A dynamic threshold-based energy storage system early warning method and system

Abstract

The embodiment of the application provides a kind of early warning method and system of energy storage system based on dynamic threshold, belong to safety monitoring technical field.The method includes: the multi-source operating data of energy storage system is collected, and monitoring data set is formed;Feature parameters are extracted based on the monitoring data set, and fusion feature vector is generated based on the extracted feature parameters;According to the fusion feature vector, dynamic threshold determination is executed, and the determined dynamic threshold is obtained;The safety state of the monitoring data set is evaluated based on the dynamic threshold, and the early warning result of energy storage system is output.The application scheme can realize more accurate identification and prompt in the early risk of energy storage system thermal runaway, thereby significantly improve the reliability and adaptability of early warning result, reduce the delay of operation and maintenance intervention and the probability of accident.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology, specifically to an early warning method and system for energy storage systems based on dynamic thresholds. Background Technology

[0002] With the construction and operation of large-scale energy storage power stations, the advantages of lithium batteries in terms of high energy density, long lifespan, and rapid response are gradually becoming apparent. However, their operation also brings potential safety hazards, especially thermal runaway and the resulting fires. Once thermal runaway occurs, the internal temperature of the battery will rise sharply, accompanied by the release of large amounts of harmful gases, which can easily trigger a chain reaction, causing the entire energy storage system to fail and even resulting in serious personal injury and property damage. Therefore, how to monitor the operating status of energy storage systems in real time and provide reliable early warnings in the early stages of thermal runaway is a crucial problem that the industry urgently needs to solve.

[0003] Existing safety monitoring and early warning technologies for energy storage systems suffer from the following shortcomings: Firstly, traditional monitoring strategies often rely on data from single sensors, such as single-point temperature measurement or single gas concentration detection. Due to the complex environment and high signal noise within the battery compartment, single parameters are prone to false alarms or missed alarms. For example, fluctuations in ambient temperature may cause abnormal output from temperature sensors, resulting in false alarms, while gas concentration sensors are also prone to drift when affected by local airflow, making it difficult to reflect the true risk. Secondly, existing solutions generally adopt a fixed threshold judgment mode, directly comparing monitoring data with a preset threshold to trigger an alarm. This static threshold method lacks adaptability when facing energy storage power stations with variable operating conditions and significant differences in external environments: setting the threshold too low will frequently trigger invalid alarms, increasing the maintenance burden; setting the threshold too high may miss early signs of thermal runaway, reducing system safety.

[0004] Furthermore, while some improved methods attempt to introduce multi-source data fusion, they still rely on fixed patterns in threshold setting and judgment mechanisms, failing to effectively consider the interrelationships and temporal evolution characteristics among multiple parameters, resulting in insufficient reliability of the assessment results. In summary, the existing safety early warning system for energy storage power stations still has significant shortcomings in the utilization of multi-source data and dynamic threshold adjustment, which is the main reason why it is difficult to simultaneously achieve stability and accuracy in the early warning system. Summary of the Invention

[0005] The purpose of this invention is to provide an early warning method and system for energy storage systems based on dynamic thresholds, so as to at least solve the problems of high false alarm rate and high risk of missed alarm caused by relying on fixed thresholds in the early warning process of existing energy storage systems.

[0006] To achieve the above objectives, the first aspect of the present invention provides an early warning method for energy storage systems based on dynamic thresholds. The method includes: collecting multi-source operating data of the energy storage system to form a monitoring dataset; extracting feature parameters based on the monitoring dataset and generating a fused feature vector based on the extracted feature parameters; performing dynamic threshold determination based on the fused feature vector to obtain a determined dynamic threshold; and performing a safety status assessment on the monitoring dataset based on the dynamic threshold, and outputting an early warning result for the energy storage system.

[0007] Optionally, the multi-source operational data includes any one or more of the following: gas monitoring data, temperature monitoring data, electrical monitoring data, humidity detection data, and pressure monitoring data inside the energy storage cabinet; the rules for forming the monitoring dataset are as follows: each monitoring data is sampled synchronously according to a unified time reference, the collected data is denoised and filtered, and after denoising and filtering, each monitoring data is standardized in units and calibrated in range; timestamp information and data source identifier are added to each calibrated monitoring data, and the data is stored in a preset data cache area to form a monitoring dataset.

[0008] Optionally, feature parameters are extracted based on the monitoring dataset, including: performing a sliding window calculation on the gas monitoring data to obtain the concentration surge rate; performing differential processing on the temperature monitoring data to obtain the temperature rise rate; performing a fluctuation detection algorithm on the electrical monitoring data to obtain the voltage fluctuation amplitude; performing a working condition correction operation based on the humidity detection data and the pressure monitoring data, substituting the deviation between the humidity detection data and the pressure monitoring data into a preset environmental correction model to obtain an environmental compensation factor for correcting the gas monitoring data and the temperature monitoring data; and using the concentration surge rate, the temperature rise rate, the voltage fluctuation amplitude, and the environmental compensation factor as the extracted feature parameters.

[0009] Optionally, generating a fused feature vector based on the extracted feature parameters includes: performing normalization processing on each feature parameter except for the environmental compensation factor, and aligning the feature parameters at different time scales based on a dynamic time warping algorithm after the normalization processing is completed; wherein, the dynamic time warping algorithm is a constraint band-based dynamic time warping algorithm, used to limit the offset range of the time matching path by setting a preset constraint band width when aligning feature parameters at different time scales; adjusting the weights of the other aligned feature parameters in the fusion process based on the environmental compensation factor to obtain the fused feature vector.

[0010] Optionally, dynamic threshold determination is performed based on the fused feature vector to obtain the determined dynamic threshold, including: extracting the corresponding gas concentration surge rate parameter and temperature rise duration parameter from the fused feature vector; establishing a two-factor criterion based on the concentration surge rate parameter and the temperature rise duration parameter; comparing the two-factor criterion with a preset reference threshold, and correcting the reference threshold when the comparison result meets the trigger condition to obtain the determined dynamic threshold; otherwise, when the comparison result does not meet the trigger condition, the preset reference threshold is used as the determined dynamic threshold.

[0011] Optionally, extracting the corresponding gas concentration surge rate parameter and temperature rise duration parameter from the fused feature vector includes: performing sliding window filtering on the gas monitoring component in the fused feature vector to eliminate instantaneous disturbances, and then calculating the slope of the filtered gas concentration curve to obtain the gas concentration surge rate parameter; setting a predefined time window for the temperature monitoring component in the fused feature vector and detecting the length of the continuous exceeding interval to determine the temperature rise duration parameter; the triggering condition is: if the gas concentration surge rate parameter exceeds a first preset rate threshold and the temperature rise duration parameter is greater than a second preset duration threshold within a consecutive preset sampling period, then the triggering condition is determined to be met.

[0012] Optionally, the rule for correcting the reference threshold is as follows: the environmental compensation factor in the extracted feature parameters is used as the gating parameter of a preset gated linear interaction model, and the gas concentration surge rate parameter and the temperature rise duration parameter are substituted into the gated linear interaction model to obtain the threshold correction amount; wherein, the preset gated linear interaction model is expressed as:

[0013]

[0014] in, This is the threshold correction amount obtained in the k-th sampling period; For environmental compensation gating factors; These are the coefficients obtained from the offline calibration of the exponential decay function model; The rate of sudden increase in gas concentration in the fused feature vector Rate threshold relative to the preset reference threshold Dimensionless excess; The duration of temperature rise in the fused feature vector The duration threshold relative to the preset reference threshold Dimensionless excess; The normalized value of the alignment residual for the kth sampling period is used; the preset reference threshold is adjusted based on the threshold correction amount to obtain the determined dynamic threshold.

[0015] Optionally, a safety status assessment of the monitoring dataset is performed based on the dynamic threshold, and an early warning result of the energy storage system is output, including: comparing the monitoring data with the dynamic threshold point by point within the assessment window to obtain the window average excess, generating a stable indicator based on stability and hysteresis criteria, recursively accumulating the historical assessment results using time decay weighting to form a risk score, and outputting the risk score and the maximum excess component as the assessment result; converting the assessment result into an early warning result containing a level identifier and a trigger source identifier according to the early warning level mapping table and outputting it.

[0016] Optionally, within the evaluation window, the monitoring data is compared point-by-point with the dynamic threshold to obtain the window average excess, and a stable indicator is generated based on stability and hysteresis criteria. A risk integral is formed by recursively accumulating historical evaluation results using time decay weighting, and the risk integral and the maximum excess component are output as the evaluation result. This includes: within the evaluation window, comparing the target component used for evaluation in the monitoring data set with the corresponding threshold in the dynamic threshold to obtain the corresponding excess sequence and calculating the window average excess; applying stability and hysteresis criteria to the window average excess, including counting the number of periods where excess is established within a continuous sampling period and setting a fallback threshold, generating a stable indicator to indicate whether a stable excess state has been entered; and recursively accumulating historical evaluation results using time decay weighting based on the stable indicator to obtain the risk integral at the current moment, and outputting the current risk integral and the target component that caused the maximum excess as the evaluation result.

[0017] A second aspect of the present invention provides an early warning system for energy storage systems based on dynamic thresholds. The system is applied to the aforementioned early warning method for energy storage systems based on dynamic thresholds. The system includes: a data acquisition unit for acquiring multi-source operational data of the energy storage system to form a monitoring dataset; a processing unit for extracting feature parameters from the monitoring dataset and generating a fused feature vector based on the extracted feature parameters; a threshold determination unit for performing dynamic threshold determination based on the fused feature vector to obtain a determined dynamic threshold; and an evaluation unit for performing a safety status evaluation on the monitoring dataset based on the dynamic threshold and outputting an early warning result for the energy storage system.

[0018] Through the above technical solution, this invention collects multi-source operational data from the energy storage system and forms a monitoring dataset, which comprehensively reflects the multi-dimensional state of the battery compartment during operation, including temperature, gas, and electrical parameters. Based on this, feature parameters are extracted and a fused feature vector is generated, achieving a unified representation of different types of monitoring signals and avoiding biases caused by single-parameter judgments. Furthermore, by performing dynamic threshold determination through the fused feature vector, the warning threshold can adaptively change with operating conditions and environmental conditions, overcoming the rigidity, false alarms, or missed alarms of fixed threshold methods. Finally, based on the dynamic threshold, a safety status assessment of the monitoring dataset is performed, and warning results are output. This enables the energy storage system to more accurately identify and alert to risks such as thermal runaway at an early stage, thereby significantly improving the reliability and adaptability of warning results and reducing operation and maintenance intervention delays and the probability of accidents.

[0019] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0021] Figure 1 This is a flowchart of the steps of an early warning method for energy storage systems based on dynamic thresholds provided in one embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the gas concentration surge rate parameter and dynamic threshold adjustment process provided in one embodiment of the present invention;

[0023] Figure 3 This is a schematic diagram of the temperature rise duration parameter and dynamic threshold adjustment process provided in one embodiment of the present invention;

[0024] Figure 4 This is a system structure diagram of an energy storage system early warning system based on dynamic thresholds provided in one embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0026] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0027] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0028] Figure 1 This is a flowchart of a method for early warning of energy storage systems based on dynamic thresholds, provided by one embodiment of the present invention. Figure 1 As shown, this invention provides an early warning method for energy storage systems based on dynamic thresholds, the method comprising:

[0029] Step S10: Collect multi-source operation data of the energy storage system to form a monitoring dataset.

[0030] Specifically, the multi-source operational data includes any one or more of the following: gas monitoring data, temperature monitoring data, electrical monitoring data, humidity detection data, and pressure monitoring data inside the energy storage cabinet; the rules for forming the monitoring dataset are as follows: each monitoring data is sampled synchronously according to a unified time reference, the collected data is denoised and filtered, and after denoising and filtering, each monitoring data is standardized in units and calibrated in range; timestamp information and data source identifiers are added to each calibrated monitoring data, and the data is stored in a preset data cache area to form a monitoring dataset.

[0031] In this embodiment of the invention, the internal operating conditions of the energy storage power station are complex during actual operation. Relying on a single parameter is often insufficient to accurately reflect the safety status of the battery cell. Therefore, in this embodiment, it is necessary to first collect multi-source operating data inside the energy storage system and use this data to form the monitoring dataset required for subsequent analysis.

[0032] Multi-source operational data refers to various monitoring signals that can reflect the internal environment of the energy storage cabinet and the operating status of the battery. Specifically, it includes, but is not limited to, any one or more of the following: gas monitoring data, temperature monitoring data, electrical monitoring data, humidity detection data, and pressure monitoring data.

[0033] Gas monitoring data is typically collected by sensors placed inside the battery compartment, reflecting the concentration trends of gases such as carbon monoxide and carbon dioxide; temperature monitoring data reflects the thermal state of the battery module and the surrounding air, especially in the early stage of thermal runaway; electrical monitoring data mainly includes parameters such as voltage and current, used to characterize the charging and discharging state and fluctuations of the battery; humidity and pressure data are mainly used to reflect changes in the internal environment of the energy storage cabinet, providing basic information for subsequent environmental compensation.

[0034] To ensure that data from different sources are consistent in time, the rules for forming the monitoring dataset clearly stipulate that all types of data must be sampled synchronously according to a unified time benchmark. For example, the same clock signal can be used to align different acquisition channels, so that the gas concentration, temperature, and voltage values ​​acquired within the same sampling period are strictly correlated in time.

[0035] After data acquisition, the raw data needs to be denoised and filtered to eliminate abnormal fluctuations caused by electromagnetic interference, sensor jitter, and other factors. Common practices include using moving average filtering, median filtering, or low-pass filtering to suppress high-frequency noise while preserving the main trends in parameter changes. Even after denoising, the resulting data may still have inconsistencies in units or ranges. Therefore, it is necessary to standardize the units and calibrate the ranges of all monitoring data. For example, gas concentrations should be converted to ppm, temperatures to degrees Celsius, and voltage and current should be normalized according to set reference values.

[0036] After completing the above calibration, to ensure data traceability and the orderly processing of subsequent data, it is necessary to add timestamp information and data source identifiers to each monitoring data point. The timestamp ensures that each data point corresponds to a specific acquisition time, while the source identifier indicates the data acquisition channel or sensor location, thereby facilitating the tracing of data sources and the identification of anomalies during subsequent analysis.

[0037] Finally, the monitoring data, with timestamps and source identifiers appended, is stored in a pre-defined data cache, forming a complete monitoring dataset. This cache can be managed using a circular queue or a sliding window to ensure effective storage and real-time retrieval of recent data even under memory constraints.

[0038] Step S20: Extract feature parameters based on the monitoring dataset, and generate a fused feature vector based on the extracted feature parameters.

[0039] Specifically, feature parameters are extracted based on the monitoring dataset, including: performing sliding window calculation on the gas monitoring data to obtain the concentration surge rate; performing differential processing on the temperature monitoring data to obtain the temperature rise rate; performing a fluctuation detection algorithm on the electrical monitoring data to obtain the voltage fluctuation amplitude; performing operating condition correction calculation based on the humidity detection data and the pressure monitoring data, substituting the deviations corresponding to the humidity detection data and the pressure monitoring data into a preset environmental correction model to obtain an environmental compensation factor for correcting the gas monitoring data and temperature monitoring data; and using the concentration surge rate, the temperature rise rate, the voltage fluctuation amplitude, and the environmental compensation factor as extracted feature parameters.

[0040] Furthermore, generating a fused feature vector based on the extracted feature parameters includes: performing normalization processing on each feature parameter except for the environmental compensation factor, and aligning the feature parameters at different time scales based on a dynamic time warping algorithm after the normalization processing is completed; wherein, the dynamic time warping algorithm is a constraint band-based dynamic time warping algorithm, used to limit the offset range of the time matching path by setting a preset constraint band width when aligning feature parameters at different time scales; adjusting the weights of the other aligned feature parameters in the fusion process based on the environmental compensation factor to obtain the fused feature vector.

[0041] In this embodiment of the invention, after constructing the monitoring dataset, it is necessary to extract key feature parameters that reflect the operating status of the battery compartment, and generate a fusion feature vector for subsequent analysis. The process of extracting feature parameters is not simply reading numerical values, but rather combining the physical meaning and variation patterns of various monitoring data, selecting appropriate algorithms for processing, and obtaining feature quantities with practical reference value. Specifically, it can be implemented according to the following rules: First, for gas monitoring data, in order to reflect the sudden change trend of gas concentration in a short period of time, a sliding time window of length N can be set, and the concentration sequence within the window is denoted as... The rate of change of concentration over time is obtained through linear fitting or difference calculation, thereby extracting the rate of concentration surge. This rate can directly characterize whether there is a rapid gas release phenomenon inside the energy storage tank, and is an important indicator in the early stages of thermal runaway.

[0042] For temperature monitoring data, considering that the temperature rise of battery modules often has a continuous and cumulative effect, differential processing can be used to obtain the temperature rise rate. That is, the temperature values ​​of adjacent sampling points are calculated to obtain... The temperature rise rate is obtained by dividing by the sampling interval. If further stabilization is needed, a moving average of the temperature rise rate over a period of time can be taken to suppress the influence of single-point fluctuations. The resulting temperature rise rate not only reflects the absolute change in temperature but also reveals the rate of temperature increase, which is of great significance in distinguishing between slow rises caused by normal charging and discharging and rapid rises before thermal runaway.

[0043] For electrical monitoring data, the key is to identify fluctuation amplitude from the voltage curve. In practice, fluctuation detection algorithms can be used, such as those based on the difference between the maximum and minimum values ​​within a sliding window, or wavelet transform to extract high-frequency components, thus obtaining the characteristic parameter of voltage fluctuation amplitude. Voltage fluctuation amplitude can reveal whether there are abnormal oscillations during battery charging and discharging. Especially before thermal runaway occurs, the voltage often drops sharply or fluctuates abnormally; therefore, this parameter has high early warning value.

[0044] Furthermore, to ensure that changes in gas and temperature accurately reflect actual operating conditions, the influence of environmental factors must be considered. Specifically, this involves performing operating condition correction calculations using humidity and pressure monitoring data. First, the deviation between the current humidity and a reference humidity, and the deviation between the current pressure and a reference pressure, are calculated. These are then substituted into a pre-defined environmental correction model to obtain an environmental compensation factor. This factor can be understood as a weighted correction parameter used to compensate for sensor response deviations caused by excessively high humidity or large pressure fluctuations. For example, when humidity is high, the concentration value output by the gas sensor may be too high; after correction using the environmental compensation factor, it can more closely approximate the true value. Similarly, when pressure fluctuations are large, the temperature sensor may experience delays or anomalies; the environmental compensation factor can also reduce this impact. Finally, the concentration surge rate, temperature rise rate, voltage fluctuation amplitude, and environmental compensation factor are used as extracted feature parameters.

[0045] After extracting the feature parameters, it is necessary to fuse these feature parameters with different dimensions and time scales to generate a fused feature vector. To achieve this, the first issue to address is the inconsistency in dimensions. Specifically, except for the environmental compensation factor, all other feature parameters need to be normalized. This can be done using z-score normalization (subtracting the mean from the parameter and then dividing by the standard deviation) or min-max normalization (mapping the parameter values ​​to the [0,1] interval). The benefit of normalization is that it allows parameters with different dimensions to be compared and fused within the same numerical range, preventing one type of parameter from dominating the fusion process due to excessively large values.

[0046] A further issue to address is the inconsistency in time scales. Changes in gas concentration can occur dramatically within seconds, while temperature increases often take minutes or even longer, and voltage fluctuations may fall somewhere in between. To align these characteristic parameters across different time scales, a Dynamic Time Warping (DTW) algorithm is required. Here, a DTW algorithm with constraint bands is used. By setting a preset constraint band width, the offset range of the time matching path is limited, thus avoiding over-stretching or compression of the match. For example, when aligning the temperature rise rate and the gas surge rate, without constraints, DTW might match points that are far apart, leading to distortion; however, by setting constraint bands, the rationality of the matching path can be ensured, improving alignment accuracy.

[0047] After normalization and alignment, the weighting effect of environmental factors needs to be considered. Specifically, an environmental compensation factor is used to weight the aligned feature parameters. This can be designed so that when the environmental compensation factor is close to 1, the original weights are maintained; when the environmental compensation factor deviates from 1, the fusion weights of each parameter are dynamically adjusted according to the degree of deviation. For example, in high humidity, the weights of gas sensor data are reduced, while the weights of temperature and electrical parameters are increased, thereby reducing environmental interference with the results. Finally, the feature parameters are linearly combined or vector-concatenated after weight adjustment to obtain the fused feature vector.

[0048] The fused feature vector generated through the above process contains dynamic information from multiple dimensions such as gas, temperature, and electrical properties, and ensures comparability and consistency between data through normalization, time alignment, and weight adjustment.

[0049] Step S30: Perform dynamic threshold determination based on the fused feature vector to obtain the determined dynamic threshold.

[0050] Specifically, dynamic threshold determination is performed based on the fused feature vector to obtain the determined dynamic threshold, including: extracting the corresponding gas concentration surge rate parameter and temperature rise duration parameter from the fused feature vector; establishing a two-factor criterion based on the concentration surge rate parameter and the temperature rise duration parameter; comparing the two-factor criterion with a preset reference threshold, and correcting the reference threshold when the comparison result meets the trigger condition to obtain the determined dynamic threshold; otherwise, when the comparison result does not meet the trigger condition, the preset reference threshold is used as the determined dynamic threshold.

[0051] Furthermore, extracting the corresponding gas concentration surge rate parameter and temperature rise duration parameter from the fused feature vector includes: performing sliding window filtering on the gas monitoring component in the fused feature vector to eliminate instantaneous disturbances, and then calculating the slope of the filtered gas concentration curve to obtain the gas concentration surge rate parameter; setting a predefined time window for the temperature monitoring component in the fused feature vector and detecting the length of the continuous exceeding interval to determine the temperature rise duration parameter; the triggering condition is: within a consecutive preset number of sampling periods, if the gas concentration surge rate parameter exceeds a first preset rate threshold and the temperature rise duration parameter is greater than a second preset duration threshold, then the triggering condition is determined to be met.

[0052] Specifically, the rule for correcting the reference threshold is as follows: the environmental compensation factor in the extracted feature parameters is used as the gate parameter of a preset gated linear interaction model, and the gas concentration surge rate parameter and the temperature rise duration parameter are substituted into the gated linear interaction model to obtain the threshold correction amount; wherein, the preset gated linear interaction model is expressed as:

[0053]

[0054] in, This is the threshold correction amount obtained in the k-th sampling period; For environmental compensation gating factors; These are the coefficients obtained from the offline calibration of the exponential decay function model; The rate of sudden increase in gas concentration in the fused feature vector Rate threshold relative to the preset reference threshold Dimensionless excess; The duration of temperature rise in the fused feature vector The duration threshold relative to the preset reference threshold Dimensionless excess; The normalized value of the alignment residual for the kth sampling period is used; the preset reference threshold is adjusted based on the threshold correction amount to obtain the determined dynamic threshold.

[0055] In this embodiment of the invention, after obtaining the fused feature vector, it is necessary to further determine the dynamic threshold based on this vector in order to obtain a determined dynamic threshold that can be used for subsequent safety status assessment. The emphasis on "dynamic" is because the operating state within the energy storage cabinet is not static; the gas release rate and temperature rise rate vary significantly under different operating conditions. If a fixed threshold is still used for judgment, it often leads to either overly conservative approaches resulting in a high false alarm rate, or overly lenient approaches resulting in a high risk of missed alarms. Therefore, it is necessary to dynamically adjust the reference threshold based on the real-time performance of different feature parameters, ensuring that the threshold maintains adaptability and sensitivity throughout the entire operation.

[0056] Furthermore, key discrimination parameters are extracted from the fused feature vector, focusing on the gas concentration surge rate parameter and the temperature rise duration parameter. These two parameters reflect the rate of gas release and the duration of temperature rise inside the energy storage tank, respectively, and are the most sensitive and direct signals in the early stages of thermal runaway. The extraction method is as follows: For the gas monitoring component in the fused feature vector, a sliding window filter is first performed to eliminate high-frequency fluctuations caused by airflow disturbances or instantaneous sensor jitter. The filtered gas concentration curve is smoother and more stable. Then, the slope is calculated, for example, by dividing the difference between adjacent sampling points by the time interval to obtain the rate of concentration change per unit time, i.e., the gas concentration surge rate parameter.

[0057] On the other hand, for the temperature monitoring component in the fused feature vector, a predefined time window is set, and the length of the continuous over-limit interval is detected within this window, that is, the duration for which the temperature continuously exceeds a certain benchmark threshold, thereby obtaining the temperature rise duration parameter. The two parameters extracted in this way, one is rate-type and the other is duration-type, complement each other and can more comprehensively characterize the signs of thermal runaway.

[0058] After the parameters are extracted, a two-factor criterion needs to be established based on these two parameters. The so-called two-factor criterion refers to considering the combined conditions of the gas concentration surge rate and the duration of the temperature rise, rather than relying on a single parameter for judgment.

[0059] Specifically, the gas concentration surge rate is compared with a first preset rate threshold, and the temperature rise duration is compared with a second preset duration threshold. The two-factor criterion is deemed valid only if both exceed their respective thresholds simultaneously. This design avoids false triggers caused by fluctuations in a single factor. For example, a high gas concentration surge rate at a certain moment may be merely an occasional disturbance if its duration is short, and does not represent a true thermal runaway trend. Conversely, a prolonged temperature rise without a significant gas concentration surge may be a normal load heating process. The two-factor criterion, by combining both rate and duration conditions, improves the reliability of the determination.

[0060] Furthermore, the two-factor criterion needs to be compared with a preset reference threshold to determine whether the reference threshold needs to be corrected. In implementation, a trigger condition can be set: if, within a continuously preset sampling period, the gas concentration surge rate parameter consistently exceeds a first preset rate threshold, and the temperature rise duration parameter is greater than a second preset duration threshold, then the trigger condition is considered met. The emphasis on the "continuous sampling period" requirement is to further improve the robustness of the judgment and avoid erroneous triggering due to a single anomaly. When the trigger condition is met, it indicates that the operating state inside the energy storage cabinet has shown an abnormal trend. At this point, the reference threshold needs to be corrected to obtain a determined dynamic threshold.

[0061] As for the correction method, various forms are employed. In a simple implementation, the reference threshold can be adjusted downwards by a preset margin to improve warning sensitivity. In a more complex implementation, the correction amount can be calculated based on the degree to which the concentration surge rate and temperature rise duration exceed limits. For example, the correction amount can be set as a weighted sum of the difference between the rate parameter and the threshold and the difference between the duration parameter and the threshold, with the weights obtained through offline calibration. This correction amount is applied to the reference threshold to obtain a new dynamic threshold. If the triggering condition is not met, the reference threshold remains unchanged and is used as the determined dynamic threshold. This achieves the process of dynamically adjusting the threshold according to the operating state.

[0062] Specifically, when correcting the reference threshold, a gated linear interaction model is introduced to combine multiple key feature quantities according to a set interaction relationship to obtain a dynamic correction amount. The basic idea of ​​this model is to use an environmental compensation factor as a gating parameter, adjusting its value range to control the influence of the gas concentration surge rate and temperature rise duration on the correction amount. In this way, under favorable environmental conditions, the gating factor allows gas and temperature features to play a major role; while when there are environmental deviations, the gating factor reduces the weight of these features to prevent false triggering caused by environmental disturbances.

[0063] In the gated linear interaction model, the environmental compensation factor, as a gating parameter, only applies to excessive gas concentration surge rate, excessive temperature rise duration, and their interaction terms. This is because these parameters directly correspond to the main physical criteria for thermal runaway risk and need adaptive adjustment based on environmental conditions to avoid false amplification or false triggering when humidity or pressure deviations are significant. The normalized alignment residual value only reflects the alignment error and fluctuation deviation of monitoring data on a short timescale; it is essentially a statistical correction, not the primary risk driver. Placing it within the gating system would easily lead to loss of correction due to environmental interference. Therefore, this residual term is introduced into the model in an independent weighted form, fine-tuning the correction without gating constraints. This achieves functional separation between the primary risk criterion and error correction, ensuring that threshold adjustment is both environmentally adaptive and maintains independent correction capabilities for short-term fluctuations, thereby improving the stability and accuracy of dynamic threshold determination.

[0064] Specifically, the model first calculates the relative excess between the gas concentration surge rate parameter and its reference threshold, obtaining a dimensionless rate excess; simultaneously, it calculates the relative excess between the temperature rise duration parameter and its reference duration, obtaining a dimensionless duration excess. These two excesses quantify the degree to which the current state deviates from the normal reference level. Subsequently, these two excesses are linearly combined with the residual normalization parameter. The residual normalization parameter reflects the difference between short-term fluctuations in the monitoring data and the expected trend, and is used to add a correction factor in the overall correction process, thereby avoiding error amplification caused by single-point noise. Finally, the above linear combination, under the adjustment of the gating factor, forms the threshold correction amount.

[0065] The significance of gated linear interaction lies in the fact that, through a combination of linear combinations and gating adjustments, it maintains the transparency and interpretability of the computational structure while enabling the correction process to adaptively adjust based on the stability of the external environment. Unlike traditional fixed weighting, the introduction of the gating mechanism allows the model to automatically reduce the sensitivity of threshold correction when environmental factors such as humidity and pressure are abnormal, avoiding false alarms caused by environmental noise; while in stable environments, it allows excess gas and temperature to have a stronger impact on the threshold, thereby achieving a sensitive risk response.

[0066] Furthermore, it should be noted that the operation of this model must meet a series of constraints to ensure the stability of the calculation and the rationality of the results. These include: the dimensionless excess of the gas concentration surge rate only participates in the correction when it is greater than zero, avoiding erroneous adjustments caused by negative values; the dimensionless excess of the temperature rise duration must also be limited to non-negative values ​​to ensure its physical meaning; the normalized value of the residual must be controlled in the [0,1] interval to reflect the magnitude of the relative deviation without infinitely amplifying it; the environmental compensation factor, as a gating parameter, must be limited to the (0,1) interval so that it only serves as a weight adjustment and does not reverse the contribution direction of the characteristic parameters; and the coefficients in the linear combination must be calibrated to positive values ​​through experiments or historical data to ensure that the correction amount increases monotonically with the increase of the excess.

[0067] By satisfying the above constraints, the gated linear interaction model can stably output a reasonable threshold correction amount and achieve dynamic adjustment of the reference threshold in practical applications. Its technical advantage lies in ensuring that threshold changes consider both the degree of anomaly in key features and environmental conditions and data fluctuations, thereby significantly reducing the probability of false alarms and missed alarms in the early risk identification of energy storage cabinets and improving the reliability and practicality of the overall early warning results.

[0068] Step S40: Based on the dynamic threshold, perform a safety status assessment on the monitoring dataset and output the early warning result of the energy storage system.

[0069] Specifically, within the evaluation window, the monitoring data is compared point by point with the dynamic threshold to obtain the window average excess, and a stable indicator is generated based on the stability and hysteresis criteria. The historical evaluation results are recursively accumulated using time decay weighting to form a risk score, and the risk score and the maximum excess component are output as the evaluation result. The evaluation result is converted into an early warning result containing a level identifier and a trigger source identifier according to the early warning level mapping table and then output.

[0070] Furthermore, within the evaluation window, the monitoring data is compared point-by-point with the dynamic threshold to obtain the window average excess. A stable indicator is generated based on stability and hysteresis criteria. A risk integral is formed by recursively accumulating historical evaluation results using time decay weighting. The risk integral and the maximum excess component are output as the evaluation result. This includes: within the evaluation window, comparing the target component used for evaluation in the monitoring data set with the corresponding threshold in the dynamic threshold to obtain the corresponding excess sequence and calculating the window average excess; applying stability and hysteresis criteria to the window average excess, including counting the number of periods where excess is established within a continuous sampling period and setting a fallback threshold to generate a stable indicator indicating whether a stable excess state has been entered; recursively accumulating historical evaluation results using time decay weighting based on the stable indicator to obtain the risk integral at the current moment, and outputting the current risk integral and the target component that caused the maximum excess as the evaluation result.

[0071] In this embodiment of the invention, after determining the dynamic threshold, it is necessary to further assess the safety status of the energy storage system's operating data based on this threshold, so as to output clear early warning results at the early stage of abnormal signs. Unlike traditional methods that rely solely on single-point data for judgment, this method employs a sliding evaluation window mechanism, comparing monitoring data over a period of time with the corresponding dynamic threshold point by point, thereby forming a complete state criterion over time. This not only avoids misjudgments caused by single-point fluctuations but also better captures the trend of continuous parameter anomalies.

[0072] Specifically, within the evaluation window, the target component used for evaluation in the monitoring dataset is compared point-by-point with the corresponding threshold in the dynamic threshold. The range of the target component can include gas concentration monitoring components, temperature monitoring components, electrical monitoring components, or the aforementioned data after environmental compensation correction. The result of the point-by-point comparison is a binary excess sequence, that is, at each sampling time, it is determined whether the component exceeds the corresponding dynamic threshold. If it exceeds, it is recorded as 1; otherwise, it is recorded as 0. By statistically analyzing the values ​​of the excess sequence throughout the entire evaluation window and further calculating the average value within the window, the window-average excess is obtained. The window-average excess can intuitively reflect the degree of excess of the component within a certain time range, and is more robust than single-point values.

[0073] Subsequently, stability and hysteresis criteria need to be applied to the window average overshoot. The core idea of ​​the stability criterion is to determine whether the overshoot phenomenon is persistent, rather than an occasional instantaneous fluctuation. Specifically, the number of periods where overshoot occurs within a continuous sampling period is counted. If this number exceeds a preset threshold, the system is considered to have entered a stable overshoot state. Simultaneously, a hysteresis criterion is introduced, setting a fallback threshold. Only when the overshoot significantly falls below the threshold and persists for a certain period is the system considered to have exited the stable overshoot state. The advantage of this design is to avoid "jitter," meaning that when data fluctuates around the threshold, alarms are not frequently triggered or cleared; instead, a certain degree of stability is maintained through the hysteresis mechanism, thereby reducing false alarms. Finally, based on the stability and hysteresis criteria, a stability indicator is generated, which is either 0 or 1, to indicate whether the system is currently in a stable overshoot state.

[0074] After obtaining the stable indication quantity, it is necessary to recursively accumulate the historical evaluation results. Considering that the safety evaluation is not a one-time judgment but requires gradually accumulating evidence in the time dimension, a time decay weighting method is introduced to calculate the risk integral. The basic idea is as follows: for the stable indication quantity obtained in each sampling period, a weight that decays with time is assigned. The closer the sampling period is to the current moment, the higher the weight; the farther the period is from the current moment, the weight gradually decreases. Through this decay mechanism, it can be ensured that the risk integral is more sensitive to recent abnormal performances and gradually weakens the influence of earlier abnormalities. The process of recursive accumulation can be achieved through weighted summation, that is, at each moment, the risk integral is equal to the integral of the previous moment multiplied by a decay factor, plus the weight value of the stable indication quantity at the current moment. The finally obtained risk integral is an index that evolves with time and can dynamically reflect the current risk level of the energy storage system.

[0075] Furthermore, when outputting the evaluation result, not only the risk integral needs to be given, but it is also necessary to clearly indicate the target component that causes the maximum overage. The specific approach is as follows: among all target components, compare the magnitudes of their window average overages, select the component with the largest median value as the main source of risk, and output the identifier of this component together with the current risk integral. This can avoid only giving a vague "high risk" conclusion in the evaluation result, but can intuitively show whether the risk mainly comes from gas, temperature or electrical fluctuations, facilitating maintenance personnel to take targeted measures.

[0076] After obtaining the risk integral and the component with the maximum overage, it is also necessary to map this evaluation result to a specific warning level for visual display and linkage control at the operation and maintenance end. For this purpose, a warning level mapping table is preset in advance, dividing the numerical range of the risk integral into multiple levels, such as low risk, general risk, high risk, and severe risk, etc. Each level not only corresponds to a numerical range but also corresponds to a level identifier and a trigger source identifier. The level identifier is used to intuitively reflect the degree of risk, and the trigger source identifier is used to indicate which type of parameter causes this risk level. Through this mapping, the finally output warning result not only includes a score or a numerical value, but gives the level and source in a structured form, facilitating subsequent automated processing and human decision-making.

[0077] Embodiment:

[0078] For the thermal runaway safety risk during the operation of an energy storage power station, a fire warning method for an energy storage cabinet based on multi-source sensing data fusion is provided. In this embodiment, an immersion liquid-cooled energy storage system is used as the application scenario, and an integrated gas sensor module is configured inside the energy storage cabinet. The sensor module can simultaneously monitor parameters such as carbon monoxide (CO), carbon dioxide (CO2), and ambient temperature, so as to achieve multi-dimensional early identification when an abnormality occurs in the battery compartment.

[0079] Specifically, the integrated gas sensor includes a CO2 concentration detection unit with a monitoring range of 0 ppm–10000 ppm and a measurement accuracy of ±(50 ppm + 5% of the reading); a CO concentration detection unit with a monitoring range of 1–1000 ppm and a measurement accuracy of ±40 ppm; and a temperature detection unit with a monitoring range of -40℃–125℃ and an accuracy of ±1℃. All of these sensors are installed in key locations inside the energy storage cabinet to ensure real-time detection of abnormal phenomena such as gas leaks and heat accumulation within the battery compartment. The data collected by the sensors is synchronously sampled and uploaded to form an operational monitoring dataset inside the energy storage cabinet.

[0080] To ensure the validity of the monitoring data, this embodiment performs filtering and noise reduction on the collected raw data, and uses range calibration and unit unification to ensure that the data from different sensing units have a consistent time base and numerical scale. The processed monitoring data is stored in a buffer with a timestamp and source identifier for subsequent feature parameter extraction.

[0081] In the data analysis process, a sliding window calculation is first performed on the gas concentration sequence to obtain the gas concentration surge rate parameter; the temperature sequence is then differentially processed to obtain the temperature rise rate or temperature rise duration parameter; simultaneously, an environmental compensation factor is calculated based on humidity and pressure data to correct the reliability of the gas and temperature components. Subsequently, after normalization and dynamic time warping algorithms, the above feature parameters are aligned and fused to generate a fused feature vector.

[0082] like Figure 2 As can be seen, during the period from t=12 to t=15, the gas concentration surge rate parameter rises rapidly, reaching a peak at t=13, significantly exceeding the fixed reference threshold. At this point, due to the fulfillment of the conditions of exceeding the surge rate and sustained temperature rise, a dynamic threshold correction mechanism is triggered, causing the dynamic threshold to be lowered within the high-risk range, thereby amplifying the excess difference between the parameter and the threshold. As the gas concentration surge rate gradually decreases, the dynamic threshold slowly recovers to the reference threshold level according to a preset recovery strategy. Through this mechanism, not only can the occurrence of exceedance events be accurately reflected, but the early warning sensitivity in the risk integration process can also be enhanced, realizing the transformation from fixed threshold judgment to dynamic judgment.

[0083] Similarly, such as Figure 3As can be seen, during the period from t=10 to t=18, the temperature rise duration parameter gradually accumulates and reaches a peak around t=15, significantly exceeding the fixed reference threshold of 3 minutes. At this point, the dynamic threshold adjustment mechanism is triggered, and the threshold is gradually lowered from the reference value to around 2 minutes, further amplifying the excess difference and thus increasing the sensitivity to the risk of persistent temperature anomalies. As the temperature rise duration parameter gradually decreases, the dynamic threshold gradually recovers to the reference value level through a recovery strategy. This process avoids the early warning lag caused by an overly rigid fixed threshold, enabling timely detection of temperature rise anomalies in the early stages of thermal runaway in the energy storage system, improving the real-time performance and accuracy of safety status assessment.

[0084] In the early warning and judgment process of actual energy storage systems, a single parameter is often insufficient to fully reflect the development trend of thermal runaway. Therefore, this embodiment introduces a dual-factor joint mechanism of the gas concentration surge rate parameter and the temperature rise duration parameter. Figure 2 and Figure 3 As shown, the gas concentration surge rate parameter can quickly capture the gas release phenomenon caused by decomposition reactions inside the battery. Its value changes steeply and reaches its peak in a short time, making it suitable as a sensitive indicator of sudden anomalies. The temperature rise duration parameter reflects the duration of the temperature anomaly; its increase often accompanies the inability of heat to dissipate effectively inside the battery, making it a suitable indicator for judging whether thermal runaway is continuously escalating. By combining the two, a dynamic threshold correction mechanism is triggered simultaneously when the concentration surge rate reaches the reference threshold and the duration parameter exceeds the set value. This ensures that the warning result considers both rapid gas release and the persistence of the temperature anomaly. This dual-factor combined effect avoids false alarms or missed alarms that may be caused by a single indicator, improves the sensitivity and reliability of the risk integral, and thus achieves a more accurate early warning of thermal runaway risk in energy storage systems.

[0085] Furthermore, during the safety assessment process, dynamic thresholds are determined based on fused feature vectors. Specifically, when the rate of sudden increase in gas concentration exceeds a reference threshold and the duration of temperature rise is greater than a preset duration, the system triggers threshold correction logic. This logic dynamically adjusts the reference threshold through a gated linear interaction model to obtain a dynamic threshold that better suits the operating environment.

[0086] Finally, within the evaluation window, the monitoring data is compared point-by-point with the dynamic threshold to form an excess sequence, and the average excess within the window is calculated. A stable indicator is generated using stability and hysteresis criteria, and then the historical evaluation results are recursively accumulated using a time decay weighting method to obtain a risk score. The risk score and the maximum excess component are output as the evaluation result, and converted into specific warning levels and trigger sources according to a preset warning level mapping table, ultimately providing a warning result for the energy storage cabinet that includes the risk level and source identifier.

[0087] Through the above embodiments, it can be seen that this solution not only relies on the fusion of multiple sources such as CO, CO2 and temperature, but also introduces environmental compensation, stability criteria and time weighting in data processing and threshold determination. This effectively improves the reliability of early warning results in the immersion liquid-cooled energy storage cabinet scenario, reduces the probability of false alarms and missed alarms, and achieves high-precision safety control during the operation of the energy storage system.

[0088] Figure 4 This is a system structure diagram of an energy storage system early warning system based on dynamic thresholds provided in one embodiment of the present invention. Figure 4 As shown, this invention provides an early warning system for an energy storage system based on a dynamic threshold. The system includes: a data acquisition unit for acquiring multi-source operational data of the energy storage system to form a monitoring dataset; a processing unit for extracting feature parameters based on the monitoring dataset and generating a fused feature vector based on the extracted feature parameters; a threshold determination unit for performing dynamic threshold determination based on the fused feature vector to obtain a determined dynamic threshold; and an evaluation unit for performing a safety status evaluation on the monitoring dataset based on the dynamic threshold and outputting an early warning result for the energy storage system.

[0089] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0090] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0091] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for early warning of energy storage systems based on dynamic thresholds, characterized in that, The method includes: Collect multi-source operational data from energy storage systems to form a monitoring dataset; Feature parameters are extracted from the monitoring dataset, and a fused feature vector is generated based on the extracted feature parameters; Dynamic threshold determination is performed based on the fused feature vector to obtain the determined dynamic threshold, including: Extract the corresponding gas concentration surge rate parameter and temperature rise duration parameter from the fused feature vector; establish a two-factor criterion based on the concentration surge rate parameter and the temperature rise duration parameter; compare the two-factor criterion with a preset reference threshold, and correct the reference threshold when the comparison result meets the trigger condition to obtain the determined dynamic threshold; otherwise, if the comparison result does not meet the trigger condition, use the preset reference threshold as the determined dynamic threshold. The rule for correcting the reference threshold is as follows: The environmental compensation factor from the extracted feature parameters is used as the gating parameter of a pre-defined gated linear interaction model. The gas concentration surge rate parameter and the temperature rise duration parameter are then substituted into the gated linear interaction model to obtain the threshold correction amount. The preset gated linear interaction model is represented as follows: ； in, This is the threshold correction amount obtained in the k-th sampling period; For environmental compensation gating factors; These are the coefficients obtained from the offline calibration of the exponential decay function model; The rate of sudden increase in gas concentration in the fused feature vector Rate threshold relative to the preset reference threshold Dimensionless excess; The duration of temperature rise in the fused feature vector The duration threshold relative to the preset reference threshold Dimensionless excess; The normalized value of the alignment residual in the kth sampling period is used; the preset reference threshold is adjusted based on the threshold correction amount to obtain the determined dynamic threshold; The monitoring dataset is assessed for safety status based on the dynamic threshold, and the early warning results of the energy storage system are output.

2. The early warning method for energy storage systems based on dynamic thresholds according to claim 1, characterized in that, The multi-source operational data includes: Any one or more of the following: gas monitoring data, temperature monitoring data, electrical monitoring data, humidity detection data, and pressure monitoring data inside the energy storage cabinet; The rules for forming the monitoring dataset are as follows: All monitoring data are sampled synchronously according to a unified time benchmark. The collected data are denoised and filtered. After denoising and filtering, the units of each monitoring data are unified and the range is calibrated. Add timestamp information and data source identifiers to each calibrated monitoring data and store them in a preset data cache to form a monitoring dataset.

3. The early warning method for energy storage systems based on dynamic thresholds according to claim 2, characterized in that, Feature parameters are extracted based on the monitoring dataset, including: Perform a sliding window calculation on the gas monitoring data to obtain the concentration surge rate; The temperature monitoring data is differentially processed to obtain the temperature rise rate; A fluctuation detection algorithm is executed on the electrical monitoring data to obtain the voltage fluctuation amplitude; Based on the humidity detection data and the pressure monitoring data, a working condition correction calculation is performed. The deviation between the humidity detection data and the pressure monitoring data is substituted into a preset environmental correction model to obtain an environmental compensation factor for correcting the gas monitoring data and temperature monitoring data. The concentration surge rate, the temperature rise rate, the voltage fluctuation amplitude, and the environmental compensation factor were used as extracted feature parameters.

4. The early warning method for energy storage systems based on dynamic thresholds according to claim 3, characterized in that, A fused feature vector is generated based on the extracted feature parameters, including: After normalizing all feature parameters except for the environmental compensation factor, the feature parameters at different time scales are aligned based on the dynamic time warping algorithm after normalization. The dynamic time warping algorithm is a constraint band-based dynamic time warping algorithm, which is used to limit the offset range of the time matching path by setting a preset constraint band width when aligning feature parameters at different time scales. The weights of the other feature parameters after alignment are adjusted based on the environmental compensation factor during the fusion process to obtain the fused feature vector.

5. The early warning method for energy storage systems based on dynamic thresholds according to claim 1, characterized in that, Extract the corresponding gas concentration surge rate parameter and temperature rise duration parameter from the fused feature vector, including: A sliding window filter is applied to the gas monitoring component in the fused feature vector to eliminate instantaneous disturbances, and then the slope of the filtered gas concentration curve is calculated to obtain the gas concentration surge rate parameter. A predefined time window is set for the temperature monitoring component in the fused feature vector, and the length of the continuous over-limit interval is detected to determine the temperature rise duration parameter; The triggering condition is: If, within a consecutive preset number of sampling periods, the gas concentration surge rate parameter exceeds a first preset rate threshold and the temperature rise duration parameter is greater than a second preset duration threshold, then the triggering condition is determined to be met.

6. The early warning method for energy storage systems based on dynamic thresholds according to claim 1, characterized in that, Based on the dynamic threshold, a safety status assessment is performed on the monitoring dataset, and an early warning result for the energy storage system is output, including: Within the evaluation window, the monitoring data is compared with the dynamic threshold point by point to obtain the average excess of the window. Based on the stability and hysteresis criteria, a stable indicator is generated. The historical evaluation results are recursively accumulated using time decay weighting to form a risk score. The risk score and the maximum excess component are output as the evaluation result. The assessment results are converted into warning results containing level identifiers and trigger source identifiers according to the warning level mapping table and then output.

7. The early warning method for energy storage systems based on dynamic thresholds according to claim 6, characterized in that, Within the evaluation window, monitoring data is compared point-by-point with dynamic thresholds to obtain the window average excess. A stable indicator is generated based on stability and hysteresis criteria. A risk score is formed by recursively accumulating historical evaluation results using time-decay weighting. The risk score and the maximum excess component are then output as the evaluation result, including: Within the evaluation window, the target component used for evaluation in the monitoring dataset is compared point by point with the corresponding threshold in the dynamic threshold to obtain the corresponding excess sequence and calculate the window average excess. The stability and hysteresis criteria for the average over-quantity of the window are applied, including counting the number of periods in which over-quantity is established within a continuous sampling period and setting a fallback threshold, and generating a stable indicator quantity to indicate whether a stable over-quantity state has been entered. Based on the stable indicator, the historical evaluation results are recursively accumulated using time decay weighting to obtain the risk score at the current moment. The current risk score and the target component that causes the maximum excess are then output as the evaluation result.

8. An early warning system for energy storage systems based on dynamic thresholds, characterized in that, The system is applied to the early warning method for energy storage systems based on dynamic thresholds as described in any one of claims 1-7, and the system comprises: The acquisition unit is used to collect multi-source operational data from the energy storage system to form a monitoring dataset. The processing unit is used to extract feature parameters based on the monitoring dataset and generate a fused feature vector based on the extracted feature parameters. A threshold determination unit is used to perform dynamic threshold determination based on the fused feature vector to obtain the determined dynamic threshold. The evaluation unit is used to evaluate the safety status of the monitoring dataset based on the dynamic threshold and output the early warning results of the energy storage system.

Images