Direct current arc detection method and device for energy storage system and energy storage system

Abstract

This application provides a method, apparatus, and energy storage system for detecting DC arcing in an energy storage system, relating to the field of energy storage system technology. The method includes: acquiring the current electrical parameters and current operating condition parameters of a target battery in the energy storage system; determining the confidence level of an arcing fault in the target battery based on the current electrical parameters and a detection module; determining the current confidence threshold corresponding to the target battery under the current operating condition parameters based on the current operating condition parameters and a threshold update module; the threshold update module is used to substitute the current operating condition parameters into a preset polynomial function to obtain a confidence threshold correction amount, and calculate the sum of the confidence threshold correction amount and a preset initial confidence threshold to obtain the current confidence threshold; inputting the confidence level and the current confidence threshold into a classification module to obtain the current arcing detection result of the target battery. This application can accurately detect DC arcing faults in energy storage systems, improving the accuracy and reliability of arcing detection.

Description

Technical Field

[0001] This application relates to the field of energy storage system technology, and in particular to a DC arc detection method, device and energy storage system for energy storage systems. Background Technology

[0002] As energy storage systems develop towards larger capacities and higher voltages, the risk of fires caused by DC arcing has increased significantly. Arcing is a continuous discharge phenomenon caused by the ionization of gas between charged conductors, with energy reaching thousands to tens of thousands of joules. It can easily lead to thermal runaway of batteries in energy storage systems. Therefore, accurate detection of DC arcing in energy storage systems is necessary.

[0003] In related technologies, DC arcing detection typically includes feature threshold detection and model detection methods. Feature threshold detection methods sometimes involve spectral analysis of the battery's voltage and current to extract fault frequency components and compare them with corresponding thresholds to determine if arcing has occurred. Others analyze the waveforms of the battery's voltage and current to extract features of arcing faults, such as high-frequency oscillations and pulses, and compare these with corresponding thresholds to determine if arcing has occurred. These feature analysis methods usually rely on pre-defined thresholds, limiting their detection accuracy. Model detection methods, on the other hand, typically extract features from the acquired signals and combine them with machine learning models for arcing fault detection. By learning the differences between the characteristics of arcing and normal operation, machine learning models can effectively detect arcing faults in batteries.

[0004] However, the inventors of this application have discovered that although a large amount of training data is used to train the machine learning model, the training data may not be able to cover all operating conditions of the battery, resulting in false alarms and false negatives in the trained machine learning model, which affects the accuracy and reliability of arc detection. Summary of the Invention

[0005] This application provides a method, apparatus, and energy storage system for detecting DC arcing faults in an energy storage system, so as to accurately detect DC arcing faults in the energy storage system, reduce false alarms and missed alarms of arcing faults, and improve the accuracy and reliability of arcing detection.

[0006] In a first aspect, embodiments of this application provide a DC arcing detection method for an energy storage system, the method comprising: The current electrical parameters and current operating condition parameters of the target battery in the energy storage system are obtained; wherein, the current operating condition parameters include the current state of charge, current load rate and current ambient temperature of the target battery. Based on the current electrical parameters and the detection module, the confidence level of the target battery experiencing an arcing fault is determined; wherein, the detection module is a module in the arcing detection model, and the arcing detection model further includes a threshold update module and a classification module, and the detection module is trained based on different categories of electrical parameters and corresponding arcing categories, the arcing categories including arcing fault occurrence and no arcing fault occurrence; Based on the current operating condition parameters and the threshold update module, the current confidence threshold corresponding to the target battery under the current operating condition parameters is determined; wherein, the threshold update module is used to substitute the current operating condition parameters into a preset polynomial function to obtain a confidence threshold correction amount, and calculate the sum of the confidence threshold correction amount and the preset initial confidence threshold to obtain the current confidence threshold. The confidence level and the current confidence threshold are input into the classification module to obtain the current arc detection result of the target battery.

[0007] In one possible implementation, the method further includes: Obtain the first historical electrical parameters and sample operating condition parameters of the first sample battery under various operating conditions; For each operating condition, based on the first historical operating parameters under that operating condition and the detection module, the sample confidence level of the first sample battery experiencing arcing failure under that operating condition is determined; based on the sample confidence level of the first sample battery experiencing arcing failure under that operating condition, the corresponding confidence threshold correction amount is determined; Based on the sample operating condition parameters for each operating condition and the corresponding confidence threshold correction, a polynomial function is established between the operating condition parameters and the confidence threshold correction.

[0008] In one possible implementation, determining the confidence threshold correction amount corresponding to the operating condition based on the sample confidence level of the first sample battery experiencing arcing failure under the operating condition includes: Based on the sample confidence level of the first sample battery experiencing arcing failure under this operating condition, the confidence distribution of the first sample battery experiencing arcing failure under this operating condition is obtained. The confidence threshold correction amount corresponding to the operating condition is determined based on the sample confidence level corresponding to the preset quantile in the confidence distribution.

[0009] In one possible implementation, the current electrical parameters include current current data and current voltage data; The step of determining the confidence level of the target battery experiencing an arcing fault based on the current electrical parameters and the detection module includes: Extract time-frequency domain features from the current current data and the current voltage data; The time-frequency domain features are reconstructed to obtain fused features; wherein the dimension of the fused features is smaller than the dimension of the time-frequency domain features. The fused features are input to the detection module to obtain the confidence level of the target battery having an arcing fault, output by the detection module; wherein, the detection module is used to map the fused features to a preset interval to obtain the confidence level.

[0010] In one possible implementation, the time-frequency domain features are reconstructed to obtain fused features, including: Calculate the covariance between the eigenvalues ​​of every two dimensions in the time-frequency domain features; Based on all the covariances of the time-frequency domain features, the covariance matrix is ​​obtained; The covariance matrix is ​​subjected to eigenvalue decomposition to obtain multiple eigenvalues ​​and the eigenvector corresponding to each eigenvalue; A preset number of feature vectors are selected in descending order of their feature values ​​to obtain a feature vector matrix; wherein the preset number is less than the dimension of the time-frequency domain features. Based on the time-frequency domain features and the feature vector matrix, the fusion features are determined.

[0011] In one possible implementation, determining the fused features based on the time-frequency domain features and the feature vector matrix includes: The time-frequency domain features are projected onto the space constructed by the feature vector matrix to obtain initial dimensionality-reduced features; wherein the preset number is greater than the dimension of the fused features; The fusion features are determined based on the preset projection matrix and the initial dimensionality reduction features; wherein the projection matrix is ​​calculated based on the intra-class scatter matrix and the inter-class scatter matrix, and the intra-class scatter matrix and the inter-class scatter matrix are calculated based on the initial dimensionality reduction features of different categories.

[0012] In one possible implementation, the method further includes: Obtain multiple sets of second historical electrical parameters of the second sample battery and the arcing category of each set of second historical electrical parameters; wherein, the second historical electrical parameters include historical current data and historical voltage data; For each set of second historical electrical parameters, historical time-frequency domain features are extracted from the historical current data and the historical voltage data; based on the historical time-frequency domain features, the initial historical dimensionality reduction features are determined. Based on the historical initial dimensionality reduction features of each arcing category, calculate the intra-class mean of each arcing category and the overall mean of the second sample batteries; Based on the historical initial dimensionality reduction features of each arc category and the intra-class mean of its corresponding arc category, the internal divergence matrix of each arc category is determined; The sum of the internal scatter matrices for each arc category is calculated to obtain the intra-class scatter matrix; The inter-class scatter matrix is ​​determined based on the difference between the intra-class mean and the overall mean for each arc category. The projection matrix is ​​determined based on the intra-class scatter matrix and the inter-class scatter matrix.

[0013] In one possible implementation, acquiring the current electrical parameters and current operating condition parameters of the target battery in the energy storage system includes: Obtain the current electrical parameters and current operating condition parameters of the target battery in the energy storage system within the current time period; After inputting the confidence level and the current confidence threshold into the classification module to obtain the current arc detection result of the target battery in the current time period, the method further includes: Obtain a preset number of historical arc detection results consecutively before the current time period; If both the historical arcing detection results and the current arcing detection results indicate that an arcing fault has occurred, then it is determined that the target battery has experienced an arcing fault; otherwise, it is determined that the target battery has not experienced an arcing fault.

[0014] Secondly, embodiments of this application provide a DC arcing detection device for an energy storage system, comprising: The acquisition module is used to acquire the current electrical parameters and current operating condition parameters of the target battery in the energy storage system; wherein, the current operating condition parameters include the current state of charge, current load rate and current ambient temperature of the target battery. The processing module is used to determine the confidence level of the target battery having an arcing fault based on the current electrical parameters and the detection module; wherein, the detection module is a module in the arcing detection model, and the arcing detection model further includes a threshold update module and a classification module. The detection module is trained based on different categories of electrical parameters and corresponding arcing categories, and the arcing categories include arcing fault occurrence and no arcing fault occurrence. The threshold module is used to determine the current confidence threshold corresponding to the target battery under the current operating conditions based on the current operating condition parameters and the threshold update module; wherein, the threshold update module is used to substitute the current operating condition parameters into a preset polynomial function to obtain a confidence threshold correction amount, and calculate the sum of the confidence threshold correction amount and the preset initial confidence threshold to obtain the current confidence threshold. The determination module is used to input the confidence level and the current confidence threshold into the classification module to obtain the current arc detection result of the target battery.

[0015] Thirdly, embodiments of this application provide an energy storage system including a controller and a plurality of batteries, wherein the controller is used to implement the method in the first aspect or any possible implementation thereof.

[0016] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described in the first aspect or any possible implementation thereof.

[0017] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect or any possible implementation of the first aspect.

[0018] The beneficial effects of the embodiments in this application compared with the prior art are: This application embodiment acquires the current electrical parameters and current operating condition parameters of the target battery in the energy storage system. The current operating condition parameters may include the target battery's current state of charge, current load rate, and current ambient temperature. Using the current electrical parameters and a detection module, the confidence level of an arcing fault in the target battery can be accurately determined. The detection module is a module within an arcing detection model, which also includes a threshold update module and a classification module. Then, using the current operating condition parameters and the threshold update module, the current confidence threshold corresponding to the target battery under the current operating condition parameters is determined. The threshold update module substitutes the current operating condition parameters into a preset polynomial function to obtain a confidence threshold correction amount, and calculates the sum of the confidence threshold correction amount and a preset initial confidence threshold to obtain the current confidence threshold. This fully considers the changes in the target battery's operating conditions and accurately obtains a suitable confidence threshold for the target battery. Finally, this confidence level is compared with the current confidence threshold by the classification module, which can accurately determine the current arcing detection result of the target battery, reducing false alarms and false negatives of arcing faults and improving the accuracy and reliability of arcing fault detection. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is an application scenario diagram of the DC arc detection method for energy storage systems provided in the embodiments of this application; Figure 2 This is a flowchart illustrating the implementation of the DC arc detection method for an energy storage system provided in this application embodiment; Figure 3 This is a schematic diagram of the arc detection model provided in the embodiments of this application; Figure 4 This is a schematic diagram of the effective components of the current frequency domain harmonics of a battery experiencing an arcing fault, provided in an embodiment of this application. Figure 5 This is a schematic diagram of the effective components of the current frequency domain harmonics of a battery that has not experienced arcing faults, as provided in the embodiments of this application. Figure 6 This is a schematic diagram of the structure of the DC arc detection device for the energy storage system provided in the embodiments of this application. Detailed Implementation

[0021] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0022] The inventors of this application have discovered that when using a machine learning model for DC arcing detection, the training data of the machine learning model may be insufficient to cover all operating conditions of the battery, resulting in false alarms and missed alarms in the trained machine learning model, which affects the accuracy and reliability of DC arcing detection.

[0023] In order to improve the accuracy of DC arcing detection, this application embodiment considers that the machine learning model outputs the confidence level of the result along with the output result. By using the operating conditions of the battery in the energy storage system, a confidence threshold suitable for the battery is determined. Then, by using the obtained confidence level and the confidence threshold, the arcing fault of the battery is finally determined.

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0025] Figure 1 This diagram illustrates an application scenario of the DC arcing detection method for an energy storage system provided in this embodiment. Figure 1As shown, the energy storage system includes multiple battery groups, such as a battery cluster. Each battery group is equipped with a current sampling device, which samples the current of the corresponding battery to obtain its current data. The current sampling device can employ sensors, such as tunneling magnetoresistance (TMR) sensors. The high sensitivity (μA-level detection) and wide bandwidth (0-1MHz) of TMR sensors allow for precise capture of high-frequency harmonics in the current.

[0026] All current sampling devices are connected to the energy storage system controller via a multiplexer. By adjusting the multiplexer, the current data of the selected target battery can be transmitted to the controller for subsequent DC arcing detection. A filter, such as a Butterworth bandpass filter, can also be placed between the multiplexer and the controller to filter the sampled current data and reduce noise in the signal.

[0027] In addition to current data, DC arcing detection can also utilize battery voltage data. The controller can also acquire the voltage data of the target battery, and by combining the acquired current and voltage data, DC arcing detection of the target battery can be performed.

[0028] The aforementioned controller can be implemented using a Digital Signal Processor (DSP). Furthermore, the controller can be implemented using a single DSP, such as the main control DSP in the energy storage system, which receives current and voltage data and performs DC arc detection. Alternatively, the controller can be implemented using two DSPs, such as the main control DSP in the energy storage system handling voltage data reception and DC arc detection, while another DSP receives the target battery's current data, processes it, and then transmits it to the aforementioned main control DSP for subsequent DC arc detection.

[0029] Here, by setting a multi-channel selection switch, the target battery under test and its current data can be switched simply by adjusting the multi-channel selection switch, thereby performing DC arcing detection on all batteries in the energy storage system. There is no need to set up corresponding devices such as arcing detection DSP for each group of batteries, thus reducing the detection cost.

[0030] In addition, the controller can connect to a Human Machine Interface (HMI). After determining the arcing detection result, the controller can send the result to the HMI for relevant personnel to view. Furthermore, it can also send relevant current and voltage data to the HMI.

[0031] See Figure 2The document illustrates a flowchart of the DC arc detection method for an energy storage system provided in this embodiment, which is described in detail below: Step 201: Obtain the current electrical parameters and current operating condition parameters of the target battery in the energy storage system; wherein, the current operating condition parameters include the current state of charge, current load rate and current ambient temperature of the target battery.

[0032] In this embodiment, the current electrical parameters can be data such as the voltage and current of the target battery in the energy storage system, including high-frequency timing signals such as the current current data and current voltage data of the target battery. The current current data can be acquired by a TMR sensor, and the current voltage data can be acquired by an analog-to-digital converter and transmitted to the main control DSP within the energy storage system.

[0033] The current operating parameters are the state of charge, load rate, and ambient temperature of the target battery in the energy storage system, which are slow-changing state variables of the battery.

[0034] Step 202: Determine the confidence level of the target battery having an arcing fault based on the current electrical parameters and the detection module; wherein, the detection module is a module in the arcing detection model, which also includes a threshold update module and a classification module. The detection module is trained based on different categories of electrical parameters and corresponding arcing categories, including arcing fault occurrence and no arcing fault occurrence.

[0035] In this embodiment, the detection module in the arcing detection model can output the degree of certainty of its generated results, i.e., the confidence level, when detecting and classifying the current electrical parameters. Here, the confidence level of the target battery experiencing arcing faults output by the detection module is used for subsequent judgment. The value range of this confidence level is usually 0~1, such as 0.35, 0.50, 0.78, and 0.91.

[0036] The aforementioned detection module is trained using the electrical parameters of sample batteries from different categories and their corresponding arcing categories. The detection module can be trained independently or by training the arcing detection model. When training the detection module using the arcing detection model, the threshold update module within the arcing detection model may not update the confidence threshold; it will always use the initial confidence threshold for classification.

[0037] The arcing category includes those that have experienced arcing faults and those that have not, which are known labels corresponding to the electrical parameters of the sample batteries. For example... Figure 3As shown, the arc detection model includes a detection module, a threshold update module, and a classification module. This arc detection model can employ models such as support vector machines, neural network models, and machine learning models for classification. By adding a threshold update module to these models, the arc detection model required in this embodiment can be obtained. For example, a pre-trained support vector machine with radial basis function (RBF) kernels can be used.

[0038] Step 203: Based on the current operating condition parameters and the threshold update module, determine the current confidence threshold corresponding to the target battery under the current operating condition parameters; wherein, the threshold update module is used to substitute the current operating condition parameters into a preset polynomial function to obtain the confidence threshold correction amount, and calculate the sum of the confidence threshold correction amount and the preset initial confidence threshold to obtain the current confidence threshold.

[0039] The confidence threshold is the boundary threshold for determining whether a battery has experienced an arcing fault. Because batteries operate under different conditions, their electrical parameters such as current and voltage will vary, affecting the determination of arcing faults and causing fluctuations in the confidence threshold corresponding to batteries under different operating conditions.

[0040] In this embodiment, the preset polynomial function represents the correspondence between the current operating condition and the confidence threshold correction. The threshold update module can substitute the current operating condition parameters into the preset polynomial function to calculate the confidence threshold correction. Then, it can calculate the sum of the confidence threshold correction and the preset initial confidence threshold to obtain the current confidence threshold. This allows the module to find the current confidence threshold corresponding to the current operating condition parameters of the target battery, so as to accurately determine whether an arcing fault has occurred and avoid false alarms when the battery is lightly loaded and missed alarms when the battery is heavily loaded.

[0041] In addition to setting a polynomial function, the threshold update module can also use a multidimensional table mapping operating parameters to confidence threshold corrections. By looking up the value corresponding to the current operating parameter in this multidimensional table, the corresponding confidence threshold correction can be obtained. Then, the current confidence threshold is obtained by calculating the sum of the confidence threshold correction and the preset initial confidence threshold.

[0042] The aforementioned polynomial functions or tabular correspondences can be obtained based on the sample confidence levels of sample batteries under different operating conditions. Specifically, for each operating condition, using the historical electrical parameters of the sample batteries under that condition and the arcing detection model (or the detection module within the arcing detection model), the sample confidence level at which batteries without arcing faults are identified as having arcing faults can be obtained. This allows for the accurate determination of the fluctuation in the detection of batteries without arcing faults under that operating condition, i.e., the confidence threshold correction amount corresponding to that operating condition.

[0043] Here, the current operating parameters include the target battery's current State of Charge (SOC), current load rate, and current ambient temperature. The current SOC, representing the percentage of the battery's current remaining charge relative to its rated capacity, is a core parameter indicating the battery's chemical state. Different SOCs result in different internal resistances and terminal voltage responses. The current load rate is the ratio of the target battery's current current to its rated current, representing the power flow intensity. The load rate affects the target battery's loop noise and arc characteristic intensity. The current ambient temperature affects the characteristics of devices within the target battery, sensor noise characteristics, and arc combustion patterns.

[0044] In the polynomial function, the operating condition parameters are the independent variables, and the confidence threshold correction is the dependent variable.

[0045] For example, the polynomial function can be expressed as: In the formula, This represents the dependent variable, specifically the confidence threshold correction amount; Indicates the state of charge of the battery. Indicates the battery load rate. Indicates the ambient temperature of the battery; , , , , ... represent polynomial coefficients. Here, the polynomial function can be a linear polynomial, a quadratic polynomial, a cubic polynomial, etc.

[0046] In this embodiment, the threshold update module obtains the corresponding function value, i.e., the confidence threshold correction amount, by substituting the current state of charge, current load rate, and current ambient temperature of the target battery into a preset polynomial function. This confidence threshold correction amount is then superimposed on the initial confidence threshold to correct the initial confidence threshold, thus obtaining the current confidence threshold of the target battery under the current operating conditions.

[0047] The current state of charge (SOC), current load rate, and current ambient temperature of the target battery can be continuous, true values ​​or discrete ranges. For example, the SOC range could include 0-5%, 5%-20%, 20%-80%, 80%-95%, and 95%-100%; the load rate range could include 0-20%, 20%-50%, 50%-80%, and 80%-100%; and the ambient temperature range could include 0℃-15℃, 15℃-30℃, 30℃-45℃, and 45℃-60℃, etc. For batteries with the same SOC, load rate, and ambient temperature ranges, the corresponding confidence threshold and confidence threshold correction amount are also the same.

[0048] Here, characteristic values ​​can be selected for each interval, and these selected characteristic values ​​can be used as corresponding operating condition parameters to establish a polynomial function. These characteristic values ​​can be the minimum, maximum, or median of each interval. When the target battery's current state of charge, current load rate, and current ambient temperature fall within the corresponding interval, the confidence threshold correction can be calculated based on the characteristic value corresponding to that interval.

[0049] In addition, multidimensional tables can be created for each interval of the operating conditions. The index of the multidimensional table is the interval of state of charge, the interval of load rate, and the interval of ambient temperature. The mapped value is the confidence threshold correction amount corresponding to the combination of the intervals of state of charge, load rate, and ambient temperature.

[0050] Step 204: Input the confidence level and the current confidence threshold into the classification module to obtain the current arc detection result of the target battery.

[0051] In this embodiment, the classification module compares the confidence level with the current confidence threshold. If the confidence level of the arcing fault output by the detection module is greater than or equal to the current confidence threshold, it indicates that the target battery has a high degree of confidence in the arcing fault, and the current arcing detection result of the target battery can be determined to be an arcing fault.

[0052] If the confidence level of the arcing fault output by the detection module is less than the current confidence threshold, it indicates that the credibility of the arcing fault in the target battery is low, and it can be determined that the current arcing detection result of the target battery is that no arcing fault has occurred.

[0053] Here, the final output of the classification module, the current arc detection result, is also the final output of the arc detection model.

[0054] This application embodiment acquires the current electrical parameters and current operating condition parameters of the target battery in the energy storage system. The current operating condition parameters may include the target battery's current state of charge, current load rate, and current ambient temperature. Using the current electrical parameters and a detection module, the confidence level of an arcing fault in the target battery can be accurately determined. The detection module is a module within an arcing detection model, which also includes a threshold update module and a classification module. Then, using the current operating condition parameters and the threshold update module, the current confidence threshold corresponding to the target battery under the current operating condition parameters is determined. The threshold update module substitutes the current operating condition parameters into a preset polynomial function to obtain a confidence threshold correction amount, and calculates the sum of the confidence threshold correction amount and a preset initial confidence threshold to obtain the current confidence threshold. This fully considers the changes in the target battery's operating conditions and accurately obtains a suitable confidence threshold for the target battery. Finally, this confidence level is compared with the current confidence threshold by the classification module, which can accurately determine the current arcing detection result of the target battery, reducing false alarms and false negatives of arcing faults and improving the accuracy and reliability of arcing fault detection.

[0055] In some embodiments, the DC arcing detection method for an energy storage system provided in this embodiment further includes: firstly acquiring the first historical electrical parameters and sample operating condition parameters of a first sample battery under multiple operating conditions; then, for each operating condition, determining the sample confidence level of arcing fault occurrence of the first sample battery under that operating condition based on the first historical operating parameters and the detection module; and determining the confidence threshold correction amount corresponding to that operating condition based on the sample confidence level of arcing fault occurrence of the first sample battery under that operating condition. Finally, establishing a polynomial function between the operating condition parameters and the confidence threshold correction amount based on the sample operating condition parameters and the corresponding confidence threshold correction amount for each operating condition.

[0056] The first sample battery is a normal battery that has been confirmed to be free of arcing faults in laboratory or field calibration. This serves as a benchmark for learning the confidence fluctuation boundary of normal behavior in the event of arcing faults.

[0057] For continuous parameter operating conditions, different sample operating condition parameters correspond to different operating conditions. For each operating condition, the first historical electrical parameters of the first sample battery are obtained. The first historical electrical parameters can be the current data and voltage data of the same first sample battery at different times under the operating condition, or the current data and voltage data of different first sample batteries under the operating condition.

[0058] For discrete operating conditions, different sample operating condition parameters may correspond to the same operating condition. By obtaining the first historical electrical parameters and sample operating condition parameters of the first sample battery under multiple operating conditions, the first historical electrical parameters of the first sample battery for each operating condition can be obtained based on the sample operating condition data. These first historical electrical parameters can be current and voltage data of the same first sample battery at different times under that operating condition, or current and voltage data of different first sample batteries under that operating condition. Here, the sample operating condition data of the first sample battery under the same operating condition may differ, but they all belong to that operating condition.

[0059] In this embodiment, for each operating condition, a trained detection module is used to define the noise baseline for the confidence level of a battery that is normal (i.e., has not experienced an arcing fault) when it is determined to have experienced an arcing fault. Since the first historical electrical parameters come from batteries confirmed to be free of arcing faults, fluctuations in the confidence levels corresponding to different historical electrical parameters within the same operating condition are not due to battery arcing, but rather to inherent system noise and changes in operating condition parameters. This baseline clarifies the boundary between the confidence level fluctuations when an arcing fault is detected as occurring, allowing for the determination of the confidence threshold correction for each operating condition. This establishes a correspondence between operating condition parameters and the confidence threshold correction.

[0060] For each operating condition, the detection module can be used to detect the first historical operating parameters under that condition to obtain the sample confidence level of the first sample battery experiencing arcing failure under that operating condition. Each operating condition typically includes multiple sets of first historical operating parameters, thus yielding multiple sample confidence levels for each operating condition. These sample confidence levels will fluctuate within a certain range, allowing the determination of the confidence threshold correction amount for that operating condition based on the sample confidence levels under the same operating condition. For example, for each operating condition, the highest sample confidence level can be selected as the confidence threshold correction amount for that operating condition, and then added to the initial confidence threshold to obtain the confidence threshold for that operating condition.

[0061] Once the confidence threshold correction for each operating condition is obtained, a polynomial function relating the operating condition parameters to the confidence threshold correction can be established based on the sample operating condition data for each condition and the corresponding confidence threshold correction. Additionally, a table of discrete calibration points can also be created.

[0062] For continuous parameter operating conditions, different sample operating condition parameters correspond to different operating conditions. However, when obtaining the first historical electrical parameters and sample operating condition parameters, it may be difficult to exhaust all operating conditions. Therefore, discrete points of a fixed length can be selected to determine a finite number of confidence threshold corrections for operating conditions. Then, polynomial fitting can be used to obtain the confidence threshold corrections for different operating conditions. For example, the selection interval for the state of charge can be 1%, 3%, 5%, and 10%, the selection interval for the load rate can be 1%, 3%, 5%, and 10%, and the selection interval for the ambient temperature can be 1℃, 2℃, 3℃, and 5℃.

[0063] Furthermore, for operating conditions in discrete intervals, a polynomial function and a multidimensional table can be established using the characteristic values ​​of the operating condition parameters selected for each interval as the dependent variable. Alternatively, a multidimensional table can be established using the interval range of the operating condition parameters for each interval as the dependent variable.

[0064] Optionally, the confidence threshold correction amount corresponding to the operating condition can be determined based on the sample confidence level of the first sample battery experiencing arcing failure under the operating condition. This can be achieved by first obtaining the confidence distribution of the first sample battery experiencing arcing failure under the operating condition based on the sample confidence level of the first sample battery experiencing arcing failure under the operating condition; and then determining the confidence threshold correction amount corresponding to the operating condition based on the sample confidence level corresponding to the preset quantile in the confidence distribution.

[0065] The confidence distribution described above represents the probability distribution of confidence scores for all samples of the first sample battery experiencing an arcing fault under a given operating condition. It can be represented by a histogram or probability density curve. The shape of the confidence distribution reflects the normal fluctuations under the operating condition.

[0066] A preset quantile is a pre-selected percentage value between 0 and 1, representing the statistical tolerance standard for confidence levels. For example, you can choose the 99th quantile, which represents the confidence level at the 99th percentile. You can also choose the 95th quantile, which represents the confidence level at the 95th percentile. This preset quantile can be selected according to specific needs.

[0067] In this embodiment, for each operating condition, a histogram or probability density curve is plotted using the confidence scores of all samples under that operating condition to obtain the confidence distribution for that operating condition. Since the aforementioned sample confidence scores represent the confidence scores of batteries that have not experienced arcing faults when they are determined to have experienced arcing faults, the resulting confidence distribution is typically a distribution curve with a peak skewed to the left (close to 0) and a long tail on the right. Most sample confidence scores are low, for example, concentrated around 0.05. Due to noise, a few sample confidence scores are high, for example, reaching 0.15 or even 0.20.

[0068] Here, considering that directly taking the maximum value of the confidence distribution as the threshold correction amount means that if even one normal sample is heavily polluted by interference, the final confidence threshold will be significantly inflated by this outlier, causing the confidence threshold to lose sensitivity. Therefore, by setting a preset quantile, the highest level of the vast majority of data in all normal samples that have not experienced arcing faults can be selected as the boundary, discarding the most extreme abnormal noise points at the tail end, such as the highest 1%, actively excluding the most extreme outlier values ​​that may be caused by unknown sporadic interference, reducing the probability that the confidence threshold is affected by individual samples, and ensuring the rationality of the final confidence threshold. For example, if the preset quantile is the 99th quantile, and a certain operating condition includes 1000 sample confidence levels, then the 1000 sample confidence levels are sorted from low to high, and the confidence level of the 990th sample is the value of the 99th quantile. If the confidence level of the 990th sample is 0.12, then the corresponding confidence threshold correction amount for that operating condition can be determined accordingly.

[0069] When determining the confidence threshold for an operating condition, the sample confidence level corresponding to the preset quantile (i.e., the confidence threshold correction) can be added to the initial confidence threshold to obtain the confidence threshold corresponding to that operating condition. For example, if the initial confidence threshold is 0.5 and the sample confidence level corresponding to the preset quantile is 0.12, then the confidence threshold corresponding to that operating condition is 0.62.

[0070] In some embodiments, the current electrical parameters include current current data and current voltage data. This embodiment determines the confidence level of a target battery experiencing an arcing fault based on the current electrical parameters and the detection module. This can be achieved by: first extracting time-frequency domain features from the current current and current voltage data; then reconstructing the time-frequency domain features to obtain fused features; wherein the dimension of the fused features is smaller than the dimension of the time-frequency domain features; finally, inputting the fused features to the detection module to obtain the confidence level of the target battery experiencing an arcing fault output by the detection module; wherein the detection module maps the fused features to a preset interval to obtain the confidence level.

[0071] Current electrical parameters include the target battery's current current and voltage data. Time-domain and frequency-domain features are extracted from the current current and voltage data respectively, yielding time-frequency domain features. These features can include current time-domain features, current frequency-domain features, voltage time-domain features, and voltage frequency-domain features. The current current and voltage data can be acquired using a sliding window method.

[0072] Current time-domain characteristics can be derived by extracting statistics such as variance, peak factor, and impulse index of the AC component of the current within a sliding window, characterizing the random fluctuations and impulse characteristics of the current on the time axis. (See also...) Figure 4 and Figure 5In the figure, the horizontal axis represents frequency and the vertical axis represents amplitude. It is evident that the current of a battery experiencing arcing faults differs significantly in amplitude at different frequencies from that of a battery without arcing faults. Current frequency domain characteristics can be obtained by performing a Fast Fourier Transform (FFT) on the current data to extract the total harmonic energy proportion (1-100kHz), the sub-band energy proportion (10-30kHz), and the frequency domain peak frequency (after excluding switching frequencies), thus characterizing the high-frequency discharge spectrum characteristics caused by arcing. Specifically, the TMR sensor can collect only the AC current, extract the current time-domain statistical characteristics, perform an FFT on it, extract high-frequency harmonic components (such as the 1-100kHz band), and calculate the harmonic energy proportion.

[0073] Voltage time-domain features can be extracted by considering the voltage range within a sliding window and the maximum rate of change between adjacent sampling points, thus characterizing the voltage drop amplitude and steepness of the sudden change during arcing. Voltage frequency-domain features can be obtained by performing an FFT on the current voltage data to extract the proportion of harmonic energy in the 1-100kHz range, spectral dispersion, etc., thus characterizing the coupled high-frequency noise components in the voltage.

[0074] Here, the variance, range, and maximum rate of change of the time-domain features can all be calculated simultaneously in a single sliding window traversal; the peak factor and pulse index share intermediate results. The overall computational complexity of time-domain feature extraction is O(N), where N is the number of sampling points within the window. In the frequency-domain features, only one FFT of the current data and one FFT of the voltage data are performed. All frequency-domain features are accumulated, compared, and subjected to simple algebraic operations on the FFT results, without increasing the number of Fourier transforms. The weighted summation of the spectral dispersion can also be quickly completed using the amplitude array output by the FFT. Therefore, the computational overhead on the DSP is only two FFTs, with the rest being linear traversal and accumulation operations, which can be completed entirely within the DSP's control cycle, meeting the requirements for real-time arc detection.

[0075] In this embodiment, the aforementioned time-domain features, current-frequency-domain features, voltage-time-domain features, and voltage-frequency-domain features together constitute a high-dimensional original feature vector, namely, the time-frequency domain features. This time-frequency domain feature fully records the behavioral information of current and voltage within each analysis domain within the sliding window, but it contains significant redundancy and is coupled with operating condition interference. Some features are highly correlated, such as current variance and total current harmonic energy. These redundant features not only increase computational overhead but also reduce the generalization ability of the classifier. Normal operations such as battery balancing, switching actions, and power surges simultaneously generate arc-like feature responses in multiple dimensions of the time and frequency domains. Simple combination does not decouple and reconstruct the information of each dimension of features, resulting in interference signals highly overlapping with real arcs in some dimensions, insufficient feature discrimination, and consequently a high false alarm rate. In the multi-dimensional time-frequency domain features, key information sensitive to arcing is often diluted by a large number of irrelevant or weakly correlated dimensions. If the aforementioned time-frequency domain features are directly input into the arc detection model, the redundancy and interference will lead to blurred classification boundaries, making it difficult to form a clear and compact decision boundary.

[0076] Therefore, high-dimensional original features are used for information decoupling and reconstruction to generate low-dimensional fused features. These fused features are no longer simple representations of a single signal source, but rather a compact expression of the information about current and voltage that has the highest correlation with arcing in each analysis domain.

[0077] Since the fused features have removed redundancy, suppressed interference, and focused on key information, the low-dimensional fused features obtained through dimensionality reduction can be used as input to the trained detection module. This allows for high-accuracy binary classification in this feature space, outputting the confidence level of an arcing fault as a probability. Specifically, the detection module can perform feature calculations on the fused features to obtain a score for arcing fault occurrence. This score is then mapped to a range of 0-1 to obtain the probability of an arcing fault, which is the confidence level output by the detection module.

[0078] Optionally, in this embodiment, feature reconstruction of the time-frequency domain features to obtain fused features can be performed as follows: First, calculate the covariance between the eigenvalues ​​of every two dimensions in the time-frequency domain features; then, obtain the covariance matrix based on all the covariances of the time-frequency domain features; perform eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues ​​and eigenvectors corresponding to each eigenvalue; select a preset number of eigenvectors in descending order of eigenvalues ​​to obtain an eigenvector matrix; wherein the preset number is less than the dimension of the time-frequency domain features; finally, determine the fused features based on the time-frequency domain features and the eigenvector matrix.

[0079] In this embodiment, principal component analysis can be used to reduce the dimensionality of the time-frequency domain features. By decomposing the covariance matrix, the original 10-dimensional features are projected onto the principal component direction with the largest variance, retaining the top principal components (e.g., 3-dimensional) with the highest cumulative variance contribution rate, thus achieving dimensionality compression.

[0080] The covariance mentioned above is a statistical measure of the cooperative relationship between two feature dimensions. The covariance matrix is ​​a square matrix. For example, if there are 10 original features, the covariance matrix is ​​a 10×10 square matrix, where the element in the i-th row and j-th column is the covariance between the i-th and j-th features. The elements on the diagonal are the covariance of each feature with itself, i.e., the variance.

[0081] Eigenvalues ​​represent the magnitude of variance along the direction of the corresponding eigenvector. Larger eigenvalues ​​indicate a more dispersed distribution of data along that direction, containing more information. Eigenvectors represent the direction of the new coordinate axes and are linear combination coefficients of the original features. By performing eigenvalue decomposition on the covariance matrix, the eigenvalues ​​and eigenvectors of the covariance matrix can be obtained.

[0082] Here, the feature values ​​can be sorted from largest to smallest, and the feature vectors corresponding to the largest preset number of feature values ​​are used as row vectors to obtain a feature vector matrix. The time-frequency domain features are then transformed into a new space constructed from the feature vector matrix to obtain the final fused features. The number of selected feature values, i.e., the preset number, can be the same as the dimension of the fused features. For example, if the dimension of the fused features is 3, then the preset number can be 3. Furthermore, the preset number is usually smaller than the dimension of the time-frequency domain features.

[0083] By reducing the dimensionality of the time-frequency domain features, fused features are obtained. Redundant features can be removed, highly correlated original dimensions can be merged, and the new feature dimensions are approximately independent, enabling the arc detection model to capture the decision boundary with a simpler structure.

[0084] Arc-like responses generated by operating conditions such as balancing and switching often exhibit specific covariant patterns across multiple original dimensions. Principal component analysis separates these responses from the main variation patterns of arcing by maximizing variance, thus suppressing operating condition interference into secondary dimensions in the dimensionality reduction space. This significantly weakens the projection values ​​of interference components in the retained principal components, thereby improving the anti-interference effect of arcing detection.

[0085] Furthermore, the dimensionality reduction process is essentially a weighted recombination of the original features. Dimensions with high sensitivity to arcing (such as the high-frequency subband energy of current, voltage range and rate of change) will receive greater weights and be prominently expressed in the fused features, thereby improving the feature discrimination and enabling the arcing detection model to perform accurate classification.

[0086] Experimental verification was conducted using the arc detection model constructed with support vector machines. After reducing the original 10-dimensional time-frequency domain features to 3-dimensional fused features, the separability of samples with and without arc faults in the 3-dimensional space was significantly improved. The arc detection model can obtain a clearer and more stable decision hyperplane.

[0087] Optionally, in this embodiment, the fusion features are determined based on time-frequency domain features and feature vector matrices. This can be achieved by first projecting the time-frequency domain features onto the space constructed by the feature vector matrix to obtain initial dimensionality-reduced features, wherein the preset number is greater than the dimension of the fusion features; then, the fusion features are determined based on the preset projection matrix and the initial dimensionality-reduced features, wherein the projection matrix is ​​calculated based on the intra-class scatter matrix and the inter-class scatter matrix, and the intra-class scatter matrix and the inter-class scatter matrix are calculated based on the initial dimensionality-reduced features of different categories.

[0088] In this embodiment, a combined dimensionality reduction approach can also be used to reduce the dimensionality of time-frequency domain features. For example, principal component analysis can be used to reduce the dimensionality of time-frequency domain features to obtain initial dimensionality-reduced features, and then linear discriminant analysis (LDA) can be used to further reduce the dimensionality of the initial dimensionality-reduced features to obtain the final fused features.

[0089] The LDA mentioned above is a supervised learning dimensionality reduction technique. Its idea is to maximize the inter-class mean and minimize the intra-class variance, projecting the data onto a low dimension, and ensuring that the projected points of data of the same class are as close as possible, while the center points of the projected points of data of different classes are as far apart as possible.

[0090] The projection matrix here is a matrix composed of the first k eigenvectors obtained by solving the intra-class scatter matrix and the inter-class scatter matrix. Here, k is the second preset number, and the preset number corresponding to the eigenvector matrix is ​​the first preset number. The second preset number is usually less than the first preset number. The second preset number is the same as the dimension of the fused feature, such as the dimension of the fused feature can be 3 or 4, etc.

[0091] The projection matrix was pre-calculated in an offline environment using historical data of known arcing categories. Specifically, it acquires the initial dimensionality-reduced features of batteries that experienced arcing faults and the initial dimensionality-reduced features of batteries that did not experience arcing faults. Each sample of the initial dimensional feature is labeled with its arcing category.

[0092] The intra-class scatter matrix measures the dispersion of sample points within the same class relative to the class mean, while the inter-class scatter matrix measures the dispersion of the means of different classes relative to the population mean. This is achieved by calculating the matrices. By finding the k largest eigenvalues ​​and k eigenvectors, we can obtain the corresponding projection matrix, where... Represents the within-class scatter matrix. This represents the inter-class scatter matrix.

[0093] Finally, projecting the initial dimensionality-reduced features onto the space constructed by the projection matrix yields the fused features, achieving high information compression and improving the processing speed of the arc detection model. Furthermore, principal component analysis (PCA) dimensionality reduction removes random noise and linear correlations between features, ensuring clean and non-singular input data. By purposefully reshaping the feature space using LDA, the distance between normal samples without arcing faults and samples with arcing faults is maximized in the dimensionality-reduced space, simplifying the decision boundary of the arc detection model and improving detection accuracy.

[0094] Optionally, the process of pre-determining the projection matrix in the above embodiments can be as follows: First, obtain multiple sets of second historical electrical parameters of the second sample battery and the arcing category of each set of second historical electrical parameters; wherein, the second historical electrical parameters include historical current data and historical voltage data. For each set of second historical electrical parameters, extract historical time-frequency domain features from the historical current data and historical voltage data; determine the historical initial dimensionality reduction features based on the historical time-frequency domain features. Then, calculate the intra-class mean of each arcing category and the overall mean of the second sample battery based on the historical initial dimensionality reduction features of each arcing category. Next, determine the internal divergence matrix of each arcing category based on the historical initial dimensionality reduction features of each arcing category and its corresponding intra-class mean of the arcing category, calculate the sum of the internal divergence matrices of each arcing category to obtain the intra-class divergence matrix; and determine the inter-class divergence matrix based on the class difference between the intra-class mean and the overall mean of each arcing category; finally, determine the projection matrix based on the intra-class divergence matrix and the inter-class divergence matrix.

[0095] The second sample batteries mentioned above include batteries that experienced arcing failures and batteries that did not experience arcing failures. The determination of time-frequency domain features and initial dimensionality reduction features can refer to the process in the above embodiments. After feature extraction, a first dimensionality reduction is performed using principal component analysis to ensure that the initial dimensionality reduction features used in determining the projection matrix are consistent with the method of obtaining the initial dimensionality reduction features when using the projection matrix.

[0096] The within-class mean is the mean vector of all samples within the same class, specifically the mean vector of the historical initial dimensionality-reduced features for the class that experienced arcing faults, and the mean vector of the historical initial dimensionality-reduced features for the class that did not experience arcing faults. The overall mean is the total mean vector of all samples, i.e., the mean vector of all historical initial dimensionality-reduced features.

[0097] The internal scatter matrix can be represented as In the formula, Indicates the first One type of arc drawing, Indicates the first Each type of arc The internal scatter matrix, Indicates the first Each type of arc The first in One sample (historical initial dimensionality reduction features). Indicates the arc type as within-class mean Represents the transpose of a vector. It can be 1, 2..., such as A value of 1 indicates that no arcing fault has occurred. A value of 2 indicates an arcing fault has occurred. Specifically, for each arcing category, the internal scatter matrix of that category is determined using the deviation vector from the intra-class mean to each historical initial dimensionality-reduced feature of that category. Correspondingly, the sum of the internal scatter matrices for each arcing category yields the intra-class scatter matrix.

[0098] The inter-class scatter matrix can be represented as In the formula, Indicates the first Each type of arc The number of samples (the number of initial historical dimensionality reduction features). Indicates the first Each type of arc within-class mean This represents the overall mean.

[0099] Finally, the matrix obtained by analyzing the intra-class scatter matrix and the inter-class scatter matrix... By decomposing the data, we can obtain multiple eigenvalues ​​and the eigenvectors corresponding to each eigenvalue. By selecting the eigenvectors corresponding to the k largest eigenvalues, we can construct the corresponding projection matrix.

[0100] In some embodiments, obtaining the current electrical parameters and current operating condition parameters of the target battery in the energy storage system can be achieved by obtaining the current electrical parameters and current operating condition parameters of the target battery in the energy storage system within the current time period. The aforementioned current arcing detection result is determined for the current time period in which the obtained current electrical parameters and current operating condition parameters are located, indicating whether an arcing fault has occurred in the target battery within the current time period.

[0101] After inputting the confidence level and the current confidence threshold into the classification module to obtain the current arcing detection result of the target battery in the current time period, it is also possible to: first obtain a preset number of historical arcing detection results before the current time period; if both the historical arcing detection results and the current arcing detection results indicate that an arcing fault has occurred, then it is determined that the target battery has an arcing fault; otherwise, it is determined that the target battery has not experienced an arcing fault.

[0102] In this embodiment, considering that arcing is usually a continuous physical phenomenon rather than a brief noise spike, the historical arcing detection results of multiple consecutive time periods before the current time period can be obtained. By using the continuous arcing detection results, the true arcing fault and transient noise can be distinguished.

[0103] Here, the number of historical arc detection results obtained is a preset number, such as 1, 2, 3, 4, and 5, etc., which can be set according to needs.

[0104] If both historical and current arcing detection results indicate an arcing fault, it means the target battery has been identified as experiencing an arcing fault across multiple consecutive time periods. Therefore, it can be confirmed that the target battery and its corresponding branch have experienced an arcing fault. Once an arcing fault is confirmed, the faulty branch containing the target battery can be disconnected within milliseconds, and an alarm can be triggered to ensure the safety of the energy storage system.

[0105] If the historical arcing detection results and the current arcing detection results are not both indicative of an arcing fault, it indicates that the arcing detection model may be misjudged due to sudden load changes in the target battery or external electromagnetic interference. In this case, it can be determined that the target battery has not yet experienced an arcing fault, and continuous monitoring of the target battery should continue.

[0106] Furthermore, after obtaining the arcing detection results for each time period, the arcing detection results identified as having an arcing fault can be continuously counted. This count value is then compared with a preset number to determine whether an arcing fault has occurred. For example, if the arcing detection result indicates an arcing fault has occurred, the count value is incremented by 1; if the arcing detection result indicates no arcing fault has occurred, the count value is updated to 0. When the count value is the same as the preset number, it can be determined that the target battery has experienced an arcing fault.

[0107] This embodiment filters the arcing detection results from multiple time periods, which can reduce the impact of occasional noise in a single time period, reduce false alarms of arcing faults, and improve the accuracy and reliability of arcing fault detection.

[0108] This application embodiment acquires the current electrical parameters and current operating condition parameters of the target battery in the energy storage system. The current operating condition parameters may include the target battery's current state of charge, current load rate, and current ambient temperature. Using the current electrical parameters and a detection module, the confidence level of an arcing fault in the target battery can be accurately determined. The detection module is a module within an arcing detection model, which also includes a threshold update module and a classification module. Then, using the current operating condition parameters and the threshold update module, the current confidence threshold corresponding to the target battery under the current operating condition parameters is determined. The threshold update module substitutes the current operating condition parameters into a preset polynomial function to obtain a confidence threshold correction amount, and calculates the sum of the confidence threshold correction amount and a preset initial confidence threshold to obtain the current confidence threshold. This fully considers the changes in the target battery's operating conditions and accurately obtains a suitable confidence threshold for the target battery. Finally, this confidence level is compared with the current confidence threshold by the classification module, which can accurately determine the current arcing detection result of the target battery, reducing false alarms and false negatives of arcing faults and improving the accuracy and reliability of arcing fault detection.

[0109] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0110] The following are device embodiments of this application. For details not described in detail, please refer to the corresponding method embodiments described above.

[0111] Figure 6 A schematic diagram of the charging module distribution device for a charging pile provided in an embodiment of this application is shown. For ease of explanation, only the parts related to the embodiment of this application are shown, and are described in detail below: like Figure 6 As shown, the charging module distribution device 60 of the charging pile includes: The acquisition module 61 is used to acquire the current electrical parameters and current operating condition parameters of the target battery in the energy storage system; wherein, the current operating condition parameters include the current state of charge, current load rate and current ambient temperature of the target battery. The processing module 62 is used to determine the confidence level of the target battery having an arcing fault based on the current electrical parameters and the detection module; wherein, the detection module is a module in the arcing detection model, which also includes a threshold update module and a classification module. The detection module is trained based on different categories of electrical parameters and corresponding arcing categories, including arcing fault occurrence and no arcing fault occurrence. The threshold module 63 is used to determine the current confidence threshold corresponding to the target battery under the current operating conditions based on the current operating condition parameters and the threshold update module; wherein, the threshold update module is used to substitute the current operating condition parameters into a preset polynomial function to obtain the confidence threshold correction amount, and calculate the sum of the confidence threshold correction amount and the preset initial confidence threshold to obtain the current confidence threshold. The determination module 64 is used to input the confidence level and the current confidence threshold into the classification module to obtain the current arc detection result of the target battery.

[0112] In one possible implementation, the threshold module 63 is also used for: Obtain the first historical electrical parameters and sample operating condition parameters of the first sample battery under various operating conditions; For each operating condition, based on the first historical operating parameters and detection module under that operating condition, the sample confidence level of the first sample battery experiencing arcing failure under that operating condition is determined; based on the sample confidence level of the first sample battery experiencing arcing failure under that operating condition, the corresponding confidence threshold correction amount is determined; Based on the sample operating condition parameters for each operating condition and the corresponding confidence threshold correction, a polynomial function is established between the operating condition parameters and the confidence threshold correction.

[0113] In one possible implementation, the threshold module 63 is specifically used for: Based on the sample confidence level of the first sample battery experiencing arcing failure under this operating condition, the confidence distribution of the first sample battery experiencing arcing failure under this operating condition is obtained. Based on the sample confidence level corresponding to the preset quantile in the confidence distribution, determine the confidence threshold correction amount corresponding to this operating condition.

[0114] In one possible implementation, the current electrical parameters include current current data and current voltage data; The detection module 62 is specifically used for: Extract time-frequency domain features from current current data and current voltage data; The time-frequency domain features are reconstructed to obtain fused features; the dimension of the fused features is smaller than the dimension of the time-frequency domain features. The fused features are input into the detection module to obtain the confidence level of the target battery arcing fault output by the detection module; wherein, the detection module is used to map the fused features to a preset interval to obtain the confidence level.

[0115] In one possible implementation, the detection module 62 is specifically used for: Calculate the covariance between the eigenvalues ​​of every two dimensions in the time-frequency domain features; Based on all the covariances of the time-frequency domain features, the covariance matrix is ​​obtained; The covariance matrix is ​​decomposed into eigenvalues ​​to obtain multiple eigenvalues ​​and the corresponding eigenvectors for each eigenvalue. A predetermined number of eigenvectors are selected in descending order of eigenvalues ​​to obtain an eigenvector matrix; wherein the predetermined number is less than the dimension of the time-frequency domain features. Based on time-frequency domain features and eigenvector matrices, the fusion features are determined.

[0116] In one possible implementation, the detection module 62 is specifically used for: The time-frequency domain features are projected onto the space constructed by the feature vector matrix to obtain the initial dimensionality-reduced features; where the preset number is greater than the dimension of the fused features; The fusion features are determined based on the preset projection matrix and initial dimensionality reduction features; wherein, the projection matrix is ​​calculated based on the intra-class scatter matrix and the inter-class scatter matrix, and the intra-class scatter matrix and the inter-class scatter matrix are calculated based on the initial dimensionality reduction features of different categories.

[0117] In one possible implementation, the detection module 62 is further used for: Obtain multiple sets of second historical electrical parameters of the second sample battery and the arcing category of each set of second historical electrical parameters; wherein, the second historical electrical parameters include historical current data and historical voltage data; For each group of second historical electrical parameters, historical time-frequency domain features are extracted from historical current data and historical voltage data; based on the historical time-frequency domain features, the initial historical dimensionality reduction features are determined. Based on the historical initial dimensionality reduction features of each arcing category, calculate the intra-class mean of each arcing category and the overall mean of the second sample batteries; Based on the historical initial dimensionality reduction features of each arc category and the intra-class mean of its corresponding arc category, the internal divergence matrix of each arc category is determined; Calculate the sum of the internal scatter matrices for each arc category to obtain the intra-class scatter matrix; The inter-class scatter matrix is ​​determined based on the difference between the intra-class mean and the overall mean for each arc type. The projection matrix is ​​determined based on the intra-class scatter matrix and the inter-class scatter matrix.

[0118] In one possible implementation, module 61 is used for: Obtain the current electrical parameters and current operating condition parameters of the target battery in the energy storage system within the current time period; Module 64 is also used for: Obtain a preset number of historical arc detection results consecutively before the current time period; If both the historical and current arcing detection results indicate an arcing fault, then the target battery is determined to have an arcing fault; otherwise, the target battery is determined not to have an arcing fault.

[0119] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0120] This application also provides an energy storage system. For details not described in detail, please refer to the corresponding method embodiments described above.

[0121] In some embodiments, the energy storage system includes a controller and a plurality of batteries. The controller is used to implement the methods in the various method embodiments described above.

[0122] This application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0123] This application also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the methods in the above-described method embodiments.

[0124] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0125] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0126] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for detecting DC arcing in an energy storage system, characterized in that, include: The current electrical parameters and current operating condition parameters of the target battery in the energy storage system are obtained; wherein, the current operating condition parameters include the current state of charge, current load rate and current ambient temperature of the target battery. Based on the current electrical parameters and the detection module, the confidence level of the target battery experiencing an arcing fault is determined; wherein, the detection module is a module in the arcing detection model, and the arcing detection model further includes a threshold update module and a classification module, and the detection module is trained based on different categories of electrical parameters and corresponding arcing categories, the arcing categories including arcing fault occurrence and no arcing fault occurrence; Based on the current operating condition parameters and the threshold update module, the current confidence threshold corresponding to the target battery under the current operating condition parameters is determined; wherein, the threshold update module is used to substitute the current operating condition parameters into a preset polynomial function to obtain a confidence threshold correction amount, and calculate the sum of the confidence threshold correction amount and the preset initial confidence threshold to obtain the current confidence threshold. The confidence level and the current confidence threshold are input into the classification module to obtain the current arc detection result of the target battery.

2. The DC arcing detection method for an energy storage system according to claim 1, characterized in that, The method further includes: Obtain the first historical electrical parameters and sample operating condition parameters of the first sample battery under various operating conditions; For each operating condition, based on the first historical electrical parameters under that operating condition and the detection module, the sample confidence level of the first sample battery experiencing arcing failure under that operating condition is determined; based on the sample confidence level of the first sample battery experiencing arcing failure under that operating condition, the corresponding confidence threshold correction amount is determined. Based on the sample operating condition parameters for each operating condition and the corresponding confidence threshold correction, a polynomial function is established between the operating condition parameters and the confidence threshold correction.

3. The DC arcing detection method for an energy storage system according to claim 2, characterized in that, The step of determining the confidence threshold correction amount corresponding to the operating condition based on the sample confidence level of the first sample battery experiencing arcing failure under the operating condition includes: Based on the sample confidence level of the first sample battery experiencing arcing failure under this operating condition, the confidence distribution of the first sample battery experiencing arcing failure under this operating condition is obtained. The confidence threshold correction amount corresponding to the operating condition is determined based on the sample confidence level corresponding to the preset quantile in the confidence distribution.

4. The DC arcing detection method for an energy storage system according to any one of claims 1 to 3, characterized in that, The current electrical parameters include current current data and current voltage data; The step of determining the confidence level of the target battery experiencing an arcing fault based on the current electrical parameters and the detection module includes: Extract time-frequency domain features from the current current data and the current voltage data; The time-frequency domain features are reconstructed to obtain fused features; wherein the dimension of the fused features is smaller than the dimension of the time-frequency domain features. The fused features are input to the detection module to obtain the confidence level of the target battery having an arcing fault, output by the detection module; wherein, the detection module is used to map the fused features to a preset interval to obtain the confidence level.

5. The DC arcing detection method for an energy storage system according to claim 4, characterized in that, The time-frequency domain features are reconstructed to obtain fused features, including: Calculate the covariance between the eigenvalues ​​of every two dimensions in the time-frequency domain features; Based on all the covariances of the time-frequency domain features, the covariance matrix is ​​obtained; The covariance matrix is ​​subjected to eigenvalue decomposition to obtain multiple eigenvalues ​​and the eigenvector corresponding to each eigenvalue; A preset number of feature vectors are selected in descending order of their feature values ​​to obtain a feature vector matrix; wherein the preset number is less than the dimension of the time-frequency domain features. Based on the time-frequency domain features and the feature vector matrix, the fusion features are determined.

6. The DC arcing detection method for an energy storage system according to claim 5, characterized in that, The determination of fused features based on the time-frequency domain features and the feature vector matrix includes: The time-frequency domain features are projected onto the space constructed by the feature vector matrix to obtain initial dimensionality-reduced features; wherein the preset number is greater than the dimension of the fused features; The fusion features are determined based on the preset projection matrix and the initial dimensionality reduction features; wherein the projection matrix is ​​calculated based on the intra-class scatter matrix and the inter-class scatter matrix, and the intra-class scatter matrix and the inter-class scatter matrix are calculated based on the initial dimensionality reduction features of different categories.

7. The DC arcing detection method for an energy storage system according to claim 6, characterized in that, The method further includes: Obtain multiple sets of second historical electrical parameters of the second sample battery and the arcing category of each set of second historical electrical parameters; wherein, the second historical electrical parameters include historical current data and historical voltage data; For each set of second historical electrical parameters, historical time-frequency domain features are extracted from the historical current data and the historical voltage data; based on the historical time-frequency domain features, the initial historical dimensionality reduction features are determined. Based on the historical initial dimensionality reduction features of each arcing category, calculate the intra-class mean of each arcing category and the overall mean of the second sample batteries; Based on the historical initial dimensionality reduction features of each arc category and the intra-class mean of its corresponding arc category, the internal divergence matrix of each arc category is determined; The sum of the internal scatter matrices for each arc category is calculated to obtain the intra-class scatter matrix; The inter-class scatter matrix is ​​determined based on the difference between the intra-class mean and the overall mean for each arc category. The projection matrix is ​​determined based on the intra-class scatter matrix and the inter-class scatter matrix.

8. The DC arcing detection method for an energy storage system according to any one of claims 1 to 3, characterized in that, The acquisition of the current electrical parameters and current operating condition parameters of the target battery in the energy storage system includes: Obtain the current electrical parameters and current operating condition parameters of the target battery in the energy storage system within the current time period; After inputting the confidence level and the current confidence threshold into the classification module to obtain the current arc detection result of the target battery, the method further includes: Obtain a preset number of historical arc detection results consecutively before the current time period; If both the historical arcing detection results and the current arcing detection results indicate that an arcing fault has occurred, then it is determined that the target battery has experienced an arcing fault; otherwise, it is determined that the target battery has not experienced an arcing fault.

9. A DC arc detection device for an energy storage system, characterized in that, include: The acquisition module is used to acquire the current electrical parameters and current operating condition parameters of the target battery in the energy storage system; wherein, the current operating condition parameters include the current state of charge, current load rate and current ambient temperature of the target battery. The processing module is used to determine the confidence level of the target battery having an arcing fault based on the current electrical parameters and the detection module; wherein, the detection module is a module in the arcing detection model, and the arcing detection model further includes a threshold update module and a classification module. The detection module is trained based on different categories of electrical parameters and corresponding arcing categories, and the arcing categories include arcing fault occurrence and no arcing fault occurrence. The threshold module is used to determine the current confidence threshold corresponding to the target battery under the current operating conditions based on the current operating condition parameters and the threshold update module; wherein, the threshold update module is used to substitute the current operating condition parameters into a preset polynomial function to obtain a confidence threshold correction amount, and calculate the sum of the confidence threshold correction amount and the preset initial confidence threshold to obtain the current confidence threshold. The determination module is used to input the confidence level and the current confidence threshold into the classification module to obtain the current arc detection result of the target battery.

10. An energy storage system, characterized in that, It includes a controller and multiple batteries, the controller being used to implement the method as described in any one of claims 1 to 8.

Images