A method and device for detecting faults in batteries of energy storage plants

Abstract

The application provides a kind of battery fault detection method and device of energy storage power station, it is related to battery fault detection technical field, including setting time window, the battery parameter of battery cluster and its inside corresponding monomer battery is sampled in time window, and pretreatment is carried out, constructs battery output state coefficient;Based on historical battery parameter, train LightGBM model, and input battery output state coefficient into the LightGBM model trained, obtain failure probability;Based on failure probability, realize over-threshold alarm using dynamic threshold adaptive algorithm.In time window, battery cluster and monomer battery are sampled, pretreatment is carried out to sampling point data, constructs output state coefficient, the output state coefficient of battery cluster and monomer battery output state coefficient are input into LightGBM model respectively, obtain battery cluster failure probability and monomer battery failure probability, based on the failure probability of battery cluster and monomer battery, realize dynamic threshold alarm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery fault detection methods and devices, in particular to a battery fault detection method and device based on an energy storage power station. BACKGROUND

[0002] Traditional energy storage power stations mainly include battery clusters and single battery cells, wherein each battery cluster includes multiple single battery cells. Since the energy storage power station has high power consumption and a long service life, it is prone to faults. Therefore, the state of the battery cluster and the single battery cell needs to be evaluated separately. Traditional battery fault detection methods and devices mainly use threshold detection and traditional machine learning, and gradually develop towards multi-parameter fusion, unsupervised learning and dynamic adaptive models, aiming to solve the problems of early detection, complex fault identification and real-time performance.

[0003] In the prior art, a method for determining the fault condition of a battery based on an unsupervised isolation forest algorithm is disclosed in CN117949828A. The battery SOC from a preset time to the current time is obtained and dimensionally divided to obtain the current SOC of each dimension. The current SOC is input into an isolation forest algorithm model to further determine whether the battery has failed. However, this method has low efficiency for multi-dimensional time series data segmentation. Battery data usually includes multiple parameters such as voltage, current, temperature, and internal resistance. Isolation forest has difficulty in effectively capturing the complex relationship between multiple variables during random feature segmentation, resulting in a decrease in abnormal detection accuracy and a high false positive rate. Therefore, battery data needs to be sampled in a time window to obtain battery parameter sampling point data, and the multi-variable data is sorted into a vector that can be input into the model, effectively avoiding the difficulty in distinguishing complex relationships between multiple variables. In addition, the battery cluster and the single battery cell should be detected using separate models to improve detection accuracy and reduce false positives.

[0004] The above information disclosed in the background section is only intended to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute the prior art known to those of ordinary skill in the art. SUMMARY

[0005] The present application aims to provide a battery fault detection method and device based on an energy storage power station to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] A battery fault detection method based on an energy storage power station, comprising the following specific steps:

[0008] S1: set a time window, sample and preprocess the battery parameters of the battery cluster and the corresponding single battery cells inside it in the time window, and construct a battery output state coefficient, which includes the output state coefficient of the battery cluster and the output state coefficient of the single battery cell;

[0009] S2: based on the historical battery parameters, train a LightGBM model, and input the battery output state coefficient into the trained LightGBM model to obtain the failure probability of the battery cluster and the failure probability of each single battery cell inside it;

[0010] S3: based on the failure probability of the battery cluster and the failure probability of the single battery cell, count the number of single battery cell failures, set an alarm threshold, and realize dynamic alarm of the battery cluster.

[0011] Further, the time window is [T0-T, T0], wherein T0 represents the timestamp of the current time, and T represents the length of the time window.

[0012] When collecting battery parameters, the battery parameters of the battery cluster are collected at a time interval of 10k, and the battery parameters of the single battery cell are collected at a time interval of k.

[0013] Further, the battery parameters of the single battery cell include single voltage, single current, single temperature, single battery internal resistance and single self-discharge rate, and the battery parameters of the battery cluster include total voltage, total current and temperature difference, wherein the temperature difference is the difference between the highest single temperature and the lowest single temperature of the single battery cells inside the battery cluster.

[0014] Further, the preprocessing step is: for the battery parameters of the battery cluster, set the window of the median filter to a+1, and perform median filter processing on the voltage, total current and temperature difference data of the battery cluster, and for the voltage, total current and temperature difference data of the battery cluster after median filter processing, respectively, according to the type of data, perform normalization processing;

[0015] For the battery parameters of the single battery cell, set the window of the median filter to 2a+1, and perform median filter processing on the single voltage, single current, single temperature, single battery internal resistance and single self-discharge rate, and for the single voltage, single current, single temperature, single battery internal resistance and single self-discharge rate after median filter processing, respectively, according to the type of data, perform normalization processing.

[0016] Further, the logic for constructing the battery cluster output state coefficient is:

[0017] Based on the preprocessed battery parameters of the battery cluster, a feature matrix of the battery cluster is constructed, and the specific formula is:

[0018] Further, the logic for constructing the battery cluster output state coefficient is:

[0019] wherein X represents the feature matrix of the battery cluster, V" c (j), I" c (j) and ΔT" c (j) represent the pre-processed total voltage, total current and temperature difference data of the jth sampling point, respectively, and j and N represent the number of sampling points and the total number of sampling points when collecting the battery parameters of the battery cluster, respectively;

[0020] Based on the pre-processed total voltage, total current and temperature difference data of each sampling point, the battery cluster state index of each sampling point is generated, and the output state coefficient of each battery cluster is constituted, and the formula is:

[0021]

[0022] wherein Xcl represents the output state coefficient of the battery cluster, x j represents the battery cluster state index of the jth sampling point;

[0023]

[0024] wherein T0 represents the set temperature difference threshold.

[0025] Further, the logic for constructing the output state coefficient of the single battery cell is:

[0026] Based on the pre-processed battery parameters of the single battery cell, the feature matrix of the single battery cell is constructed, and the specific formula is

[0027]

[0028] wherein Xce represents the feature matrix of the single battery cell; V" n (i) and I" n (i) represent the pre-processed single voltage value and single current value of the ith sampling point of the single battery cell, respectively; T" n (i), R" n (i) and S" n (i) represent the pre-processed single temperature, single internal resistance and single self-discharge rate of the ith sampling point of the single battery cell, respectively; i and M represent the number of sampling points and the total number of sampling points when collecting the battery parameters of the single battery cell, respectively;

[0029] Based on the pre-processed single voltage, single current, single temperature, single battery internal resistance and single self-discharge rate data of each sampling point, the cell state index of each sampling point is generated, and the output state coefficient of each single battery cell is constituted, and the formula is:

[0030]

[0031] wherein Y cell represents the output state coefficient of a monomer cell, y i represents the state index of the i-th sampling point, and:

[0032]

[0033] wherein C represents a monomer internal resistance adjustment constant; T op represents the optimal working temperature of the cell.

[0034] Further, the LightGBM model comprises a battery cluster model and a monomer cell model, the historical battery parameters comprise normal battery parameter samples and fault battery parameter samples, the normal battery parameter samples comprise output state coefficient samples of normal battery clusters and output state coefficient samples of normal monomer cells, and the fault battery parameter samples comprise output state coefficient samples of fault battery clusters and output state coefficient samples of fault monomer cells.

[0035] When training the LightGBM model, the output state coefficient samples of the battery clusters are taken as the feature input of the battery cluster model, and whether the battery cluster is faulty is taken as the feature output, and the battery cluster model is trained.

[0036] The output state coefficient samples of the monomer cells are taken as the feature input of the monomer cell model, and whether the monomer cell is faulty is taken as the feature output of the monomer cell model, and the monomer cell model is trained.

[0037] The output state coefficient of the battery cluster is input into the trained battery cluster model to obtain the fault probability of each battery cluster, and the output state coefficient of each monomer cell corresponding to the battery cluster is input into the trained monomer cell model to obtain the fault probability of each monomer cell.

[0038] Further, the logic for realizing the dynamic alarm of the battery cluster is:

[0039] According to the fault probability of the battery cluster and the fault probability of each monomer cell inside the battery cluster, the overall fault probability of the battery cluster is obtained, and the formula is:

[0040]

[0041] wherein P total represents the overall fault probability of the battery, represents the fault probability of the q-th monomer cell inside the battery cluster, n represents the total number of monomer cells inside the battery cluster, q represents the monomer cell index number, and p c represents the fault probability of the battery cluster.

[0042] The number of single battery cells with a failure probability greater than 0.5 is counted and marked as A, a dynamic alarm threshold is set, and the total battery failure probability and the alarm threshold are compared, if a battery cluster failure alarm signal is sent, and T represents the set threshold.

[0043] The present application also provides a battery failure detection device for an energy storage power station, which is used to perform the above detection method, comprising:

[0044] A state assessment module is configured to set a time window, sample battery parameters of the battery cluster and corresponding single battery cells within the time window, and preprocess the battery parameters to construct battery output state coefficients, including battery cluster output state coefficients and single battery cell output state coefficients.

[0045] A failure prediction module is configured to train a LightGBM model based on historical battery parameters, and input the battery output state coefficients into the trained LightGBM model to obtain the failure probability of the battery cluster and the failure probability of each single battery cell.

[0046] A failure warning module is configured to count the number of single battery cells based on the failure probability of the battery cluster and the failure probability of the single battery cells, set an alarm threshold, and realize dynamic alarm of the battery cluster.

[0047] Compared with the prior art, the present application has the following advantages:

[0048] The time window is introduced, the battery cluster and the single battery cell are sampled within the time window, the sampling point data is subjected to median filtering and normalization processing, noise interference is effectively eliminated, key failure features are retained, output state coefficients are constructed, the battery cluster output state coefficients and the single battery cell output state coefficients are input into the LightGBM model to obtain the battery cluster failure probability and the single battery cell failure probability, cross-level interference is avoided, the failure positioning accuracy is effectively improved, and dynamic threshold alarm is realized based on the failure probability of the battery cluster and the single battery cell. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 The present application is a whole method flowchart;

[0050] Figure 2 The present application is a whole device structure schematic diagram. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below with specific examples.

[0052] It should be noted that, unless otherwise defined, technical terms or scientific terms used in the present application shall be understood as their ordinary meaning to those having ordinary skill in the art to which the present application pertains. The terms "first", "second", and similar terms used in the present application do not denote any order, number, or importance, but are only used to distinguish different components. The terms "comprise", "include" and similar terms mean that the elements or objects before the term encompass the elements or objects listed after the term and their equivalents, without excluding other elements or objects. The terms "connected" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "up", "down", "left", "right", and the like only represent relative positional relationships, which can change accordingly when the absolute position of the described object changes.

[0053] Embodiments:

[0054] Please refer to Figure 1 The present application provides a technical solution:

[0055] A coal mine supervision method based on video images, the specific steps comprising:

[0056] S1: Set a time window, sample the battery parameters of the battery cluster and the corresponding single battery inside it within the time window, and pre-process, construct the battery output state coefficient, the battery output state coefficient includes the output state coefficient of the battery cluster and the output state coefficient of the single battery.

[0057] The time window is [T0-T, T0], wherein T0 represents the timestamp of the current time, T represents the length of the time window, and the battery parameters are collected at a time interval of 10k for the battery cluster and at a time interval of k for the single battery; wherein k∈N + .

[0058] The battery parameters of the single battery include single voltage, single current, single temperature, single battery internal resistance and single self-discharge rate, and the battery parameters of the battery cluster include total voltage, total current and temperature difference, the temperature difference is the difference between the highest single temperature and the lowest single temperature of the single battery inside the battery cluster.

[0059] Collecting the above battery parameters is because the battery parameters can comprehensively represent the running state of the battery, and have high sensitivity to fault characteristics. For single battery, single voltage, single current, single temperature, single battery internal resistance and single self-discharge rate can accurately locate the fault battery; for the battery cluster, the total voltage, total current and temperature difference can judge the system level fault risk; and the single battery parameters change rapidly, which requires high frequency sampling, while the battery cluster parameters change slowly, which requires low frequency sampling;

[0060] The preprocessing step is: for the battery cluster parameter, set the window of median filtering to a+1, median filter the voltage, total current and temperature difference data of the battery cluster, and normalize the median filtered voltage, total current and temperature difference data of the battery cluster according to the type of data respectively;

[0061] Since the sampling of the total voltage, total current and temperature difference of the battery cluster is low-frequency sampling, a smaller window of median filtering is selected;

[0062] The number of battery cluster parameter sampling points within the time window is denoted as N, N=T / 10k-1, the filtering window width is set to a+1, and the battery cluster parameter of each sampling point j is median filtered:

[0063] V′ ic (j)=median({V ic (j-a),V ic (j-a+1),…,V ic (j)});

[0064] I′ ic (j)=median({I ic (j-a),I ic (j-a+1),…,I ic (j)});

[0065] ΔT′ ic (j)=median({ΔT ic (j-a),ΔT ic (j-a+1),…,ΔT ic (j)});

[0066] Wherein, j∈[1,N];V′ ic (j) represents the median filtered total voltage value of the jth sampling point of the ith battery cluster; I′ ic (j) represents the median filtered total current value of the jth sampling point of the ith battery cluster; ΔT′ ic (j) represents the median filtered temperature difference value of the jth sampling point of the ith battery cluster; V ic (j) represents the original total voltage of the jth sampling point of the ith battery cluster; I ic (j) represents the original total current of the jth sampling point of the ith battery cluster; ΔT ic (j) represents the original temperature difference of the jth sampling point of the ith battery cluster;

[0067] The median filter is used to process the total voltage, total current and temperature difference of the battery cluster, which can filter out the jump of the cluster level parameters caused by external interference, such as sudden drop of total voltage and abnormal temperature difference, and the median filter only eliminates isolated noise points, and retains the real changes of continuous data; based on the filtered parameter sequence, the maximum and minimum values of each parameter in the time window are linearly normalized:

[0068]

[0069] m represents the sampling point index variable; V" ic (j) represents the normalized total voltage value of the jth sampling point of the ith battery cluster; I" ic (j) represents the normalized total current value of the jth sampling point of the ith battery cluster; ΔT" ic (j) represents the normalized temperature difference value of the jth sampling point of the ith battery cluster; V' ic (m) represents the filtered total voltage value of the ith battery cluster sampling point m in the time window; I' ic (m) represents the filtered total current value of the ith battery cluster sampling point m in the time window; ΔT' ic (m) represents the filtered temperature difference value of the ith battery cluster sampling point m in the time window;

[0070] The normalized filtered total voltage, total current and temperature difference are to adjust the dynamic range of the filtered data, so that the input data is closer to the real situation, and the false alarm rate of the battery cluster model is reduced;

[0071] For the battery parameters of the single cell, the window of the median filter is set to 2a+1, and the median filter is used to process the single cell voltage, single cell current, single cell temperature, single cell internal resistance and single cell self-discharge rate. For the single cell voltage, single cell current, single cell temperature, single cell internal resistance and single cell self-discharge rate processed by the median filter, the normalization processing is carried out according to the type of data respectively;

[0072] Since the sampling of the single cell voltage, single cell current, single cell temperature, single cell internal resistance and single cell self-discharge rate is high frequency sampling, a larger median filter window is selected; the number of single cell parameter sampling points in the time window is marked as M, M=T / k-1, the filter window width is set to 2a+1, and the median filter is used to process the single cell parameter of each sampling point l:

[0073] V′ n (i)=median({V(i-a),V(i-a+1),…,V(i+a)});

[0074] I′ n(i) = median({I(i-a), I(i-a+1),..., I(i+a)}) ;

[0075] T′ n (i) = median({T(i-a), T(i-a+1),..., T(i+a)}) ;

[0076] R′ n (i) = median({R(i-a), R(i-a+1),..., R(i+a)}) ;

[0077] S′ n (i) = median({S(i-a), S(i-a+1),..., S(i+a)}) ;

[0078] wherein, i ∈ [1, M]; V′ n (i) represents the median filtered voltage value of the i-th sampling point of the n-th single battery cell; I′ n (i) represents the median filtered current value of the i-th sampling point of the n-th single battery cell; T′ n (i) represents the median filtered temperature value of the i-th sampling point of the n-th single battery cell; R′ n (i) represents the median filtered internal resistance value of the i-th sampling point of the n-th single battery cell; S′ n (i) represents the median filtered self-discharge rate value of the i-th sampling point of the n-th single battery cell; V(i+a) represents the voltage value of the i+a-th sampling point of the single battery cell; I(i+a) represents the current value of the i+a-th sampling point of the single battery cell; T(i+a) represents the temperature value of the i+a-th sampling point of the single battery cell; R(i+a) represents the internal resistance value of the i+a-th sampling point of the single battery cell; S(i+a) represents the self-discharge rate value of the i+a-th sampling point of the single battery cell;

[0079] The median filtering is adopted to process the single voltage, single current, single temperature, single battery internal resistance and single self-discharge rate, so as to suppress high-frequency sampling noise, retain single battery cell fault characteristics, effectively eliminate random spikes of single voltage and single current, and avoid misjudgment of connection faults;

[0080] Based on the filtered parameter sequence, linear normalization is performed according to the maximum value and the minimum value of each parameter in the time window:

[0081]

[0082] V″ n (i) and I″ n (i) respectively represent the single voltage value and the single current value of the i-th sampling point of the single battery cell after preprocessing; T″n (i), R" n (i) and S" n (i) respectively represent the monomer battery cell i-th sampling point after pretreatment of monomer temperature value, monomer internal resistance value and monomer self-discharge rate value; i represents the number of sampling points and the total number of sampling points when collecting the battery parameters of the monomer battery cell; V' n (m) represents the filtered voltage value of the monomer battery cell sampling point m in the time window; I' n (m) represents the filtered current value of the monomer battery cell sampling point m in the time window; T' n (m) represents the filtered temperature value of the monomer battery cell sampling point m in the time window; R' n (m) represents the filtered internal resistance value of the monomer battery cell sampling point m in the time window; S' n (m) represents the filtered self-discharge rate value of the monomer battery cell sampling point m in the time window;

[0083] The normalized monomer voltage, monomer current, monomer temperature, monomer battery internal resistance and monomer self-discharge rate are to adjust the dynamic range of the filtered data, so that the input data is closer to the real situation, and the false positive rate of the monomer battery cell model is reduced;

[0084] S2: based on the historical battery parameters, training LightGBM model, and inputting the battery output state coefficient into the trained LightGBM model, obtaining the fault probability of the battery cluster and the fault probability of each monomer battery cell inside;

[0085] The logic for constructing the battery cluster output state coefficient is:

[0086] Based on the pretreated battery parameters of the battery cluster, the feature matrix of the battery cluster is constructed, and the specific formula is:

[0087]

[0088] Wherein, X represents the feature matrix of the battery cluster, V" c (j), I" c (j) and ΔT" c (j) respectively represent the total voltage, total current and temperature difference data of the j-th sampling point after pretreatment, and j and N respectively represent the number of sampling points and the total number of sampling points when collecting the battery parameters of the battery cluster;

[0089] The feature matrix of the battery cluster is constructed in the form of sampling points, which effectively preserves the continuity of the original data and provides a basis for the construction of the battery cluster state index;

[0090] The total voltage, total current and temperature difference data of each sampling point after preprocessing are processed to generate a battery cluster state index of each sampling point, and an output state coefficient of each battery cluster is formed, and the formula is:

[0091]

[0092] Wherein, Xcl represents the output state coefficient of the battery cluster, x j represents the battery cluster state index of the jth sampling point.

[0093]

[0094] The performance of the battery can be measured by power, which is defined as the product of voltage and current. This product can directly reflect the working state of the battery at a certain time. The changes of voltage and current can indicate the health status of the battery. Under normal circumstances, the voltage and current of the battery should be maintained within a certain range, and when a fault occurs, they may decrease or fluctuate. Therefore, combining them as a product can more comprehensively reflect the overall state of the battery. The total voltage V c (j) and the product of the total current I c (j) reflect the power output capability of the battery cluster.

[0095] The temperature difference reflects the temperature uniformity inside the battery cluster. A larger temperature difference means that some battery cells have faults or uneven discharge. In the actual operation of the battery, the temperature exceeding a certain threshold will cause the battery performance to decrease significantly or increase the risk of failure. By setting a temperature difference threshold, the influence of temperature on the state of the battery can be clearly defined, which is convenient for understanding and use. The exponential function form is used to introduce the nonlinear influence of temperature difference on the state index. The change of temperature has a nonlinear influence on the performance of the battery. The use of exponential function can effectively capture this nonlinear characteristic. T0 in this part is a temperature difference threshold, which is used to distinguish the degree of influence of temperature difference on the state index. When the temperature difference is greater than T0, the state index will rise rapidly, indicating that the battery may be in a higher risk state, which can better reflect the rapid and significant influence of excessively high or low temperature on the performance of the battery in the actual situation.

[0096] The form of the denominator is which can effectively prevent the denominator from being zero. Since the value of is always non-negative (i.e. greater than or equal to 0), the denominator is always greater than or equal to 1, thereby avoiding uncertainty and potential mathematical errors in calculation, ensuring that when the temperature difference is 0 or in a lower state, the denominator will not be too small, thereby keeping the numerical range of the state index within a reasonable level. Even in the case of small temperature difference, the state index can remain at a stable base value.

[0097] When the temperature difference is relatively low compared to the temperature difference threshold, The value of the state index will be close to 1, and the state index will be close to The change of the state index is gradual rather than abrupt when the temperature difference is close to the threshold, which reflects the response characteristics of the battery when the temperature difference changes, making the state assessment more smooth and avoiding unnecessary sharp fluctuations. When the temperature difference is relatively high compared to the temperature difference threshold, the value of the battery cluster state index decreases rapidly, resulting in a significant increase in the battery cluster state index, reflecting the significant impact of temperature difference on the state of the battery cluster, which can reflect the significant impact of temperature difference on the performance and health status of the battery in the real world.

[0098] The logic for constructing the output state coefficient of the single cell is as follows:

[0099] Based on the pre-processed battery parameters of the single cell, a feature matrix of the single cell is constructed, and the specific formula is

[0100]

[0101] Where Xce represents the feature matrix of the single cell; M represents the total number of sampling points when collecting the battery parameters of the single cell; V" n (i) and I" n (i) represent the pre-processed single cell voltage and single cell current of the i-th sampling point; T" n (i), R" n (i) and S" n (i) represent the pre-processed single cell temperature, single cell internal resistance and single cell self-discharge rate of the i-th sampling point; i and M represent the number of sampling points and the total number of sampling points when collecting the battery parameters of the single cell;

[0102] The feature matrix of the single cell is constructed in the form of sampling points, effectively preserving the continuity of the original data, providing a basis for the construction of the cell state index later;

[0103] Based on the pre-processed single cell voltage, single cell current, single cell temperature, single cell internal resistance and single cell self-discharge rate data of each sampling point, the cell state index of each sampling point is generated, and the output state coefficient of each single cell is constructed. The formula is as follows:

[0104]

[0105] Where Y cell represents the output state coefficient of the single cell, y i represents the cell state index of the i-th sampling point, and:

[0106]

[0107] g[V″ n (i)] represents a temperature influence function, C represents a constant, T op represents the optimal working temperature of the battery cell;

[0108] The product of the monomer voltage and the monomer current V″ n (i)*I″ n (i) represents the output power of the monomer battery cell, the power output of the monomer battery cell is a direct indicator of its performance. The power can reflect the output capacity of the battery at a certain moment, under normal circumstances, the monomer battery should be able to output a certain monomer voltage and monomer current, if these two values decrease significantly, it may mean that the battery has a fault, therefore, the product of the monomer voltage and the monomer current can effectively indicate the health status of the battery cell, the monomer internal resistance is an important indicator of battery health, the larger the internal resistance, the greater the power loss, and the more heat generated, affecting the efficiency and safety of the battery, putting the monomer internal resistance into the denominator means that the increase of the monomer internal resistance will reduce the state index, reflecting the trend of the battery performance decline, in order to avoid division by zero error, a small constant C is added, to ensure that even when the monomer internal resistance is zero, the formula can still be calculated.

[0109] The monomer self-discharge rate represents the speed at which the battery loses electricity without load, a too high monomer self-discharge rate means that the battery performance is poor, it is part of the denominator, indicating that the increase of the monomer self-discharge rate will lead to the decrease of the state index, reflecting the decline of the effective utilization rate of the battery, 1+S″ n (i) in the form of adding a constant 1 ensures that the denominator is always greater than or equal to 1 regardless of the value of the self-discharge rate, which can prevent the formula from collapsing in extreme cases, providing mathematical stability, using 1+S″ n (i) ensures that the influence of the monomer self-discharge rate is non-negative, even when the monomer self-discharge rate is 0, the state index can still reflect the performance of the battery, when the monomer self-discharge rate increases, S″ n (i) will increase, thus reducing the state index, reflecting the inhibitory effect of the monomer self-discharge rate on the performance of the battery cell, which is in line with the actual situation.

[0110] The state of the monomer battery cell is not determined by a single factor, the monomer internal resistance and the monomer self-discharge rate jointly affect the effective energy output of the battery cell, the multiplication form effectively reflects the linkage between the two, even if the monomer internal resistance is small or the monomer self-discharge rate is low, the influence of the other factor will be amplified or reduced accordingly, the monomer internal resistance and the monomer self-discharge rate can be regarded as two aspects of the effective energy loss of the battery, the monomer internal resistance reflects the loss in use, while the monomer self-discharge rate reflects the loss in the static state, by combining the two, a composite energy loss indicator is formed, providing a comprehensive battery health status evaluation.

[0111] The performance of the battery is significantly affected by the cell temperature, and excessively high or low temperature can accelerate the aging of the battery and reduce the efficiency, the cell temperature influence function selects an exponential function form, T op represents the optimal working temperature of the battery cell, at which the performance of the battery cell is optimal, the temperature influence function has central symmetry, meaning that when the cell temperature is in the optimal range, the function value is close to 1, and when the cell temperature deviates slightly, the function value decreases rapidly, reflecting the nonlinear influence of the cell temperature on the performance of the battery.

[0112] The cell temperature influence function is essentially a Gaussian function, which has good smoothness and continuity, and the influence of temperature on the performance of the battery is nonlinear and changes with temperature, and the Gaussian function can naturally capture this change, and the Gaussian function is symmetric about its center, that is, it performs best near the optimal working temperature, and the function value decreases rapidly when it deviates from the temperature, which is consistent with the physical characteristics of the battery cell performing best near the optimal working temperature.

[0113] By [T″ n (i)-T op ] 2 It emphasizes that the influence of temperature deviation from the optimal value is quadratic, that is, when the cell temperature deviates slightly, the influence is small, and when the cell temperature deviates greatly, the influence increases rapidly, so that the influence of the cell temperature on the performance of the battery is gradually increasing, and the square term makes the influence of the cell temperature deviation present nonlinear characteristics, that is, the greater the deviation, the more significant the influence, which can effectively reflect the sensitivity of the performance of the battery to the change of the cell temperature in reality, for example, at extreme temperatures, the performance of the battery may decrease sharply, and the square term can well reflect this phenomenon, when the cell temperature is too low, the chemical reaction rate of the battery decreases, resulting in a decrease in power output, and when the cell temperature is too high, it may cause the battery to overheat, accelerate aging or damage. Therefore, the influence of the cell temperature on the performance of the battery presents similar Gaussian characteristics.

[0114] The LightGBM model includes a battery cluster model and a cell model, and the historical battery parameters include normal battery parameter samples and fault battery parameter samples, the normal battery parameter samples include output state coefficient samples of normal battery clusters and output state coefficient samples of normal cell, and the fault battery parameter samples include output state coefficient samples of fault battery clusters and output state coefficient samples of fault cell;

[0115] When training the LightGBM model, the output state coefficient samples of the battery cluster are input as the feature of the battery cluster model, and whether the battery cluster is faulty is taken as the feature output, and the battery cluster model is trained.

[0116] The output state coefficient sample of the single battery cell is input as a feature of the single battery cell model, and whether the single battery cell is faulty is output as a feature of the single battery cell model, and the single battery cell model is trained;

[0117] The output state coefficient of the battery cluster is input into the trained battery cluster model to obtain the fault probability of each battery cluster, and the output state coefficient of each single battery cell corresponding to the battery cluster is input into the trained single battery cell model to obtain the fault probability of each single battery cell.

[0118] S3: Based on the fault probability of the battery cluster and the fault probability of the single battery cell, the number of faults of the single battery cell is counted, the alarm threshold is set, and the dynamic alarm of the battery cluster is realized;

[0119] The logic for realizing the dynamic alarm of the battery cluster is:

[0120] According to the fault probability of the battery cluster and the fault probability of each single battery cell inside the battery cluster, the overall fault probability of the battery cluster is obtained, and the formula is:

[0121]

[0122] P total represents the overall fault probability of the battery, represents the fault probability of the qth single battery cell inside the battery cluster, n represents the total number of single battery cells inside the battery cluster, q represents the index number of the single battery cell, and p c represents the fault probability of the battery cluster;

[0123] The number of single battery cells with a fault probability greater than 0.5 is counted and marked as A, the alarm threshold is dynamically set, and the overall fault probability of the battery is compared with the alarm threshold, if a battery cluster fault alarm signal is sent, and T represents the set threshold;

[0124] P total The overall fault probability of the battery is generated by combining the fault probability of the battery cluster and the fault probability of each single battery cell, which can comprehensively reflect the overall fault probability of a single battery cluster. By considering the relatively independent fault probability of each single battery cell, the probability that the single battery cell does not fail is calculated in the form of product, and then the overall fault probability is calculated by subtracting the probability that the battery does not fail from 1. In order to realize dynamic threshold alarm, the number of single battery cells with a fault probability greater than 0.5 is counted, a trigonometric function is introduced based on the known number of single battery cells, the value of the trigonometric function varies between 0 and 1, the threshold is set according to expert experience or historical fault data, and the two are expressed in the form of product to realize dynamic threshold alarm.

[0125] Please refer to Figure 2The application further provides a fault detection device of a battery of an energy storage power station, the detection device being used for executing the detection method, and comprising:

[0126] a state evaluation module, configured to set a time window, sample and pre-process battery parameters of the battery cluster and corresponding single battery cells in the battery cluster in the time window, and construct battery output state coefficients, the battery output state coefficients comprising output state coefficients of the battery cluster and output state coefficients of the single battery cells;

[0127] a fault prediction module, configured to train a LightGBM model based on historical battery parameters, and input the battery output state coefficients into the trained LightGBM model to obtain fault probabilities of the battery cluster and the single battery cells;

[0128] a fault early warning module, configured to count fault numbers of the single battery cells based on the fault probabilities of the battery cluster and the single battery cells, set an alarm threshold, and realize dynamic alarm of the battery cluster.

[0129] The above formulas are all dimensionless values, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the nearest real situation, and preset parameters in the formula are set by a person skilled in the art according to actual conditions.

[0130] The above embodiments can be realized wholly or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized in the form of a computer program product wholly or partially. Those skilled in the art can realize that units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software methods depends on specific application and design constraints of the technical solutions.

[0131] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, and can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0132] The above is merely specific implementation manners of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for fault detection of batteries in an energy storage power station, characterized in that, include: S1: Set a time window, sample the battery parameters of the battery cluster and its corresponding individual cells within the time window, perform preprocessing, and construct the battery output state coefficient, which includes the output state coefficient of the battery cluster and the output state coefficient of the individual cells. S2: Based on historical battery parameters, train the LightGBM model and input the battery output state coefficient into the trained LightGBM model to obtain the failure probability of the battery cluster and the failure probability of each individual cell inside. S3: Based on the failure probability of battery clusters and individual cells, count the number of failures in individual cells, set alarm thresholds, and realize dynamic alarm for battery clusters. The overall failure probability of the battery cluster is obtained based on the failure probability of the battery cluster and the failure probability of each individual cell within the battery cluster. The formula used is as follows: Among them, P total Indicates the overall probability of battery failure. p represents the failure probability of the q-th individual cell within the battery cluster, where n represents the total number of individual cells in the battery cluster; q represents the index number of the individual cell; c Indicates the probability of battery cluster failure; The number of individual battery cells with a failure probability greater than 0.5 is counted and labeled as A. An alarm threshold is dynamically set, and the overall battery failure probability is compared with the alarm threshold. If... Then a battery cluster fault alarm signal will be issued, where T represents the set threshold. The time window is [T0-T, T0], where T0 represents the timestamp of the current moment and T represents the length of the time window; When collecting battery parameters, the battery parameters of the battery cluster are collected at 10k time intervals, and the battery parameters of the individual cell are collected at k time intervals. The battery parameters of a single cell include single cell voltage, single cell current, single cell temperature, single cell internal resistance, and single cell self-discharge rate. The battery parameters of a battery cluster include total voltage, total current, and temperature difference, wherein the temperature difference is the difference between the highest and lowest single cell temperatures within the battery cluster. The logic for constructing the output state coefficients of the battery cluster is as follows: Based on the preprocessed battery parameters of the battery cluster, the feature matrix of the battery cluster is constructed, and the specific formula is as follows: Where X represents the feature matrix of the battery cluster, V″ c (j), I″ c (j) and ΔT″ c (j) represent the total voltage, total current and temperature difference data of the j-th sampling point after preprocessing, respectively, and j and N represent the sampling point number and the total number of sampling points when collecting battery parameters of the battery cluster, respectively; Based on the preprocessed total voltage, total current, and temperature difference data of each sampling point, a battery cluster state index is generated for each sampling point, constituting the output state coefficient of each battery cluster. The formula used is as follows: Where Xcl represents the output state coefficient of the battery cluster, x j This represents the battery cluster state index at the j-th sampling point; Where T0 represents the set temperature difference threshold.

2. The method for fault detection of batteries in an energy storage power station according to claim 1, characterized in that, The preprocessing steps are as follows: for the battery parameters of the battery cluster, the median filtering window is set to a+1, the voltage, total current and temperature difference data of the battery cluster are processed by median filtering, and the voltage, total current and temperature difference data of the battery cluster after median filtering are normalized according to the data type. For the battery parameters of a single cell, the median filter window is set to 2a+1. Median filtering is performed on the cell voltage, cell current, cell temperature, cell internal resistance, and cell self-discharge rate. The cell voltage, cell current, cell temperature, cell internal resistance, and cell self-discharge rate after median filtering are then normalized according to the data type.

3. The method for fault detection of batteries in an energy storage power station according to claim 2, characterized in that, The logic for constructing the output state coefficient of a single battery cell is as follows: Based on the preprocessed battery parameters of the individual cells, the feature matrix of the individual cells is constructed, and the specific formula is as follows: Where Xce represents the feature matrix of a single battery cell; V″ n (i) and I″ n (i) represent the preprocessed voltage and current values ​​of a single cell at the i-th sampling point; T″ n (i), R″ n (i) and S″ n (i) represent the cell temperature, cell internal resistance, and cell self-discharge rate of the i-th sampling point after preprocessing, respectively; i and M represent the sampling point number and the total number of sampling points when collecting the battery parameters of the cell, respectively. Based on the preprocessed data of cell voltage, cell current, cell temperature, cell internal resistance, and cell self-discharge rate at each sampling point, a cell state index is generated for each sampling point, constituting the output state coefficient of each cell. The formula used is as follows: Among them, Y cell y represents the output state coefficient of a single battery cell. i Let represent the cell state index at the i-th sampling point, and: Where C represents the monomer internal resistance adjustment constant; T op This indicates the optimal operating temperature of the battery cell.

4. The method for fault detection of batteries in an energy storage power station according to claim 1, characterized in that, The LightGBM model includes a battery cluster model and a single cell model. The historical battery parameters include normal battery parameter samples and faulty battery parameter samples. The normal battery parameter samples include output state coefficient samples of normal battery clusters and output state coefficient samples of normal single cells. The faulty battery parameter samples include output state coefficient samples of faulty battery clusters and output state coefficient samples of faulty single cells. When training the LightGBM model, the output state coefficient samples of the battery cluster are used as the feature input of the battery cluster model, and whether the battery cluster is faulty is used as the feature output to train the battery cluster model. The output state coefficient samples of a single battery cell are used as the feature input of the single battery cell model, and whether the single battery cell is faulty is used as the feature output of the single battery cell model. The single battery cell model is then trained. The output state coefficients of the battery clusters are input into the trained battery cluster model to obtain the failure probability of each battery cluster. The output state coefficients of each individual cell in the battery cluster are input into the trained individual cell model to obtain the failure probability of each individual cell.

5. A fault detection device for batteries in an energy storage power station, characterized in that: The detection device is used to perform the detection method according to any one of claims 1-4, including: The state assessment module is used to set a time window, sample the battery parameters of the battery cluster and its corresponding individual cells within the time window, perform preprocessing, and construct the battery output state coefficient. The battery output state coefficient includes the output state coefficient of the battery cluster and the output state coefficient of the individual cells. The fault prediction module trains the LightGBM model based on historical battery parameters and inputs the battery output state coefficients into the trained LightGBM model to obtain the fault probability of the battery cluster and the fault probability of each individual cell within it. The fault warning module is used to count the number of faults in individual cells based on the fault probability of the battery cluster and the fault probability of individual cells, set alarm thresholds, and realize dynamic alarm of the battery cluster.

Images