer is 6-dimensional data. Then the network structure of the first DAE is 6-4-6. The network structure of the second DAE is set to 4-3-4, so the network structure of the stacked SDAE is 6-4-3.

[0159] Based on the regression method, the data is written into the input of SDAE, and the vector output by the second output layer can be obtained through the same steps as in training. The features in the output layer of the second network extracted through matrix transformation are the influencing factors that are reduced in dimension and strongly correlated with the lithium battery terminal voltage V required by this patent. After the vector is extracted and converted into influencing factors that are strongly correlated with the lithium battery terminal voltage through matrix transformation, a regression prediction model for the lithium battery terminal voltage is established based on the lithium battery terminal voltage v and its strongly correlated influencing factors.

[0160] The predicted value a of the lithium battery terminal voltage is calculated based on the characteristic variables, the stack denoising autoencoder network and the regression prediction model.

[0161] The predicted value a of the lithium battery terminal voltage is compared with the actual value b of the lithium battery terminal voltage to obtain an error value c. If the numerical error between the two is within the error range specified by the system, it can be regarded as normal operation; if the numerical error between the two is not within the error range specified by the system, the monitoring system will issue an alarm, and the staff will then conduct maintenance on it, thus realizing real-time monitoring of the lithium battery terminal voltage.

[0162] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0163] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A lithium battery energy storage system monitoring method based on noise reduction autoencoding, characterized in that: The method comprises: Step S1: Calculate the mutual information value between the lithium battery terminal voltage and each parameter in the measurement parameter group and / or the physical parameter group; Step S2: constructing a data sample based on the mutual information value; Step S3: constructing a stacked denoising autoencoder network based on the data sample; Step S4: Based on a regression method, a regression prediction model of the lithium battery terminal voltage is established according to the stacked denoising autoencoder network; Step S5: calculating a predicted value of a lithium battery terminal voltage in a battery energy storage system according to the stacked denoising autoencoder network and the regression prediction model; Step S6: judging whether to give an alarm according to the predicted value of the lithium battery terminal voltage in the battery energy storage system; The stack denoising autoencoder network is constructed based on the data sample, specifically comprising: Step S31: training a first denoising autoencoder based on the data sample; Step S32: using the output of the hidden layer of the trained first denoising autoencoder as the input of the input layer of the second denoising autoencoder to train the second denoising autoencoder; Step S33: stacking the trained first denoising autoencoder and the trained second denoising autoencoder to form a stacked denoising autoencoder network; The step of training the first denoising autoencoder based on the data sample specifically includes: Step S311: giving the initialization weights, number of neurons and activation function of the first denoising autoencoder input layer; Step S312: selecting a plurality of characteristic variables from the data sample; Step S313: training the first denoising autoencoder based on the initialization weights of the first denoising autoencoder input layer and the plurality of feature variables; Step S314: using the error function and BP algorithm to reversely calculate two weight changes of the first denoising autoencoder; Step S315: using the gradient descent method to calculate the weight required for the next training according to the two weight changes; Step S316: Determine whether the number of iterations is greater than or equal to the maximum number of iterations; if the number of iterations is greater than or equal to the maximum number of iterations, execute "Step S32"; if the number of iterations is less than the maximum number of iterations, use the weights required for the next training as the initialization weights of the first denoising autoencoder input layer, and return to "Step S312".

2. The method for monitoring a lithium battery energy storage system based on noise reduction autoencoding according to claim 1, characterized in that: The regression method is based on the stacked denoising autoencoder network to establish a regression prediction model for the terminal voltage of the lithium battery, specifically including: Step S41: Based on a regression method, the data sample is input into the stacked denoising autoencoder network to obtain a vector of an output layer of the stacked denoising autoencoder network; Step S42: extracting the vector of the output layer of the stacked denoising autoencoder network through matrix transformation to obtain parameters that are strongly correlated with the terminal voltage of the lithium battery; the mutual information value is strongly correlated in the range of 0.6-1; Step S43: establishing a regression prediction model for the lithium battery terminal voltage according to the lithium battery terminal voltage and parameters strongly correlated with the lithium battery terminal voltage.

3. The method for monitoring a lithium battery energy storage system based on noise reduction autoencoding according to claim 1, characterized in that: The determining whether to issue an alarm according to the predicted value of the lithium battery terminal voltage in the battery energy storage system specifically includes: Step S61: Compare the predicted value of the lithium battery terminal voltage with the lithium battery terminal voltage to obtain an error; Step S62: Determine whether the error is within a set range; if the error is within the set range, it indicates that the lithium battery in the battery energy storage system is working normally; if the error is not within the set range, the monitoring system issues an alarm.

4. A lithium battery energy storage system monitoring system based on noise reduction autoencoding, characterized in that: The system comprises: A mutual information value calculation module, used to calculate the mutual information value between the lithium battery terminal voltage and each parameter in the measurement parameter group and / or the physical parameter group; A data sample construction module, constructing a data sample based on the mutual information value; The stack denoising autoencoder network building module builds a stack denoising autoencoder network based on data samples; A regression prediction model determination module, based on a regression method, establishes a regression prediction model of the lithium battery terminal voltage according to the stack denoising autoencoder network; A lithium battery terminal voltage prediction value determination module, used to calculate the lithium battery terminal voltage prediction value in the battery energy storage system according to the stack denoising autoencoder network and the regression prediction model; An alarm judging module is used to judge whether to give an alarm according to the predicted value of the lithium battery terminal voltage in the battery energy storage system; The stack denoising autoencoder network building module specifically includes: A first training unit, training a first denoising autoencoder based on the data sample; A second training unit is used to use the output of the hidden layer of the trained first denoising autoencoder as the input of the input layer of the second denoising autoencoder to train the second denoising autoencoder; A stacking unit, configured to stack the trained first denoising autoencoder and the trained second denoising autoencoder to form a stacked denoising autoencoder network; The first training unit specifically includes: A first given subunit, used to give the initialization weights, number of neurons and activation function of the first denoising autoencoder input layer; A selection subunit, used for selecting a plurality of characteristic variables from the data sample; A first training subunit trains a first denoising autoencoder based on the initialization weights of the first denoising autoencoder input layer and the plurality of feature variables; A first weight change determination subunit is used to reversely calculate two weight changes of the first denoising autoencoder using an error function and a BP algorithm; A first weight determination subunit is used to calculate the weight required for the next training according to the two weight changes by using a gradient descent method; The first judgment subunit is used to judge whether the number of iterations is greater than or equal to the maximum number of iterations; if the number of iterations is greater than or equal to the maximum number of iterations, the "second training unit" is executed; if the number of iterations is less than the maximum number of iterations, the weights required for the next training are used as the initialization weights of the input layer of the first denoising autoencoder, and the "selection subunit" is returned.

5. The lithium battery energy storage system monitoring system based on noise reduction autoencoding according to claim 4 is characterized in that: The regression prediction model determination module specifically includes: A vector determination unit, based on a regression method, inputs the data sample into the stacked denoising autoencoder network to obtain a vector of an output layer of the stacked denoising autoencoder network; An extraction unit is used to extract the vector of the output layer of the stack denoising autoencoder network through matrix transformation to obtain a parameter strongly correlated with the terminal voltage of the lithium battery; the mutual information value is strongly correlated in the range of 0.6-1; The regression prediction model determination unit is used to establish a regression prediction model of the lithium battery terminal voltage according to the lithium battery terminal voltage and a parameter strongly correlated with the lithium battery terminal voltage.

6. The lithium battery energy storage system monitoring system based on noise reduction autoencoding according to claim 4 is characterized in that: The alarm determination module specifically includes: A comparison unit, used for comparing the predicted value of the lithium battery terminal voltage with the lithium battery terminal voltage to obtain an error; The judgment unit is used to judge whether the error is within a set range; if the error is within the set range, it means that the lithium battery in the battery energy storage system is working normally; if the error is not within the set range, the monitoring system will issue an alarm.

Images