A battery state monitoring method and system

Abstract

The application discloses a battery state monitoring method and system, and relates to the technical field of battery energy storage; the method comprises the following steps: constructing a main observer and an auxiliary observer based on the same battery dynamic model; estimating the state of the battery state of charge through the two observers respectively, and determining the state of the measurement data of the sensor according to the state estimation difference value of the two; when the determined measurement data is in a normal state, the main observer outputs the current state estimation value; when the determined measurement data is in an abnormal state, the first observation confidence parameter is updated and adjusted according to the difference between the predicted values of the sensor measurement data of the two observers, and state estimation is performed again and output. The system is used to implement the above method. The application has the beneficial effects that the differentiated response of the two observers to the same sensor data is used as the quantitative criterion of the sensor reliability, and the instantaneous reliability mutation of the sensor data can be accurately identified.

Description

Technical Field

[0001] This application relates to the field of battery energy storage technology, and in particular to a battery state monitoring method and system. Background Technology

[0002] In battery energy storage power stations and industrial and commercial energy storage systems, the Battery Management System (BMS) is the core component for ensuring safe system operation and optimizing energy dispatch. The State of Charge (SOC), as a key state quantity reflecting the remaining energy of the battery, directly determines the reliability of safety protection functions such as overcharge and over-discharge protection based on its estimation accuracy.

[0003] To improve the adaptability of SOC estimation to sensor noise, existing technologies typically employ state observers based on adaptive algorithms, adjusting observer parameters online to address uncertainties caused by sensor anomalies. However, under the complex actual operating conditions of energy storage systems, sudden sensor failures and slow mismatches in the battery model generate residual signal characteristics similar to those produced within a single observer, making it difficult for algorithms to accurately distinguish the root cause of the failure.

[0004] In particular, under high transient current surges, the dynamic characteristics of the battery, such as polarization voltage and internal resistance, undergo drastic and rapid changes, exacerbating the instantaneous "mismatch" in the battery model. Simultaneously, to measure this transient current, the current sensor is highly susceptible to instantaneous faults such as measurement value jumps and drifts due to electromagnetic interference, core saturation, and other factors. At this point, the observer residual simultaneously contains both "the actual model dynamic error" and "the measurement error caused by sensor malfunction."

[0005] Such misjudgments can trigger a chain reaction: when a sensor actually malfunctions, the algorithm may misinterpret it as a model change, causing the observer to over-rely on the contaminated sensor data and output a severely skewed SOC estimate. In this case, the overcharge / over-discharge protection logic executed by the BMS based on this erroneous SOC will completely fail, potentially leading to thermal runaway due to continuous overcharging or permanent damage due to deep over-discharge, resulting in a serious safety incident. Conversely, when the model is normal but the sensor experiences a brief disturbance, the algorithm may misinterpret it as increased noise, causing the observer to unnecessarily reduce its trust in all data, making the SOC estimate abnormally smooth. This can cause the BMS protection action to be slow or even fail to activate when the battery actually reaches the safety boundary, similarly creating a safety hazard. Summary of the Invention

[0006] One objective of this application is to provide a battery state monitoring method that can solve at least one of the defects in the aforementioned background art.

[0007] Another objective of this application is to provide a system capable of implementing a battery state monitoring method that addresses at least one of the defects in the aforementioned background art.

[0008] To achieve at least one of the above objectives, one aspect of this application provides a battery state monitoring method, comprising the following steps: constructing a master observer and an auxiliary observer based on the same battery dynamic model, wherein the master observer uses a dynamically adjusted first observation confidence parameter, and the auxiliary observer uses a second observation confidence parameter that is less than the first observation confidence parameter and is a fixed value; acquiring sensor measurement data, inputting it to the master observer and the auxiliary observer respectively to obtain two sets of battery state of charge state estimates, and determining the state of the sensor measurement data based on the difference between the two sets of state estimates; if the sensor measurement data is determined to be in a normal state, the master observer maintains the current first observation confidence parameter and outputs the corresponding state estimate value; if the sensor measurement data is determined to be in an abnormal state, updating and adjusting the first observation confidence parameter based on the difference between the predicted values ​​of the sensor measurement data by the master observer and the auxiliary observer; and the master observer re-evaluating the state using the updated first observation confidence parameter and outputting the state estimate value.

[0009] Preferably, the specific process for determining the state of sensor measurement data based on the state estimation difference between the master observer and the auxiliary observer is as follows: construct a sensor anomaly detection statistic based on the state estimation difference between the master observer and the auxiliary observer; if the sensor anomaly detection statistic is greater than or equal to a set decision threshold, determine that the sensor measurement data is in an abnormal state; otherwise, determine that the sensor measurement data is in a normal state.

[0010] Preferably, the sensor anomaly detection statistics at time k. The construct expression is as follows: ; In the formula, This represents the difference in state estimates between the master observer and the auxiliary observer at time k. This represents the difference in state estimates at time k. The covariance matrix.

[0011] Preferably, in the battery state monitoring process that includes multiple consecutive monitoring cycles, a sensor anomaly detection statistic is constructed for each monitoring cycle; the sensor anomaly detection statistic is stored and statistically analyzed by setting a sliding window, thereby dynamically adjusting the decision threshold; the decision threshold at the current time k is... The expression to be adjusted is as follows: ; In the formula, Indicates the regulating factor. This represents the mean of the sensor anomaly detection statistics within the sliding window at time k. It represents the standard deviation of the sensor anomaly detection statistics within the sliding window at time k.

[0012] Preferably, the corresponding sensor anomaly detection statistics are included in the sliding window for data update only when the sensor measurement data is determined to be in a normal state, and are used for the adjustment of the decision threshold in the next monitoring cycle.

[0013] Preferably, the inverse of the measurement noise covariance matrix of the master observer is used as the first observation confidence parameter. When the sensor's measurement data is determined to be in an abnormal state, the specific process of updating and adjusting the first observation confidence parameter of the master observer is as follows: based on the difference between the predicted values ​​of the sensor's measurement data by the master observer and the auxiliary observer, the instantaneous estimate of the measurement noise covariance matrix of the master observer is calculated online; the first observation confidence parameter is updated based on the instantaneous estimate of the measurement noise covariance matrix.

[0014] Preferably, the instantaneous estimate of the measurement noise covariance matrix of the master observer at time k is... The expression is as follows: ; ; In the formula, This represents the difference between the predictions of the sensor's measurement data by the primary observer and the auxiliary observer at the current time k, calculated using a sliding window. The covariance matrix, This represents the component in the measurement prediction covariance matrix of the master observer at time k that is contributed by the uncertainty of state prediction. This represents the component in the measurement prediction covariance matrix of the auxiliary observer at time k that is contributed by the uncertainty of state prediction. Let N represent the fixed measurement noise covariance matrix of the auxiliary observer, N represent the sliding window length, and i represent the time index.

[0015] Preferably, when the sensor's measurement data is determined to be in an abnormal state, the first observation confidence parameter of the master observer at the current time k is... The expression to be updated is: ; When determining that the sensor's measurement data has recovered from an abnormal state to a normal state, the first observation confidence parameter of the master observer at the current time k is... The expression to be updated is: ; in, This represents the measurement prediction covariance matrix of the master observer at time k. and Both represent update factors, and: ; This represents the measurement prediction covariance matrix of the master observer at time k-1. This represents the instantaneous estimate of the measurement noise covariance matrix of the master observer at time k. This represents the nominal value of the measurement noise covariance matrix of the master observer.

[0016] Preferably, the battery safety protection mechanism is executed based on the state estimate output by the master observer, specifically including the following process: if the output state estimate is greater than or equal to a preset first overcharge protection threshold, or less than or equal to a preset first over-discharge protection threshold, the battery is determined to enter a warning state, a software alarm is generated, and the upper-level energy management system is notified to limit the power; if the output state estimate is greater than or equal to a preset second overcharge protection threshold, or less than or equal to a preset second over-discharge protection threshold, the battery is determined to enter an emergency protection state, and a hardware protection command is generated to drive the actuator to cut off the charging and discharging circuit; wherein, the first overcharge protection threshold is less than the second overcharge protection threshold, and the first over-discharge protection threshold is greater than the second over-discharge protection threshold.

[0017] Another aspect of this application provides a battery state monitoring system for implementing the above-described battery state monitoring method, comprising an energy storage unit, an energy storage converter, a state monitoring module, and a protection logic unit module; the energy storage unit includes at least one battery; the energy storage converter is connected between the energy storage unit and the power grid for performing charge and discharge control; the state monitoring module is used to construct a master observer and an auxiliary observer, perform state determination on sensor measurement data based on the state estimates of the master observer and the auxiliary observer, and perform normal or re-output of the state estimate of the master observer based on the state determination result; the protection logic unit module performs overcharge protection and / or over-discharge protection for the battery based on the state estimate output by the state monitoring module.

[0018] Compared with the prior art, the beneficial effects of this application are as follows: (1) This application constructs a dual observer architecture with different confidence parameters and uses the differential response of the two observers to the same sensor data as a quantitative criterion for sensor confidence. It can accurately identify the instantaneous confidence change of sensor data and avoid the protection logic disorder caused by misjudging model error as sensor failure or misjudging sensor failure as increased noise.

[0019] (2) When the sensor data is determined to be abnormal, this application adjusts the observation confidence parameter of the master observer in real time according to the prediction difference between the two observers, actively reduces the trust weight of the abnormal data, so that the output of the master observer remains physically reasonable during the sensor interference and does not produce non-physical drastic changes.

[0020] (3) In adjusting the first observation confidence parameter of the master observer, this application uses the difference in predicted values ​​of the two observers for the sensor measurement data, rather than the difference in their state estimates. Since the two observers are based on the exact same battery dynamic model, this difference in predicted values ​​naturally cancels out the influence of common model dynamics and process noise, and can more purely reflect the discrepancies in the interpretation of the same sensor data caused by different confidence settings. This makes the adjustment of the first observation confidence parameter more accurate, avoids the interference of model mismatch in parameter adjustment, and thus achieves a precise and rapid response when the sensor is abnormal. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the working steps of the battery state monitoring method in this application.

[0022] Figure 2 This is a schematic diagram of the overall workflow of the battery state monitoring method in this application.

[0023] Figure 3 This is a schematic diagram of the equivalent circuit of the battery in this application.

[0024] Figure 4 This is a schematic diagram of the battery state monitoring system in this application. Detailed Implementation

[0025] The present application will now be further described in conjunction with specific embodiments. It should be noted that, in the description of this specification, the use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicates that the specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0026] In the description of this application, it should be noted that the terms "center", "lateral", "longitudinal", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc., which indicate the orientation and positional relationship based on the orientation or positional relationship shown in the accompanying drawings, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and should not be construed as limiting the specific protection scope of this application.

[0027] It should be noted that the terms "first," "second," etc., in the specification and claims of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0028] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "joining," and "fixing," etc., should be interpreted broadly. For example, they can refer to a connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0029] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0030] The terms “comprising” and “having”, and any variations thereof, in the specification and claims of this application are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0031] To facilitate understanding of the technical solution of this application, the application scenarios of this application will be briefly described below.

[0032] Specifically, the technical solution of this application can be applied to electrochemical energy storage systems, including but not limited to lithium battery energy storage power stations, industrial and commercial energy storage systems, and residential energy storage systems. The energy storage system architecture includes energy storage units, a battery management system (BMS), an energy storage converter (PCS), and a power grid. The energy storage unit includes one or more batteries for storing and releasing electrical energy; the energy storage converter is connected between the energy storage unit and the power grid to perform charge and discharge control; the battery management system is communicatively connected to the energy storage unit to obtain the terminal voltage and charge / discharge current of the energy storage unit in real time and output control commands to achieve safe charging and discharging operation of the energy storage system.

[0033] In actual operation of energy storage systems, the terminal voltage and charging / discharging current of the energy storage units are detected by voltage and current sensors. These sensors are susceptible to electromagnetic interference, loose connections, temperature drift, and core saturation, which can cause instantaneous spikes or persistent deviations. If such abnormal sensor data is not effectively identified and isolated, it will directly lead to non-physical, drastic fluctuations in the SOC estimate within the BMS, thereby incorrectly triggering overcharge or over-discharge protection, or masking the true dangerous state and preventing it from triggering protection. Both of these actions pose serious risks to battery safety.

[0034] One aspect of this application provides a battery state monitoring method, such as... Figure 1 and Figure 2 As shown, one preferred embodiment includes the following steps: S100: Construct a master observer and an auxiliary observer based on the same battery dynamic model. The master observer uses a dynamically adjusted first observation confidence parameter, while the auxiliary observer uses a second observation confidence parameter that is less than the first observation confidence parameter and is a fixed value.

[0035] It should be understood that an observer refers to an algorithm or module that estimates the internal state of a system based on its input and output data; the observation confidence parameter refers to a parameter characterizing the observer's level of trust in the sensor measurement data. This step lays the architectural foundation for subsequent sensor anomaly detection and parameter adaptive adjustment by constructing two observers with different observation confidence parameters: the differentiated responses of the two observers to the same sensor data constitute the data basis and signal source of the technical solution in this application.

[0036] S200: Acquire sensor measurement data, input it to the main observer and auxiliary observer respectively to obtain two sets of battery state of charge state estimates, and determine the state of the sensor measurement data based on the difference between the two sets of state estimates.

[0037] It should be understood that state estimation refers to the process by which an observer calculates the estimated values ​​of the internal state variables of a system based on its inputs and outputs using a recursive algorithm. The state estimation difference value refers to the vector difference between the state estimates output by the primary observer and the state estimates output by the auxiliary observer at the same time.

[0038] Since both observers are based on the same battery dynamics model, the difference primarily reflects their different interpretations of the same sensor data due to varying confidence levels. When the sensor is operating normally, this difference remains within a biased but statistically stationary range; when the sensor malfunctions, the difference exhibits significant abrupt changes. This step, by calculating the state estimation difference between the two observers in real time, provides reliable decision-making data for triggering subsequent anomaly protection mechanisms.

[0039] S300: If the sensor's measurement data is determined to be in a normal state, the master observer maintains the current first observation confidence parameter and outputs the corresponding state estimate.

[0040] It should be understood that the sensor's measurement data being in a normal state means that the sensor's current output measurement value is within its nominal accuracy range, and is not significantly affected by factors such as electromagnetic interference, poor contact, temperature drift, or core saturation. The measurement data can accurately reflect the actual physical state of the battery. This step, by maintaining a high confidence parameter of the master observer when the sensor is normal, ensures that the SOC estimate has excellent dynamic tracking accuracy and response speed under normal operating conditions, avoiding additional lag or smoothing introduced by unnecessary parameter adjustments, and guaranteeing the accuracy and economy of energy dispatching during normal operation of the energy storage system.

[0041] S400: If the sensor's measurement data is determined to be in an abnormal state, the first observation confidence parameter is updated and adjusted based on the difference between the predicted values ​​of the sensor's measurement data by the main observer and the auxiliary observer.

[0042] It should be understood that an abnormal state of sensor measurement data refers to the sensor's current output measurement value being affected by factors such as electromagnetic interference, poor contact, temperature drift, and core saturation, resulting in instantaneous spikes or continuous deviations from the true value. Prediction discrepancy refers to the difference between the primary observer's prediction of the sensor measurement data and the auxiliary observer's prediction of the same measurement data at the same moment.

[0043] This step uses the difference in predictions from two observers as the basis for parameter adjustment, abandoning the traditional approach of relying on the residuals of a single observer. Since the two observers are based on the exact same battery dynamics model, the difference in predictions naturally cancels out the influence of shared model dynamics and process noise. Therefore, it can more purely capture and quantify the interpretation discrepancies caused by different sensor confidence settings, making the adjustment of the first observation confidence parameter more accurate and avoiding interference from model mismatch in parameter adjustment.

[0044] S500: The master observer re-estimates the state using the updated first observation confidence parameter and outputs the state estimate.

[0045] Understandably, this application constructs a dual-observer architecture with different confidence parameters and uses the differentiated responses of the two observers to the same sensor data as a quantitative criterion for sensor confidence. This can accurately identify instantaneous changes in the confidence of sensor data and fundamentally avoid the protection logic disorder caused by misjudging model errors as sensor faults or misjudging sensor faults as increased noise.

[0046] Meanwhile, when determining that the sensor data is abnormal, this application adjusts the observation confidence parameter of the master observer in real time based on the prediction difference between the two observers, actively reducing the trust weight of the abnormal data, so that the output of the master observer remains physically reasonable during sensor interference and does not produce non-physical drastic jumps.

[0047] In a specific embodiment, various battery dynamic models can be used to construct the master and auxiliary observers, such as electrochemical models, equivalent circuit models, data-driven models, and thermo-electric coupling models. The specific construction processes of these battery dynamic models are well-known techniques to those skilled in the art; for ease of understanding, the dynamic characteristics of the battery will be described below using a second-order RC equivalent circuit model as an example.

[0048] Specifically, such as Figure 3 As shown, the second-order RC equivalent circuit model uses an open-circuit voltage source V OCV The battery's terminal voltage response is characterized by two RC polarization networks: one composed of an ohmic internal resistance R0, and the other composed of an RC polarization network composed of an RC resistor R1 and a capacitor C1, and the other composed of an RC polarization network composed of an RC resistor R2 and a capacitor C2. The open-circuit voltage V0 is also used. OCV There is a definite nonlinear mapping relationship between V and the battery's state of charge (SOC). This mapping relationship can be obtained in advance through offline calibration using pulse charge-discharge experiments combined with least squares fitting methods. In the figure, V... t I represents the battery's terminal voltage. L This indicates the charging current or discharging current.

[0049] Based on this equivalent circuit model, a system dynamic equation with SOC and polarization voltage as state variables can be established, serving as a common battery dynamic model for state estimation by both the master and auxiliary observers. The identification of model parameters can be performed online or offline using methods known in the field, such as recursive least squares or recursive least squares with a forgetting factor.

[0050] In a specific embodiment, when executing step S100, the observer may specifically employ Kalman filtering and its variants, such as Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), and Capacitive Kalman Filter (CKF). These algorithms are all recursive estimation algorithms well-known in the art, and their core idea is to achieve optimal estimation of the system state through two steps of "prediction-correction" under the minimum mean square error criterion. In this embodiment, the Kalman filtering algorithm is preferably used for constructing the main observer and auxiliary observer, and subsequent descriptions will also use the Kalman filtering algorithm framework as an example.

[0051] In a specific embodiment, when performing step S100, the first observation confidence parameter used by the master observer is a parameter that can be dynamically adjusted online; this parameter characterizes the degree of trust that the master observer has in the current sensor measurement data, and the larger the value, the higher the degree of trust; during system initialization, the first observation confidence parameter is set to a high initial value.

[0052] The auxiliary observer employs a fixed, relatively conservative second observation confidence parameter. This second observation confidence parameter is typically set to a multiple of the initial value of the first observation confidence parameter corresponding to the primary observer; the multiple can be determined by those skilled in the art based on their specific needs. For example, the second observation confidence parameter is 1 / 5 to 1 / 3 of the initial value of the first observation confidence parameter. This setting of the second observation confidence parameter implies that the auxiliary observer is more skeptical of any sensor data, and its state estimation relies more heavily on the predictive dynamics of the battery dynamic model itself. Consequently, its estimation results are more robust in the event of sensor anomalies.

[0053] Understandably, the statistical characteristics of noise contained in sensor measurement data can be represented by the measurement noise covariance matrix. The value of the measurement noise covariance matrix directly reflects the observer's level of confidence in the sensor measurement data; a larger value indicates a higher noise component in the sensor measurement data. Within the Kalman filter algorithm framework, the first observation confidence parameter can be set as the inverse of the measurement noise covariance matrix of the master observer; similarly, the second observation confidence parameter can be set as the inverse of the fixed measurement noise covariance matrix of the auxiliary observer.

[0054] In a specific embodiment, when performing step S200, as follows: Figure 2As shown, the specific process of determining the state of sensor measurement data based on the state estimation difference between the master observer and the auxiliary observer is as follows: Based on the state estimation difference between the master observer and the auxiliary observer, a sensor anomaly detection statistic is constructed. If the sensor anomaly detection statistics If the sensor's measurement data is greater than or equal to the set decision threshold δ, it is determined that the sensor's measurement data is in an abnormal state; otherwise, it is determined that the sensor's measurement data is in a normal state.

[0055] Understandably, at the current time k, if the sensor is operating normally, the first observation confidence parameter of the master observer is high, allowing for full trust and utilization of the sensor's measurement data to quickly and accurately correct its state prediction, outputting a state estimate of SOC that has excellent dynamic performance and closely approximates the true value. The second observation confidence parameter of the auxiliary observer is low, indicating excessive skepticism towards the sensor's measurement data. This results in a small update gain, leading to poor state estimates. There is a certain degree of lag and smoothing. Since the two observers use the same battery dynamics model, their estimation differences under normal conditions mainly stem from steady-state bias caused by different levels of confidence in the same accurate data. This bias will remain within a biased but statistically stationary range.

[0056] At time k, if the sensor is in an abnormal operating condition, the primary observer, maintaining a high first observation confidence parameter, will over-rely on the abnormal measurement data from the sensor, leading to an incorrect state estimate. A non-physical, drastic change occurs. Conversely, the auxiliary observer, due to its fixed and conservative second observation confidence parameter, is less sensitive to sensor anomalies in measurement data, and its state estimate... The changes are relatively gradual, better reflecting the system's inertial trend. Also, since both observers use the same battery dynamic model, when an anomaly occurs, the common model dynamic components in the real-time deviation of their state estimates cancel each other out. Therefore, the deviation signal presents as a significant pulse highly specific to sensor anomalies.

[0057] Therefore, this embodiment monitors the difference in state estimates output by the primary observer and the auxiliary observer in real time, and constructs a quantitative sensor health index, namely the sensor anomaly detection statistic, based on this difference. This statistic can diagnose sensor anomalies in real time, thereby determining whether the first observation confidence parameter setting of the master observer matches the actual operating conditions, providing a key criterion for subsequent online dynamic adjustments and safety decisions. For ease of understanding, the sensor anomaly detection statistic will be explained below. The specific construction process will be described in detail.

[0058] In a specific example, after the master observer and the auxiliary observer complete the state recursion in parallel during each monitoring period, they output their respective state estimates and their estimation error covariance matrix. The state estimate of the master observer at time k can be denoted as... The corresponding estimation error covariance matrix is The state estimate of the auxiliary observer is... The corresponding estimation error covariance matrix is Then, for the sensor anomaly detection statistics corresponding to the current time k... The construction process is as follows: First, calculate the difference in state estimates between the two observers at time k. The specific calculation expression is as follows: .

[0059] Then, since the two observers operate independently, their estimation errors are uncorrelated, therefore the state estimation difference value... covariance matrix The expression is: .

[0060] Where E(·) represents the expectation.

[0061] Finally, in order to quantify the state estimation difference value The significance of the difference in state estimates can be used to determine the difference in state estimates. and its covariance matrix By combining the results, a normalized statistic for sensor anomaly detection is constructed. The specific expression is: .

[0062] It is understandable that, based on the above-mentioned sensor anomaly detection statistics... From the expression, we can see that this statistic Mathematically, this is the square of the Mahalanobis distance. The larger the value, the greater the discrepancy between the estimates of the primary and secondary observers exceeds the range of random fluctuations, indicating a lower reliability of the sensor measurement data. Specifically, this can be determined by the decision threshold at the current time k. Make a judgment, that is, the sensor anomaly detection statistics. ≥ If the sensor's measurement data is found to be in an abnormal state, it is determined that the sensor's measurement data is in a normal state.

[0063] It is important to understand that the statistical characteristics of sensor measurement data may slowly change with factors such as battery aging, ambient temperature variations, and switching of operating conditions. When judging the state of sensor measurement data, using a fixed decision threshold may result in frequent false alarms due to a threshold that is too small, or missed alarms due to a threshold that is too large. Therefore, in this embodiment, it is preferable to use a sliding window to statistically analyze historical data, thereby achieving dynamic adaptive adjustment of the decision threshold to cope with the slow changes in system operating conditions. For ease of understanding, the dynamic adaptive adjustment process of the decision threshold will be described in detail below.

[0064] In one specific embodiment, during battery state monitoring encompassing multiple consecutive monitoring cycles, a sensor anomaly detection statistic is constructed for each monitoring cycle. A sliding window is used to store the sensor anomaly detection statistic values ​​for the most recent M monitoring cycles, and the decision threshold is dynamically adjusted based on the statistical characteristics of the data within the window.

[0065] Specifically, at time k, when the sensor measurement data is normal, the difference in state estimates between the primary observer and the auxiliary observer is... It will fluctuate within a statistical range, causing the sensor anomaly detection statistics to... Small and stable, i.e., sensor anomaly detection statistics mean and standard deviation It stabilizes at a low level. When abnormal sensor measurement data leads to changes in the state estimate of the master observer... State estimation difference during anomalous transitions The number of abnormal sensor detection statistics will increase dramatically. This will produce a peak value far exceeding the normal fluctuation range, and this peak value is a strong characteristic of anomalies. Therefore, it is possible to use sensor-based anomaly detection statistics... Decision threshold dynamically adjusted based on statistical characteristics Designed as follows: .

[0066] .

[0067] .

[0068] In the formula, This represents the adjustment factor, whose value determines the decision threshold. The degree of leniency; that is, the regulatory factor. When the value is small, the decision threshold is determined. More stringent, more sensitive to anomalies but may increase false alarm rate; regulation factor When the value is large, the decision threshold is determined. A wider range of adjustment factors makes it less sensitive to anomalies but may increase the false negative rate. In practical applications, the adjustment factor... The appropriate value range can be selected according to the system's security level requirements and sensor characteristics. The typical value range is 2 to 4, and the specific value can be determined by those skilled in the art based on their actual needs. Let M represent the sensor anomaly detection statistics for the j-th monitoring period, and M represent the length of the sliding window.

[0069] It is important to note that when using a sliding window to dynamically adjust the decision threshold, if the sensor's measurement data remains abnormal, the values ​​of the sensor anomaly detection statistics during the abnormal period will be significantly higher. Including all these abnormal values ​​in the sliding window would contaminate the calculation basis of the decision threshold, i.e., it would raise the mean and standard deviation of the sensor anomaly detection statistics within the sliding window, causing the decision threshold to be abnormally raised. This would reduce the system's sensitivity to subsequent anomalies and may even lead to missed detections. Therefore, in this embodiment, the corresponding sensor anomaly detection statistics are only included in the sliding window for data updates when the sensor's measurement data is determined to be in a normal state, for use in adjusting the decision threshold for the next monitoring cycle. This mechanism ensures that the decision threshold is always established and maintained based on clean, normal operating condition data. Even during periods of continuous sensor anomaly, the sliding window still stores normal historical data from before the anomaly occurred, preventing the decision threshold from being contaminated and raised by abnormal data. This ensures that the system maintains its sensitivity to subsequent anomalies even during periods of continuous anomaly.

[0070] Based on the sensor anomaly detection statistics constructed at the current time k, as described above. and decision threshold The logic for determining the state of the sensor's measurement data is as follows: If < If the sensor's measurement data is determined to be normal, the system maintains the normal operating state of step S300; if ≥ The system determines that the sensor's measurement data is abnormal, triggering the protection mechanism in step S400. The core objective of this protection mechanism is to rapidly and intelligently adjust the first observation confidence parameter of the master observer online, drastically reducing its trust weight in the current abnormal sensor data. This isolates the impact of abnormal data on battery SOC estimation during the duration of the sensor's abnormal data, maintaining the stability and reliability of the state output. For ease of understanding, the specific implementation process of step S400 is described below.

[0071] In a specific embodiment, when performing step S400, as follows: Figure 2As shown, when the sensor's measurement data is determined to be in an abnormal state, the specific process of updating and adjusting the first observation confidence parameter of the master observer is as follows: based on the difference between the predicted values ​​of the sensor's measurement data by the master observer and the auxiliary observer, the instantaneous estimate of the measurement noise covariance matrix of the master observer is calculated online; the first observation confidence parameter is updated based on the instantaneous estimate of the measurement noise covariance matrix.

[0072] As can be understood from the foregoing, within the Kalman filtering framework, the inverse of the measurement noise covariance matrix of the master observer is the corresponding first observation confidence parameter. Therefore, the first observation confidence parameter can be updated by solving for the instantaneous estimate of the measurement noise covariance matrix. Regarding the calculation of the instantaneous estimate of the master observer's measurement noise covariance matrix, since both the master and auxiliary observers use the same battery dynamics model, the difference between their predictions naturally cancels out the shared uncertainties of the system dynamics and process model. This allows for the capture and quantification of interpretation discrepancies arising from different sensor confidence settings, i.e., the instantaneous confidence of the sensor data. Therefore, through the statistical characteristics of the difference between the two observers' predictions, an instantaneously desired observer parameter adjustment can be calculated online, i.e., the required instantaneous estimate of the master observer's measurement noise covariance matrix. For ease of understanding, the specific calculation process of the instantaneous estimate of the master observer's measurement noise covariance matrix will be described in detail below.

[0073] In a specific example, the sensor measurement data at time k can be defined as follows: This includes the battery's state of charge (SOC) terminal voltage and charge / discharge current values; the master observer's prediction of the sensor's measurement data is... The auxiliary observer's prediction of the sensor's measurement data is... The difference between the predictions of the primary observer and the secondary observer is... The calculation expression is: .

[0074] Since both observers use the same battery dynamics model, the predicted values ​​differ. By removing the influence of common model dynamics and process noise, it is insensitive to changes in the battery dynamic model and can more purely reflect the discrepancies in the interpretation of the same measurement data caused by different observation confidence settings. Therefore, it has a very high indicative value for sensor anomalies.

[0075] Subsequently, during the state recursion process of the two observers, the measurement prediction covariance matrix of the master observer and the auxiliary observer can be constructed. and The specific expression is as follows: ; .

[0076] In the formula, This represents the component in the measurement prediction covariance matrix of the master observer at time k that is contributed by the uncertainty of state prediction. This represents the measurement prediction covariance matrix of the master observer at time k. This represents the component in the measurement prediction covariance matrix of the auxiliary observer at time k that is contributed by the uncertainty of state prediction. This represents the fixed measurement noise covariance matrix of the auxiliary observer.

[0077] It is important to know that the measurement prediction covariance matrix is ​​a quantitative description of the uncertainty of the observer's prediction of the measurement data. The diagonal elements of this matrix represent the variance of each measurement data prediction, and the off-diagonal elements represent the covariance between different measurement data predictions. and The calculation process involves mapping the uncertainty of state prediction to the observation space through the observation model. The specific calculation method depends on the type of observer. For example, in the Extended Kalman Filter (EKF) framework, the Jacobian matrix is ​​used for approximation through linearization. In the Unscented Kalman Filter (UKF) and Capacitive Kalman Filter (CKF) frameworks, deterministic sampling is used, and the covariance is obtained by statistically analyzing the nonlinear propagation through Sigma points or capacitive points, respectively.

[0078] Since the two observers operate independently and the measurement update steps introduce different gains, their measurement predictions are also uncorrelated. Therefore, the difference in predictions... Theoretical covariance matrix for: .

[0079] Additionally, the difference between the predictions of the primary observer and the auxiliary observer at time k can be calculated using a sliding window. Real-time sample covariance matrix The specific expression is as follows: .

[0080] In the formula, N represents the sliding window length, and i represents the time index.

[0081] At the current time k, based on the difference in predicted values Theoretical covariance matrix With real-time sample covariance matrix Equal values ​​yield the measurement noise covariance matrix of the master observer. instantaneous estimate The specific expression is as follows: .

[0082] It should be understood that after obtaining the instantaneous estimate of the measurement noise covariance matrix of the master observer, the first observation confidence parameter used by the master observer can be updated. To ensure system stability while rapidly suppressing anomalies, an event-driven, rate-controllable smooth update strategy is adopted for the first observation confidence parameter. Event-driven means that the update of the first observation confidence parameter is not triggered according to a fixed period, but rather based on specific events, such as the determination of sensor data anomalies or the occurrence of a recovery from anomalies to normal. Rate-controllable means that the update rate of the first observation confidence parameter can be independently adjusted by the magnitude of the update factor. For ease of understanding, the specific update process of the first observation confidence parameter will be described in detail below.

[0083] In a specific embodiment, such as Figure 2 As shown, at the current time k, if the sensor's measurement data is determined to be in an abnormal state, a larger update factor can be used to adjust the first observation confidence parameter of the master observer. To perform the update, the specific update expression is: .

[0084] In the formula, This represents the measurement prediction covariance matrix of the master observer at time k. Indicates the update factor, and ; This represents the measurement prediction covariance matrix of the master observer at time k-1. This represents the instantaneous estimate of the measurement noise covariance matrix of the master observer at time k. .

[0085] At time k, if it is determined that the sensor's measurement data has recovered from an abnormal state to a normal state, a smaller update factor can be used to adjust the first observation confidence parameter of the master observer. To perform the update, the specific update expression is: .

[0086] in, Let represent the update factor, and we have: ; The nominal value of the measurement noise covariance matrix of the master observer is the optimal value obtained through offline calibration or a learning period under known healthy and stable operating conditions after system installation and commissioning. It represents the performance index of the sensor under ideal conditions.

[0087] It is important to note that the initial values ​​of the measurement prediction covariance matrix of the master observer should be calculated. When setting it up, you can configure it. Therefore, when the sensor's measurement data remains normal, the measurement prediction covariance matrix of the primary observer remains stable. When sensor measurement data is abnormal, it quickly adjusts using its instantaneous observation estimate. After the abnormal measurement is recovered, the measurement prediction covariance moment of the master observer gradually recovers to normal. This design ensures that the system can eventually recover to its optimal estimation performance. It guarantees that after anomaly removal, the system can smoothly rebuild trust in high-precision data and restore optimal estimation performance, while avoiding parameter oscillations near critical states.

[0088] It is important to know that after completing the first observation confidence parameter... After the update, the master observer is based on the updated first observation confidence parameter. It recursively calculates the battery's state of charge and outputs a robust state estimate.

[0089] It should be understood that in battery management systems, overcharge / over-discharge protection mechanisms are used to prevent irreversible damage or even thermal runaway of the battery. Existing battery management systems' overcharge / over-discharge protection logic relies entirely on the accuracy of the battery's state of charge (SOC) estimation. If the sensor experiences electromagnetic interference or hardware failure leading to measurement distortion, the overcharge / over-discharge protection logic based on the incorrect SOC value will completely fail. Therefore, in the technical solution of this application, a safety protection mechanism can be implemented based on the state estimate output by the primary observer when the sensor measurement data is normal, or the state estimate output after updating when the sensor measurement data is abnormal. For ease of understanding, a detailed description will follow.

[0090] In a specific embodiment, such as Figure 2 As shown, the battery safety protection mechanism is executed based on the state estimate output by the master observer, specifically including the following process: First, overcharge protection thresholds and over-discharge protection thresholds can be set. The overcharge protection thresholds include a first overcharge protection threshold and a second overcharge protection threshold, with the first overcharge protection threshold being smaller than the second overcharge protection threshold. Similarly, the over-discharge protection thresholds include a first over-discharge protection threshold and a second over-discharge protection threshold, with the first over-discharge protection threshold being larger than the second over-discharge protection threshold. The specific values ​​of the overcharge and over-discharge protection thresholds can be set according to the actual needs of those skilled in the art. For example, the first overcharge protection threshold can be 95% of the nominal capacity, and the second overcharge protection threshold can be 100% of the nominal capacity; the first over-discharge protection threshold can be 20% of the nominal capacity, and the second over-discharge protection threshold can be 5% of the nominal capacity.

[0091] Then, a logical judgment can be made by comparing the real-time state estimate output by the master observer with the corresponding threshold. The specific judgment process is as follows: if the output state estimate is greater than or equal to the preset first overcharge protection threshold, or less than or equal to the preset first over-discharge protection threshold, the battery is determined to enter a warning state, a software alarm is generated, and the upper-level energy management system is notified to limit the power. If the output state estimate is greater than or equal to the preset second overcharge protection threshold, or less than or equal to the preset second over-discharge protection threshold, the battery is determined to enter an emergency protection state, and a hardware protection command is generated to drive the actuator to cut off the charging and discharging circuit; the actuator includes a main circuit contactor, a fuse, etc.

[0092] Understandably, this embodiment employs a two-level protection threshold to achieve a graded and progressive safety protection strategy: during the warning stage, power limiting is implemented through software to avoid the economic losses of direct shutdown; during the emergency stage, the circuit is forcibly disconnected through hardware to ensure absolute safety. More importantly, since the state estimation value input into the battery management system already possesses the characteristics of "high accuracy under normal conditions and high robustness under abnormal conditions" in the aforementioned steps, even in the worst-case scenario of sensor failure, the protection logic can still make correct judgments based on physically continuous and reasonable state estimation values. This fundamentally eliminates missed or false protection caused by front-end sensing distortion, and prevents erroneous emergency shutdowns triggered by false high or low signals in the state estimation value, ensuring the continuous and stable operation of the energy storage power station and improving power supply reliability.

[0093] Another aspect of this application provides a battery state monitoring system for implementing the above-described battery state monitoring method, such as... Figure 4 As shown, one preferred embodiment includes an energy storage unit, an energy storage converter, a state monitoring module, and a protection logic unit module. The energy storage unit includes at least one battery; the energy storage converter is connected between the energy storage unit and the power grid for performing charge and discharge control; the state monitoring module is used to construct a master observer and an auxiliary observer, perform state determination on the sensor measurement data based on the state estimates of the master observer and the auxiliary observer, and perform normal or re-output of the state estimate of the master observer based on the state determination results; the protection logic unit module performs overcharge protection and / or over-discharge protection for the battery based on the state estimate output by the state monitoring module.

[0094] Understandably, the state monitoring module and the protection logic unit module are built into the battery management system. The signal flow between the state monitoring module and the protection logic unit module is as follows: The main observer and auxiliary observer in the state monitoring module receive measurement data from the voltage sensor and current sensor in real time. After processing by steps S100 to S500 of the battery state monitoring method described above, a robust state estimate is output to the protection logic unit module. The protection logic unit module receives this state estimate and, based on its comparison with the overcharge protection threshold and / or over-discharge protection threshold, executes corresponding protection actions; that is, in the early warning state, a software alarm is generated and the EMS is notified to perform power limiting; in the emergency protection state, a hardware protection command is generated to drive the actuator to cut off the charging and discharging circuit.

[0095] This system, through the coordinated operation of the state monitoring module and the protection logic unit module, achieves a complete closed loop from sensor signal acquisition, sensor reliability assessment, adaptive adjustment of observation confidence parameters, robust state estimation output, and safety protection logic judgment. When sensors are functioning normally, the system provides high-precision SOC estimation for accurate energy scheduling and protection prediction. When sensors malfunction, it maintains the physical rationality and stability of the SOC estimation to ensure the correct execution of protection actions, thereby significantly improving the intrinsic safety level of the energy storage system.

[0096] The basic principles, main features, and advantages of this application have been described above. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely the principles of this application. Various changes and modifications can be made to this application without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claims. The scope of protection claimed by this application is defined by the appended claims and their equivalents.

Claims

1. A battery state monitoring method, characterized in that, Includes the following steps: The same battery dynamic model is used to construct a master observer and an auxiliary observer. The master observer uses a dynamically adjusted first observation confidence parameter, while the auxiliary observer uses a second observation confidence parameter that is smaller than the first observation confidence parameter and is a fixed value. The sensor measurement data is acquired and input into the main observer and auxiliary observer respectively to obtain two sets of battery state of charge state estimates. The sensor measurement data is then used to determine the state based on the difference between the two sets of state estimates. If the sensor's measurement data is determined to be in a normal state, the master observer maintains the current first observation confidence parameter and outputs the corresponding state estimate. If the sensor's measurement data is determined to be in an abnormal state, the first observation confidence parameter is updated and adjusted based on the difference between the predicted values ​​of the sensor's measurement data by the main observer and the auxiliary observer. The master observer re-estimates the state using the updated first observation confidence parameter and outputs the state estimate.

2. The battery state monitoring method as described in claim 1, characterized in that, The specific process of determining the state of sensor measurement data based on the state estimation difference between the master observer and the auxiliary observer is as follows: Based on the state estimation differences between the master observer and the auxiliary observer, a sensor anomaly detection statistic is constructed. If the sensor anomaly detection statistic is greater than or equal to the set decision threshold, the sensor's measurement data is determined to be in an abnormal state; otherwise, the sensor's measurement data is determined to be in a normal state.

3. The battery state monitoring method as described in claim 2, characterized in that, Sensor anomaly detection statistics at time k The construct expression is as follows: ； In the formula, This represents the difference in state estimates between the master observer and the auxiliary observer at time k. This represents the difference in state estimates at time k. The covariance matrix.

4. The battery state monitoring method as described in claim 2, characterized in that, In the battery status monitoring process, which includes multiple consecutive monitoring cycles, a sensor anomaly detection statistic is constructed for each monitoring cycle. By setting a sliding window to store and statistically analyze sensor anomaly detection data, the decision threshold can be dynamically adjusted; the decision threshold at the current time k is... The expression to be adjusted is as follows: ； In the formula, Indicates the regulating factor. This represents the mean of the sensor anomaly detection statistics within the sliding window at time k. It represents the standard deviation of the sensor anomaly detection statistics within the sliding window at time k.

5. The battery state monitoring method as described in claim 4, characterized in that, Only when the sensor's measurement data is determined to be in a normal state, the corresponding sensor anomaly detection statistics are included in the sliding window for data update, which is used for the adjustment of the decision threshold in the next monitoring cycle.

6. The battery state monitoring method according to any one of claims 1-5, characterized in that, The inverse of the measurement noise covariance matrix of the master observer is used as the first observation confidence parameter; When the sensor's measurement data is determined to be in an abnormal state, the instantaneous estimate of the measurement noise covariance matrix of the master observer is calculated online based on the difference between the predicted values ​​of the sensor's measurement data by the master observer and the auxiliary observer. The confidence parameter of the first observation is updated based on the instantaneous estimate of the measurement noise covariance matrix.

7. The battery state monitoring method as described in claim 6, characterized in that, The instantaneous estimate of the measurement noise covariance matrix of the master observer at time k. The expression is as follows: ； ； In the formula, This represents the difference between the predictions of the sensor's measurement data by the primary observer and the auxiliary observer at the current time k, calculated using a sliding window. The covariance matrix, This represents the component in the measurement prediction covariance matrix of the master observer at time k that is contributed by the uncertainty of state prediction. This represents the component in the measurement prediction covariance matrix of the auxiliary observer at time k that is contributed by the uncertainty of state prediction. Let N represent the fixed measurement noise covariance matrix of the auxiliary observer, N represent the sliding window length, and i represent the time index.

8. The battery state monitoring method as described in claim 7, characterized in that, When the sensor's measurement data is determined to be in an abnormal state, the first observation confidence parameter of the master observer at the current time k is... The expression to be updated is: ； When determining that the sensor's measurement data has recovered from an abnormal state to a normal state, the first observation confidence parameter of the master observer at the current time k is... The expression to be updated is: ； in, This represents the measurement prediction covariance matrix of the master observer at time k. and Both represent update factors, and: ; This represents the measurement prediction covariance matrix of the master observer at time k-1. This represents the instantaneous estimate of the measurement noise covariance matrix of the master observer at time k. This represents the nominal value of the measurement noise covariance matrix of the master observer.

9. The battery state monitoring method as described in claim 1, characterized in that, The battery safety protection mechanism is executed based on the state estimate output by the master observer, specifically including the following process: If the output state estimate is greater than or equal to the preset first overcharge protection threshold, or less than or equal to the preset first over-discharge protection threshold, the battery is determined to enter a warning state, a software alarm is generated, and the upper-level energy management system is notified to limit the power. If the output state estimate is greater than or equal to the preset second overcharge protection threshold, or less than or equal to the preset second over-discharge protection threshold, the battery is determined to enter an emergency protection state, and a hardware protection command is generated to drive the actuator to cut off the charging and discharging circuit. Among them, the first overcharge protection threshold is less than the second overcharge protection threshold, and the first over-discharge protection threshold is greater than the second over-discharge protection threshold.

10. A battery state monitoring system, used to implement the battery state monitoring method as described in any one of claims 1-9, characterized in that, include: Energy storage unit; The energy storage unit includes at least one battery; Energy storage converter; The energy storage converter is connected between the energy storage unit and the power grid to perform charge and discharge control; State monitoring module; The state monitoring module is used to construct a master observer and an auxiliary observer, determine the state of the sensor's measurement data based on the state estimates of the master observer and the auxiliary observer, and output the state estimate of the master observer normally or again based on the state determination results. as well as Protection logic unit module; The protection logic unit module performs overcharge protection and / or over-discharge protection for the battery based on the state estimation value output by the state monitoring module.

Images