Energy storage system multi-level fault early warning method based on data-knowledge driving

Abstract

The invention discloses an energy storage system multi-level fault early warning method based on data-knowledge driving, belongs to the technical field of energy storage system fault early warning, and aims to predict temperatures of key devices (power devices and capacitors) in a battery and a converter and realize energy storage system multi-level fault early warning.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy storage system fault warning, and in particular relates to a multi-level fault warning method for energy storage systems based on data-knowledge driving. Background Art

[0002] With the rapid development of renewable energy and the widespread application of energy storage technology, energy storage systems play an increasingly important role in power systems. However, the complexity and diversity of energy storage systems make them prone to various faults during operation, affecting their safety and stability. Therefore, it is particularly important to establish an effective fault warning mechanism.

[0003] Existing fault warning methods mostly rely on traditional data analysis and empirical judgment, lacking in-depth mining and integration of system internal knowledge. This method often fails to identify potential fault risks in a timely manner, resulting in equipment damage, energy loss, and safety hazards. Therefore, there is an urgent need for a novel fault warning method that can integrate the advantages of data-driven and knowledge-driven to achieve comprehensive monitoring and warning of multi-level faults in energy storage systems.

[0004] Based on this, this patent proposes a data-knowledge driven multi-level fault warning method for energy storage systems. By combining real-time data and historical knowledge, a more intelligent and efficient fault warning mechanism is constructed. This method not only improves the accuracy of fault identification, but also enhances the system's adaptive ability, providing a strong guarantee for the safe operation of the energy storage system. Summary of the invention

[0005] The purpose of the present invention is to propose a multi-level fault warning method for energy storage systems based on data-driven.

[0006] The multi-level fault warning method for energy storage system based on data-knowledge drive mainly comprises the following steps:

[0007] S1. Establish an electrothermal coupling model for lithium batteries;

[0008] S2. Establish a thermoelectric coupling model of power devices in energy storage converters;

[0009] S3. Establish a thermoelectric coupling model of capacitors in energy storage converters;

[0010] S4. Design the temperature warning level of the energy storage system;

[0011] S5. Extract the historical operating data of lithium battery current, voltage, battery surface temperature, and battery ambient temperature;

[0012] S6. Extracting the ambient temperature and output power historical operation data of the energy storage converter;

[0013] S7. Decompose the lithium battery current, voltage, surface temperature, ambient temperature, energy storage converter ambient temperature, and output power based on the Prophet algorithm;

[0014] S8. Regression of Prophet decomposition data based on XGBoost algorithm;

[0015] S9. Realize lithium battery current, voltage, surface temperature, ambient temperature, and energy storage converter ambient temperature and output power prediction based on Prophet-XGBoost model;

[0016] S10. Predict the temperature of key components (power devices, capacitors) of lithium batteries and energy storage converters based on the prediction information;

[0017] S11. Realize multi-level fault warning of energy storage system based on the predicted temperature of lithium battery and energy storage converter.

[0018] The multi-level fault warning method of energy storage system based on data-knowledge drive establishes a lithium battery thermoelectric coupling model, such as Figure 2 As shown, according to the Bernardi heat generation model, the heat generation rate Q of lithium-ion power batteries re It can be divided into four parts: irreversible heat, reversible heat, concentration gradient heat and phase change reaction heat.

[0019]

[0020] The first part of the above formula is the irreversible heat of the battery, which represents the heat generated by the energy loss caused by the battery's potential deviating from the circuit voltage; the second part is the reversible heat generated by the battery, which represents the heat caused by entropy heat; the third part represents the concentration gradient heat Q generated by the battery mix , which indicates the heat generated by the formation and relaxation of concentration gradients caused by the different reaction rates of different parts of the active material during the reaction inside the battery; the fourth part is the heat rate Q generated by the battery phase change reaction pc , which indicates the heat generated or consumed by any chemical reaction occurring in the power lithium battery;

[0021] In the normal operation of lithium-ion power batteries, the reversible heat and irreversible heat are of the same order of magnitude, and the sum of the heat generated by these two parts accounts for the vast majority of the total heat generation. Therefore, the heat generated by the concentration gradient heat and the phase change reaction heat in equation (1) can be ignored. Therefore, the heat generation rate model can be simplified as:

[0022]

[0023] Where I is the battery charge and discharge current, V and OCV represent the battery terminal voltage and open circuit voltage respectively, T is the battery core temperature, and dOCV / dT is the battery temperature entropy coefficient;

[0024] Establishing a lumped parameter two-state thermal model for lithium-ion power batteries

[0025]

[0026] make

[0027] T c -T f =T cf (5)

[0028] T s -T f =T sf (6)

[0029] therefore

[0030]

[0031] Performing the Laplace transformation, we can get

[0032]

[0033] The multi-level fault warning method of energy storage system based on data-knowledge drive establishes a thermoelectric coupling model of power devices in energy storage converter, such as Figure 3 As shown in Figure 1, the power device electrical loss is divided into switching loss and conduction loss, and the calculation formula is:

[0034]

[0035] In the formula, is the conduction loss of the power device; δ(t) is the duty cycle; i a is the output current of the power device; T j is the junction temperature of the power device; V CE_25℃ is the rated on-state voltage drop of the power device at 25°C; r CE_25℃ K is the rated on-state resistance of the power device at 25°C; V_T K is the temperature influence coefficient of the on-state voltage drop of the power device; r_T is the temperature influence coefficient of the on-state resistance of the power device; f sw K is the frequency of turning on and off the power device module; sw K is the temperature influence coefficient of the switching loss of the power device; 1 (V dc ) is the voltage coefficient of the converter DC side voltage that affects the switching power loss of the power device; K 2 (R g ) is the gate resistance R g The resistance coefficient that affects the switching power loss of power devices;

[0036] Use the Foster model and power loss model to perform thermoelectric conversion of power devices and convert the power loss profile into a thermal stress profile

[0037]

[0038] Where, T j is the junction temperature of the power device; T a is the ambient temperature; is the power loss of the power device and the anti-parallel diode; Z tjh , Z ha is the thermal impedance parameter in the Foster model.

[0039] The multi-level fault warning method of energy storage system based on data-knowledge drive establishes a thermoelectric coupling model of capacitors in energy storage converters, such as Figure 4 As shown, the total power loss of DC capacitor can be expressed as

[0040]

[0041] In the formula, R ESR is the ESR of the DC bus capacitor, I c is the current harmonic amplitude of the capacitor, then the hot spot temperature of the DC capacitor can be expressed as

[0042] T h =P c,loss R th +T a (16)

[0043] Where, T h is the hot spot temperature of the capacitor; T a is the ambient temperature; R th is the thermal resistance of the capacitor given in the datasheet.

[0044] The multi-level fault warning method for energy storage system based on data-knowledge driven is designed to design the temperature warning level of energy storage system. Battery system failure mainly manifests in accelerated aging and thermal runaway of battery pack, which is set as serious failure, general failure and no failure.

[0045] The multi-level fault warning method for energy storage system based on data-knowledge drive extracts historical operating data of lithium battery current, voltage, battery surface temperature and battery ambient temperature.

[0046] The multi-level fault warning method for energy storage system based on data-knowledge drive extracts the ambient temperature and output power historical operation data of the energy storage converter.

[0047] The multi-level fault warning method for energy storage system based on data-knowledge drive is based on the Prophet algorithm to decompose the lithium battery current, voltage, surface temperature, ambient temperature, energy storage converter ambient temperature, and output power, such as Figure 5 As shown in the figure, Prophet is a time series trend and cycle forecasting model. Its overall structure can be composed of three parts: trend term g(t), seasonal term s(t), and holiday term h(t). Its combination formula is:

[0048] P(t)=g(t)+s(t)+h(t)+ε t (17)

[0049] In the formula, e t is the error term, which is usually assumed to follow a normal distribution with a mean of 0, and is used to reflect abnormal changes that are not reflected in the model; the trend term g(t) is used to capture the long-term trend of changes in the time series, including parameters for adjusting the smoothness of the model and assumptions of varying degrees, mainly to model non-periodic changes such as piecewise linear growth in the time series. The basic form is

[0050] g(t)=(k+α(t) T δ)t+(b+α(t) T γ)(18)

[0051] Where k is the growth rate; b is the bias parameter; d is the change in growth rate; g is the vector of the offset adjustment process at the point where the growth rate k changes; a(t) T Whether to adjust the slope or offset (1 if yes, 0 otherwise);

[0052] The seasonal term s(t) uses Fourier series to approximate the periodic variation, and its expression is:

[0053]

[0054] Where N is the total number of cycles; P is a fixed cycle; 2n is the number of cycles expected to be used in the model, and a n and b n These are parameters that need to be estimated;

[0055] The holiday term h(t) is used to consider abnormal data fluctuations that occur at special values. Such fluctuations cannot be realized by the periodic model, so independent models are established for different special values. The specific form is:

[0056] h(t)=Z(t)κ(20)

[0057] Z(t)=[I(t∈D 1 ),I(t∈D 2 ),…,I(t∈D L)](twenty one)

[0058] κ∈Normal(0,v 2 )(twenty two)

[0059] Where D i is the i-th special value set; I(t∈D i ) is the indicator function, the value of the i-th special value is 1, otherwise it is 0; k is the special value setting parameter; v is the influence factor of the special value on the prediction result.

[0060] The multi-level fault warning method for energy storage system based on data-knowledge driven is based on XGBoost algorithm to regress Prophet decomposition data, such as Figure 6 As shown, the XGBoost algorithm can be understood as a sum model of multiple decision trees. Assuming that the model has k decision trees, that is:

[0061]

[0062] Where k is the number of trees; f t is a function in the function space F; is the model prediction value; x i is the i-th data input; Represents sample x i The leaf node weight that falls to the corresponding leaf node; q(x i ) represents the sample x i Fall to the corresponding leaf node;

[0063] The XGBoost algorithm uses the addition and forward distribution algorithm, and each iteration does not affect the original model, that is,

[0064]

[0065] The XGBoost algorithm objective function L consists of two parts, the first is the loss function and the second is the regularization term.

[0066]

[0067] In the formula, is the error between the predicted value and the true value; Ω(f j ) is a regular term to prevent the model from overfitting; T is the number of leaf nodes; γ is the penalty coefficient of T; w is the leaf node weight; n is the number of samples.

[0068] When training to the tth lesson tree, the objective function is corrected to:

[0069]

[0070] Since the first t-1 regression trees are known information, that is, Ω(fj ) is a constant and has no effect on the optimization process of the tth tree, so the objective function can be simplified as:

[0071]

[0072] Rewrite the objective function into a leaf node traversal:

[0073]

[0074] According to the Taylor series expansion of the binary function, It can be expressed as:

[0075]

[0076] Simplify the objective function and remove the constant term Right now

[0077]

[0078] In the formula, when hour, Get the minimum value,

[0079]

[0080] The multi-level fault warning method for energy storage system based on data-knowledge driven is based on the Prophet-XGBoost model to realize the lithium battery current, voltage, surface temperature, ambient temperature; energy storage converter ambient temperature, output power prediction, such as Figure 7 shown.

[0081] The multi-level fault warning method for energy storage system based on data-knowledge drive predicts the temperature of lithium batteries and key components (power devices, capacitors) of energy storage converters based on prediction information.

[0082] The data-driven multi-level fault warning method for energy storage systems implements multi-level fault warning for energy storage systems based on predicted temperatures of lithium batteries and energy storage converters. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 A multi-level fault warning flow chart for energy storage systems based on data-knowledge driven;

[0084] Figure 2 Lithium battery thermoelectric coupling model;

[0085] Figure 3 Power device hot spot coupling model; Figure 4 Capacitor Hot Spot Coupling Model Figure 5Prophet Algorithm

[0086] Figure 6 XGBoost Algorithm

[0087] Figure 7 Prophet-XGBoost algorithm.

Claims

1. A multi-level fault warning method for energy storage system based on data-knowledge driven, characterized in that: The method mainly includes the following steps: S1. Establish an electrothermal coupling model for lithium batteries; S2. Establish a thermoelectric coupling model of power devices in energy storage converters; S3. Establish a thermoelectric coupling model of capacitors in energy storage converters; S4. Design the temperature warning level of the energy storage system; S5. Extract the historical operating data of lithium battery current, voltage, battery surface temperature, and battery ambient temperature; S6. Extracting the ambient temperature and output power historical operation data of the energy storage converter; S7. Decompose the lithium battery current, voltage, surface temperature, ambient temperature, energy storage converter ambient temperature, and output power based on the Prophet algorithm; S8. Regression of Prophet decomposition data based on XGBoost algorithm; S9. Realize lithium battery current, voltage, surface temperature, ambient temperature, and energy storage converter ambient temperature and output power prediction based on Prophet-XGBoost model; S10. Predict the temperature of key components (power devices, capacitors) of lithium batteries and energy storage converters based on the prediction information; S11. Realize multi-level fault warning of energy storage system based on the predicted temperature of lithium battery and energy storage converter.

2. A data-knowledge driven multi-level fault early warning method for energy storage system according to claim 1, characterized in that: Establish a thermoelectric coupling model for lithium batteries. According to the Bernardi heat generation model, the heat generation rate Q of lithium-ion power batteries re It can be divided into four parts: irreversible heat, reversible heat, concentration gradient heat and phase change reaction heat. The first part of the above formula is the irreversible heat of the battery, which represents the heat generated by the energy loss caused by the battery's potential deviating from the circuit voltage; the second part is the reversible heat generated by the battery, which represents the heat caused by entropy heat; the third part represents the concentration gradient heat Q generated by the battery mix , which indicates the heat generated by the formation and relaxation of concentration gradients caused by the different reaction rates of different parts of the active material during the reaction inside the battery; the fourth part is the heat rate Q generated by the battery phase change reaction pc , which indicates the heat generated or consumed by any chemical reaction occurring in the power lithium battery; During the normal operation of lithium-ion power batteries, the reversible heat and irreversible heat are of the same order of magnitude, and the sum of the heat generated by these two parts accounts for the vast majority of the total heat generation. Therefore, the heat generated by the concentration gradient heat and the phase change reaction heat in formula (1) is ignored. Therefore, the heat generation rate model can be simplified as follows: Where I is the battery charge and discharge current, V and OCV represent the battery terminal voltage and open circuit voltage respectively, T is the battery core temperature, and dOCV / dT is the battery temperature entropy coefficient; Establishing a lumped parameter two-state thermal model for lithium-ion power batteries make T c -T f =T cf (5) T s -T f =T sf (6) therefore Performing the Laplace transformation, we can get 3. A data-knowledge driven multi-level fault early warning method for energy storage system according to claim 1, characterized in that: The thermoelectric coupling model of the power device in the energy storage converter is established. The power device electrical loss is divided into switching loss and conduction loss, and the calculation formula is: In the formula, is the conduction loss of the power device; δ(t) is the duty cycle; i a is the output current of the power device; T j is the junction temperature of the power device; V CE_25℃ is the rated on-state voltage drop of the power device at 25°C; r CE_25℃ K is the rated on-state resistance of the power device at 25°C; V_T K is the temperature influence coefficient of the on-state voltage drop of the power device; r_T is the temperature influence coefficient of the on-state resistance of the power device; f sw K is the frequency of turning on and off the power device module; sw is the temperature influence coefficient of the switching loss of the power device; K1(V dc ) is the voltage coefficient of the converter DC side voltage that affects the switching power loss of the power device; K2(R g ) is the gate resistance R g The resistance coefficient that affects the switching power loss of power devices; Use the Foster model and power loss model to perform thermoelectric conversion of power devices and convert the power loss profile into a thermal stress profile Where, T j is the junction temperature of the power device; T a is the ambient temperature; is the power loss of the power device and the anti-parallel diode; Z tjh , Z ha is the thermal impedance parameter in the Foster model.

4. A data-knowledge driven multi-level fault early warning method for energy storage system according to claim 1, characterized in that: The thermoelectric coupling model of the capacitor in the energy storage converter is established, and the total power loss of the DC capacitor can be expressed as In the formula, R ESR is the ESR of the DC bus capacitor, I c is the current harmonic amplitude of the capacitor, then the hot spot temperature of the DC capacitor can be expressed as T h =P c,loss R th +T a (16) Where, T h is the hot spot temperature of the capacitor; T a is the ambient temperature; R th is the thermal resistance of the capacitor given in the datasheet.

5. A data-knowledge driven multi-level fault early warning method for energy storage system according to claim 1, characterized in that: Design the temperature warning level of the energy storage system. Battery system failures are mainly manifested in accelerated aging and thermal runaway of the battery pack, and are set to severe failure, general failure, and no failure.

6. A data-knowledge driven multi-level fault early warning method for energy storage system according to claim 1, characterized in that: Extract historical operating data of lithium battery current, voltage, battery surface temperature, and battery ambient temperature.

7. A data-knowledge driven multi-level fault early warning method for energy storage system according to claim 1, characterized in that: Extract the historical operating data of the ambient temperature and output power of the energy storage converter.

8. The data-knowledge driven multi-level fault early warning method for energy storage system according to claim 1, characterized in that: Based on the Prophet algorithm, the lithium battery current, voltage, surface temperature, ambient temperature, energy storage converter ambient temperature, and output power are decomposed. As a time series trend and cycle prediction model, Prophet as a whole can be composed of three parts: trend term g(t), seasonal term s(t), and holiday term h(t). The combination formula is: P(t)=g(t)+s(t)+h(t)+ε t (17) In the formula, e t It is the error term, which is usually assumed to follow a normal distribution with a mean of 0 and is used to reflect abnormal changes that are not reflected in the model; The trend term g(t) is used to capture the long-term trend of changes in the time series, including parameters for adjusting the smoothness of the model and assumptions of varying degrees. It is mainly used to model non-periodic changes such as piecewise linear growth in the time series. The basic form is g(t)=(k+α(t) T d)t+(b+a(t) T c) (18) Where k is the growth rate; b is the bias parameter; d is the change in growth rate; g is the vector of the offset adjustment process at the point where the growth rate k changes; a(t) T Whether to adjust the slope or offset (1 if yes, 0 otherwise); The seasonal term s(t) uses Fourier series to approximate the periodic variation, and its expression is: Where N is the total number of cycles; P is a fixed cycle; 2n is the number of cycles expected to be used in the model, and a n and b n These are parameters that need to be estimated; The holiday term h(t) is used to consider abnormal data fluctuations that occur at special values. Such fluctuations cannot be realized by the periodic model, so independent models are established for different special values. The specific form is: h(t)=Z(t)κ (20) Z(t)=[I(t∈D1),I(t∈D2),…,I(t∈D L )] (21) κ∈Normal(0,v 2 ) (22) Where D i is the i-th special value set; I(t∈D i ) is the indicator function, the value of the i-th special value is 1, otherwise it is 0; k is the special value setting parameter; v is the influence factor of the special value on the prediction result.

9. The data-knowledge driven multi-level fault early warning method for energy storage system according to claim 1, characterized in that: Based on the XGBoost algorithm, Prophet decomposition data is regressed. The XGBoost algorithm can be understood as a sum model of multiple decision trees. Assuming that the model has k decision trees, that is: Where k is the number of trees; f t is a function in the function space F; is the model prediction value; x i is the i-th data input; Represents sample x i The leaf node weight that falls to the corresponding leaf node; q(x i ) represents the sample x i Fall to the corresponding leaf node; The XGBoost algorithm uses the addition and forward distribution algorithm, and each iteration does not affect the original model, that is, The XGBoost algorithm objective function L consists of two parts, the first is the loss function and the second is the regularization term. In the formula, is the error between the predicted value and the true value; Ω(f j ) is a regular term to prevent the model from overfitting; T is the number of leaf nodes; γ is the penalty coefficient of T; w is the weight of the leaf node; n is the number of samples. When training to the t-th tree, the objective function is corrected to: Since the first t-1 regression trees are known information, that is, Ω(f j ) is a constant and has no effect on the optimization process of the tth tree, so the objective function can be simplified as: Rewrite the objective function into leaf node traversal: According to the Taylor series expansion of the binary function, It can be expressed as: Simplify the objective function and remove the constant term Right now In the formula, when hour, Get the minimum value, 10. A data-knowledge driven multi-level fault warning method for energy storage system according to claim 1, characterized in that: Based on the Prophet-XGBoost model, the lithium battery current, voltage, surface temperature, ambient temperature, and energy storage converter ambient temperature and output power prediction are realized.

11. A data-knowledge driven multi-level fault early warning method for energy storage system according to claim 1, characterized in that: Predict the temperature of key components (power devices, capacitors) of lithium batteries and energy storage converters based on the prediction information.

12. A data-knowledge driven multi-level fault early warning method for energy storage system according to claim 1, characterized in that: Multi-level fault warning of energy storage system is realized based on the predicted temperature of lithium battery and energy storage converter.

Images