Energy storage equipment operation state real-time early warning supervision method based on LSTM hybrid model

Abstract

The invention relates to the technical field of energy storage equipment monitoring, and discloses an energy storage equipment operation state real-time early warning and supervision method based on an LSTM hybrid model. The method comprises the following steps: firstly, collecting current fluctuation time sequence data, a voltage deviation distribution curve and a temperature gradient evolution graph of the energy storage equipment, and generating a multi-source operation parameter set; constructing a long-short-term memory network dominant capacity attenuation prediction model based on the current fluctuation time sequence data, and constructing a time sequence convolution assisted voltage anomaly identification model based on the voltage deviation distribution curve; then monitoring the capacity attenuation trajectory and internal impedance fluctuation data of the equipment in real time, and dynamically activating the target model to generate a risk early warning instruction; and finally, inputting an early warning instruction into the regulation and control unit, and adjusting a charging and discharging strategy or starting a protection mechanism in real time. According to the method, through multi-source parameter fusion and dynamic model activation, comprehensive and accurate monitoring and quick response of the operation state of the energy storage equipment are realized, and the safety and stability of equipment operation are improved.

Description

Technical Field

[0001] This invention relates to the field of energy storage equipment monitoring technology, specifically to a real-time early warning and monitoring method for the operating status of energy storage equipment based on an LSTM hybrid model. Background Technology

[0002] With the rapid development of the new energy industry, energy storage equipment, as the core carrier of energy storage and dispatch, is directly related to the safe and efficient operation of the power system. Currently, the operating environment of energy storage equipment is complex and variable. Affected by multiple factors such as charging and discharging frequency, ambient temperature, and load fluctuations, it is prone to problems such as capacity decay and abnormal internal impedance. If these problems are not detected and addressed in time, they may lead to equipment failure or even safety accidents. In existing technologies, monitoring the operating status of energy storage devices often employs single-parameter analysis methods, such as judging the device status solely based on current or voltage data. This method has significant limitations because energy storage device failures are often the result of multiple parameters acting together, and a single parameter cannot comprehensively reflect the device's true operating status. Furthermore, traditional monitoring models mostly use static analysis methods, failing to effectively capture the dynamic characteristics of parameters changing over time, leading to delayed warnings or a high false alarm rate. Existing early warning mechanisms suffer from poor linkage with equipment control units. Even when an early warning is issued, it is difficult to quickly adjust charging and discharging strategies or activate protection mechanisms, thus failing to promptly curb the further development of equipment failures. In large-scale energy storage systems, this lag can trigger a chain reaction, causing severe economic losses and safety hazards. Therefore, developing a real-time early warning and monitoring method that can integrate multi-source parameters, dynamically capture changes in equipment status, and efficiently link with control units has become a problem that needs to be solved in the current energy storage field. Summary of the Invention

[0003] The purpose of this invention is to provide a real-time early warning and monitoring method for the operating status of energy storage devices based on an LSTM hybrid model, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, this invention provides a real-time early warning and monitoring method for the operating status of energy storage devices based on an LSTM hybrid model, the method comprising: Collect multi-dimensional operating parameter sequences of energy storage devices to generate a multi-source operating parameter set, which includes current fluctuation time series data, voltage offset distribution curves, and temperature gradient evolution maps; Based on the current fluctuation time series data in the multi-source operating parameter set, a capacity decay prediction model dominated by a long short-term memory network is constructed. The capacity decay prediction model is used to capture the dynamic mapping relationship between current fluctuation and capacity loss. Based on the voltage offset distribution curve in the multi-source operating parameter set, a time-series convolution-assisted voltage anomaly identification model is constructed. The voltage anomaly identification model is used to extract the nonlinear correlation rules between voltage offset and internal impedance change. Real-time monitoring of the capacity decay trajectory and internal impedance fluctuation data of energy storage devices; dynamic activation of the target model in the capacity decay prediction model or voltage anomaly identification model; and generation of risk warning instructions through the target model. The risk warning command is input to the control unit of the energy storage device to adjust the charging and discharging strategy or activate the protection mechanism in real time.

[0005] Preferably, the multi-dimensional operating parameter sequence of the energy storage device is collected to generate a multi-source operating parameter set, including: Electrochemical characteristic data of the energy storage unit during charge-discharge cycles are collected synchronously by distributed sensors. The electrochemical characteristic data includes polarization impedance spectrum and ion diffusion rate parameters. Based on the peak distribution range in the polarization impedance spectrum, the energy storage unit is divided into at least two operating state clusters, and an initial charge and discharge threshold range and a temperature control reference value are assigned to each operating state cluster. Extract the capacity attenuation compensation coefficient and voltage offset correction parameter that match each operating state cluster from the historical parameter database; The capacity attenuation compensation coefficient is normalized in the time domain to generate an optimized capacity attenuation prediction sequence for each operating state cluster. The voltage offset correction parameters are subjected to frequency domain feature enhancement processing to generate optimized voltage anomaly identification curves corresponding to each operating state cluster; By integrating the optimized capacity decay prediction sequence and optimized voltage anomaly identification curve of each operating state cluster, the multi-source operating parameter set is generated, and the historical operating mode migration trajectory is associated with it.

[0006] Preferably, the construction of the capacity decay prediction model and voltage anomaly identification model includes: The optimized capacity decay prediction sequence is input into a bidirectional long short-term memory network, and the hidden state weights are updated through a gated recurrent unit to generate the capacity loss compensation function in the capacity decay prediction model. The optimized voltage anomaly identification curve is input into a temporal convolutional network, and the feature extraction kernel is adjusted through a dilated convolutional layer to generate an impedance fluctuation mapping table in the voltage anomaly identification model. After the bidirectional long short-term memory network and the temporal convolutional network converge, the core feature tensors of the capacity loss compensation function and the impedance fluctuation mapping table are extracted respectively. The similarity between the core feature tensor and the real-time acquired electrochemical feature data is compared to verify the adaptability of the capacity decay prediction model and the voltage anomaly identification model. If the similarity comparison result is lower than the preset tolerance threshold, the hidden state vector of the bidirectional long short-term memory network is reinitialized until the core feature tensor satisfies the adaptability constraint.

[0007] Preferably, the real-time monitoring of the capacity decay trajectory and internal impedance fluctuation data of the energy storage device, dynamically activating the target model and generating risk warning instructions, includes: Continuously track the real-time capacity decay rate curve and internal impedance abrupt change amplitude of energy storage devices; When the real-time capacity decay rate curve exceeds the preset decay threshold and the internal impedance change amplitude is in a stable fluctuation range, the capacity loss compensation function in the capacity decay prediction model is activated. Based on the attenuation compensation rule in the capacity loss compensation function, a dynamic adjustment command for the charging and discharging strategy is generated. When the amplitude of the internal impedance change exceeds the preset impedance tolerance and the real-time capacity decay rate curve is in the controllable fluctuation range, the impedance fluctuation mapping table in the voltage anomaly identification model is activated. Based on the impedance control rules in the impedance fluctuation mapping table, a protection mechanism trigger command is generated; If the real-time capacity decay rate curve and the internal impedance abrupt change amplitude both exceed the critical threshold, the dynamic adjustment instruction generated by the capacity loss compensation function will be executed first, and the trigger instruction of the impedance fluctuation mapping table will be delayed until the charging and discharging strategy completes the correction operation.

[0008] Preferably, the step of inputting the risk warning command to the control unit of the energy storage device to adjust the charging and discharging strategy or activate the protection mechanism in real time includes: Based on the attenuation compensation amount in the dynamic adjustment command, the current output amplitude in the charging and discharging strategy is adjusted in stages. After each current adjustment, the actual capacity decay data of the energy storage unit is collected and the offset is calculated with the predicted value of the capacity loss compensation function. If the offset continues to converge, maintain the current adjustment direction until the target capacity retention range is reached; If the offset shows a diverging trend, the current output amplitude is adjusted in the opposite direction and the weight iteration of the capacity loss compensation function is triggered. The action threshold of the protection mechanism is dynamically adjusted according to the impedance control parameter in the protection mechanism trigger command. During the threshold adjustment process, the internal impedance change is monitored in real time by electrochemical impedance spectroscopy, and the control coefficient in the impedance fluctuation mapping table is updated according to the monitoring results.

[0009] Preferably, the method further includes a feedback calibration phase after the early warning command is executed: Collect the final capacity retention rate data and internal impedance stability spectrum of the energy storage device after it has completed a charge-discharge cycle; The final capacity retention rate data is compared with the prediction range of the capacity decay prediction model to generate a capacity prediction error signal. The overlap between the internal impedance stability spectrum and the expected impedance template of the voltage anomaly identification model is calculated to generate an impedance identification error signal. Based on the system deviation component in the capacity prediction error signal, the attenuation compensation rule in the capacity attenuation prediction model is corrected; Based on the random fluctuation component in the impedance identification error signal, optimize the impedance control parameters in the voltage anomaly identification model; The revised attenuation compensation rules and impedance control parameters are synchronously updated to the historical parameter library of the multi-source operating parameter set.

[0010] Preferably, the process of modifying the attenuation compensation rule and optimizing the impedance control parameters includes: Identify the steady-state offset component in the capacity prediction error signal and calculate the steady-state compensation amount using a sliding time window algorithm; Adjust the reference current output value in the capacity loss compensation function according to the steady-state compensation amount; The high-frequency noise component in the impedance identification error signal is identified, and the effective correction amount is extracted using a wavelet denoising algorithm. Adjust the action threshold weights in the impedance fluctuation mapping table according to the effective correction amount; The updated capacity loss compensation function and impedance fluctuation mapping table are integrated into the early warning model parameter library.

[0011] Preferably, the method further includes performing preprocessing operations before the device is started: Analyze the cycle count encoding and health status decay segment in the charge and discharge history of energy storage devices to generate a health feature vector; The health feature vector is input into a pre-built operating mode database for similarity retrieval, and a set of candidate benchmark parameters with a matching degree exceeding a preset threshold is selected. Extract the historical early warning accuracy and false alarm suppression rate of each candidate benchmark parameter in the candidate benchmark parameter set, and calculate the comprehensive early warning effectiveness score; The candidate benchmark parameter set is sorted according to the comprehensive early warning effectiveness score, and the candidate benchmark parameter with the highest score is selected as the optimal charging and discharging benchmark template. Extract the set of impedance regulation reference curves associated with the optimal charge / discharge reference template from the operating mode database; The synergy between the charging / discharging strategy and impedance control is verified for each curve in the impedance control reference curve set. Abnormal curves with strategy conflicts or impedance abrupt changes are eliminated to generate an optimized impedance control benchmark set. Based on the historical stability index of each curve in the optimized impedance control reference set, the curve with the smallest variance is selected as the final impedance control reference curve. The optimal charge / discharge reference template and the final impedance control reference curve are time-aligned to generate an initial configuration for a multi-source operating parameter set.

[0012] Preferably, the verification of the synergy between the charge / discharge strategy and impedance regulation for each curve in the impedance regulation reference curve set includes: Extract the impedance control data of a single curve to be verified, and simultaneously obtain the time-aligned charge and discharge strategy sequence in the optimal charge and discharge reference template; Based on the phase nodes of the charge-discharge strategy sequence, a cooperative timestamp identifier is marked on the curve to be verified to generate an impedance control curve with timing markers. Traverse each cooperative timestamp identifier in the impedance control curve with time-series markers, detect whether the control change gradient between adjacent time intervals exceeds the mutation tolerance, and identify abnormal intervals with strategy conflicts. If an abnormal interval is detected, the charging and discharging strategy value of the corresponding node in the optimal charging and discharging reference template is traced back to determine whether the charging and discharging strategy direction and the impedance control direction produce reverse interference. When the reverse interference intensity exceeds the conflict threshold, the abnormal interval is marked as a strategy conflict domain, and the starting position and duration of the strategy conflict domain are calculated. Based on the spatiotemporal attributes of the strategy conflict domain, a control correction interval is defined on the curve to be verified, and an alternative control smoothing segment is generated based on historical conflict solutions. The alternative control smoothing segment is inserted into the control correction interval to generate an optimized impedance control curve, and abnormal data points that overlap with the strategy conflict domain in the original curve are deleted. The integrity of the optimized impedance control curve after the substitution insertion operation is verified, and curves with residual uncorrected conflict regions are removed. The curves that pass the merge verification are used to generate the optimized impedance control reference set.

[0013] Preferably, the acquisition process of the multidimensional operating parameter sequence includes: capturing the transient response waveform during the charging and discharging process of the energy storage device through a high-frequency sampling module, extracting the rising slope, oscillation attenuation coefficient and steady-state duration of the transient response waveform; and fusing the transient response waveform features with current fluctuation time series data, voltage offset distribution curve and temperature gradient evolution map to generate an enhanced operating feature tensor. The enhanced runtime feature tensor is input into the feature selection layer, and redundant dimensions are filtered out through an importance weight allocation algorithm to generate a dimensionality-reduced core parameter matrix, which is then updated to the multi-source runtime parameter set.

[0014] Compared with the prior art, the beneficial effects of the present invention are: By collecting multi-dimensional operating parameter sequences to generate a multi-source operating parameter set, encompassing current fluctuation time-series data, voltage offset distribution curves, and temperature gradient evolution maps, this approach overcomes the limitations of traditional single-parameter monitoring and achieves comprehensive perception of the operating status of energy storage equipment. The fusion analysis of multi-source parameters can reflect the operating characteristics of the equipment from different dimensions, reducing misjudgments caused by incomplete parameter information and making the assessment of equipment status more comprehensive and accurate. The capacity decay prediction model, dominated by Long Short-Term Memory (LSTM) networks, is specifically designed for current fluctuation time-series data and can accurately capture the dynamic mapping relationship between current fluctuations and capacity loss. Due to the advantages of LSM networks in processing time-series data, this model can effectively uncover patterns in current data changes over time, anticipate capacity decay trends, and avoid the lag of traditional static models in dynamically changing scenarios, making capacity decay predictions more forward-looking. A time-series convolution-assisted voltage anomaly identification model, built upon voltage offset distribution curves, can extract nonlinear correlation rules between voltage offset and internal impedance changes. The ability of time-series convolutional networks to handle local features and nonlinear relationships allows this model to deeply analyze the complex information contained in voltage data, promptly identifying internal impedance problems hidden behind voltage anomalies. Compared to traditional linear analysis methods, it is better suited to the complex nonlinear relationships between parameters in energy storage equipment operation, thus improving the accuracy of voltage anomaly identification. Real-time monitoring of capacity decay trajectory and internal impedance fluctuation data, along with dynamic activation of the target model to generate risk warning commands, achieves dynamic and targeted warnings. The system flexibly selects appropriate models for analysis based on the real-time operating status of the equipment, avoiding unnecessary waste of computational resources and ensuring a high degree of matching between warning commands and the current state of the equipment, making warnings more timely and effective. By inputting risk warning commands into the control unit to adjust charging and discharging strategies or activate protection mechanisms, a seamless connection between early warning and control is achieved. This linkage mechanism enables immediate action upon detection of risks, rapidly curbing the development of equipment failures, reducing the likelihood of equipment damage, ensuring the stable operation of the energy storage system, and mitigating economic losses and safety risks caused by failures. Attached Figure Description

[0015] Figure 1 This is a timing diagram of the real-time early warning and monitoring method for the operation status of energy storage devices based on the LSTM hybrid model described in this invention. Figure 2A flowchart generated for a multi-source set of operating parameters; Figure 3 Flowcharts for constructing capacity decay prediction models and voltage anomaly identification models; Figure 4 This is a flowchart of the feedback calibration phase after the early warning command is executed; Figure 5 The flowchart shows the optimization method for attenuation compensation rules and impedance control parameters. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 This invention provides a real-time early warning and monitoring method for the operating status of energy storage devices based on an LSTM hybrid model. This method collects multi-dimensional operating parameters and constructs a hybrid model to achieve accurate early warning and control of energy storage devices.

[0018] Multi-dimensional operating parameter sequences of energy storage devices are collected to generate a multi-source operating parameter set, including time-series data of current fluctuations, voltage offset distribution curves, and temperature gradient evolution maps. Based on the time-series data of current fluctuations, a capacity decay prediction model dominated by a long short-term memory network is constructed to capture the dynamic mapping relationship between current fluctuations and capacity loss. Simultaneously, based on the voltage offset distribution curves, a time-series convolution-assisted voltage anomaly identification model is constructed to extract the nonlinear correlation rules between voltage offset and internal impedance changes. The capacity decay trajectory and internal impedance fluctuation data of the energy storage devices are monitored in real time. Based on the monitoring results, the target model in either the capacity decay prediction model or the voltage anomaly identification model is dynamically activated, and a risk warning command is generated through the target model. The risk warning command is input to the control unit of the energy storage device to realize real-time adjustment of the charging and discharging strategy or activation of the protection mechanism.

[0019] Example 1: See Figure 2 and Figure 3When collecting multi-dimensional operating parameter sequences of energy storage devices to generate a multi-source operating parameter set, it is necessary to use distributed sensors to synchronously acquire electrochemical characteristic data of the energy storage unit during charge-discharge cycles. These data cover polarization impedance spectroscopy and ion diffusion rate parameters. The distributed sensor layout needs to cover different regions of the energy storage unit to ensure comprehensive capture of the operating status of each part. Polarization impedance spectroscopy reflects the change of resistance characteristics of the energy storage unit with frequency, while ion diffusion rate parameters are related to the migration ability of ions within the energy storage unit. Together, they constitute important basic data for analyzing the operating status of the energy storage unit.

[0020] Based on the peak distribution range in the polarization impedance spectrum, the energy storage unit can be divided into at least two operating state clusters. Different peak distribution ranges correspond to different internal states of the energy storage unit. For example, some peak ranges may represent that the energy storage unit is in a highly efficient operating state, while other ranges may indicate that its operating efficiency has decreased. After the division, an initial charge / discharge threshold range and a temperature control reference value are assigned to each operating state cluster. The initial charge / discharge threshold range should be set with reference to the design parameters of the energy storage unit to avoid damage caused by overcharging and discharging; the temperature control reference value is determined based on the suitable operating temperature range of the energy storage unit in this operating state, providing an initial basis for subsequent temperature regulation.

[0021] Capacity attenuation compensation coefficients and voltage offset correction parameters matching each operating state cluster are extracted from the historical parameter database. The historical parameter database stores past operating data for this energy storage device and similar devices. By searching historical records similar to the current operating state cluster, the corresponding compensation coefficients and correction parameters can be obtained. The capacity attenuation compensation coefficient is used to adjust the capacity attenuation prediction results to reduce prediction deviations caused by various factors; the voltage offset correction parameter corrects the voltage offset data, making subsequent voltage anomaly identification more accurate.

[0022] The capacity degradation compensation coefficients are time-domain normalized to generate optimized capacity degradation prediction sequences for each operating state cluster. Time-domain normalization adjusts the capacity degradation compensation coefficients at different time scales to a unified time base, eliminating the influence of time factors on the data and making the compensation coefficients from different time periods comparable. The optimized capacity degradation prediction sequences obtained after this processing more clearly reflect the changing trend of capacity degradation, providing more reliable input data for the capacity degradation prediction model.

[0023] Frequency domain feature enhancement processing is applied to the voltage offset correction parameters to generate optimized voltage anomaly identification curves for each operating state cluster. This frequency domain feature enhancement process analyzes the voltage offset correction parameters in the frequency domain, highlighting features relevant to voltage anomalies and suppressing irrelevant interference signals. The processed optimized voltage anomaly identification curves more accurately represent the characteristics of voltage offset, helping the voltage anomaly identification model to better extract the correlation rules between voltage offset and internal impedance changes.

[0024] By fusing optimized capacity decay prediction sequences and optimized voltage anomaly identification curves from various operating state clusters, a multi-source operating parameter set is generated, which is then correlated with historical operating mode migration trajectories. The fusion process requires integrating data from different operating state clusters to ensure data consistency and integrity. The historical operating mode migration trajectory records the changes in the operating status of energy storage devices over different time periods. Correlating this with the multi-source operating parameter set allows the set to not only include current operating parameter information but also incorporate historical changes, providing a more comprehensive reference for subsequent model building and early warning analysis.

[0025] When constructing the capacity decay prediction model and voltage anomaly identification model, the optimized capacity decay prediction sequence is input into a bidirectional long short-term memory (LSTM) network. The bidirectional LSM network has the advantage of processing time-series data, enabling it to analyze sequence data using both past and future information simultaneously. By updating the hidden state weights through gated recurrent units (ROUs), the RNUs can selectively retain or forget historical information, allowing the network to better capture long-term dependencies in the sequence and thus generate the capacity loss compensation function in the capacity decay prediction model. This function can calculate the relevant parameters used to compensate for capacity loss based on the input optimized capacity decay prediction sequence.

[0026] The optimized voltage anomaly identification curve is input into a temporal convolutional network (TCNN). This TCNN processes the sequence data through multiple convolutional layers, effectively extracting both local and global features. By adjusting the feature extraction kernel using dilated convolutional layers, the receptive field can be expanded without increasing the number of parameters, enabling the network to capture features from greater distances. This results in an impedance fluctuation mapping table in the voltage anomaly identification model. This mapping table records the correspondence between voltage shifts and internal impedance changes, providing a basis for identifying voltage anomalies.

[0027] After the bidirectional long short-term memory network and the temporal convolutional network converge, the core feature tensors of the capacity loss compensation function and the impedance fluctuation map are extracted, respectively. Network convergence means that the model has reached a relatively stable state through training, and the extracted core feature tensors at this time can represent the key information learned by the model. The core feature tensors contain the most representative features of the capacity loss compensation function and the impedance fluctuation map, and are important bases for evaluating model performance and conducting fitness verification.

[0028] The similarity of the core feature tensor with real-time acquired electrochemical feature data is compared to verify the adaptability of the capacity decay prediction model and the voltage anomaly identification model. The similarity comparison is achieved by calculating the similarity between the core feature tensor and the real-time electrochemical feature data. A higher similarity indicates a better match between the model and the current operating state of the energy storage device. If the similarity comparison result is lower than a preset tolerance threshold, it indicates insufficient adaptability of the model to the current operating state. In this case, the hidden state vector of the bidirectional long short-term memory network needs to be reinitialized. Reinitializing the hidden state vector can eliminate any adverse effects that may have existed during previous training, allowing the network to relearn new data features until the core feature tensor meets the adaptability constraints, ensuring that the model can be accurately applied to the early warning and monitoring of the current energy storage device's operating state.

[0029] Example 2: When monitoring the capacity decay trajectory and internal impedance fluctuation data of energy storage devices in real time, dynamically activating the target model, and generating risk warning commands, it is necessary to continuously track the real-time capacity decay rate curve and the amplitude of internal impedance changes. The real-time capacity decay rate curve is plotted by continuously recording the amount of capacity reduction of the energy storage device per unit time, which can intuitively reflect the rate of capacity decay. The amplitude of internal impedance changes is obtained by monitoring the difference in internal impedance at different times, and is used to capture sudden changes in impedance. Real-time tracking of these two indicators requires high-precision sensing equipment to ensure the continuity and accuracy of data acquisition, covering the entire operation cycle of the energy storage device, including all stages of charging and discharging as well as the resting state.

[0030] When the real-time capacity decay rate curve exceeds the preset decay threshold and the internal impedance mutation amplitude is within a stable fluctuation range, the capacity loss compensation function in the capacity decay prediction model is activated. The preset decay threshold is a critical value determined based on the design life of the energy storage device, material characteristics, and actual operating experience. When the decay rate exceeds this value, it means that the capacity decay has entered a stage requiring intervention. The internal impedance mutation amplitude being within a stable fluctuation range indicates that the impedance change is relatively gradual and will not pose an urgent threat to the device's operation. In this case, the capacity loss compensation function is prioritized to address the capacity decay problem. The capacity loss compensation function includes compensation rules corresponding to different decay rates. These rules are formulated based on historical data and model training results, and can calculate a suitable compensation scheme according to the current decay situation, thereby generating dynamic adjustment instructions for the charging and discharging strategy.

[0031] When the internal impedance surge exceeds the preset impedance tolerance and the real-time capacity decay rate curve is within a controllable fluctuation range, the impedance fluctuation mapping table in the voltage anomaly identification model is activated. The preset impedance tolerance takes into account the maximum allowable fluctuation range of the internal impedance during normal operation of the energy storage device; exceeding this range may indicate an internal fault or anomaly. The real-time capacity decay rate curve being within a controllable fluctuation range indicates that the capacity decay is relatively stable and does not require priority handling. The impedance fluctuation mapping table records impedance control rules corresponding to different impedance surge amplitudes. These rules are derived by analyzing a large amount of historical impedance anomaly data and corresponding handling measures, and can generate corresponding protection mechanism trigger commands based on the current impedance surge situation.

[0032] If both the real-time capacity decay rate curve and the internal impedance abrupt change exceed the critical threshold, the dynamic adjustment command generated by the capacity loss compensation function will be executed first, while the trigger command for the impedance fluctuation mapping table will be delayed until the charge / discharge strategy completes its correction operation. This priority setting is based on the fact that handling capacity decay issues often requires a continuous adjustment process. Executing two commands simultaneously may lead to control conflicts and affect the adjustment effect. Delaying the execution of the impedance fluctuation mapping table trigger command ensures that the correction operation of the charge / discharge strategy is not disturbed. After the capacity decay situation is initially controlled, the relevant commands of the protection mechanism will be executed to achieve comprehensive protection of the equipment.

[0033] When a risk warning command is input to the control unit of the energy storage device, and the charging / discharging strategy is adjusted in real time or a protection mechanism is activated, the current output amplitude in the charging / discharging strategy is adjusted in stages according to the attenuation compensation amount in the dynamic adjustment command. The attenuation compensation amount is the current amplitude that needs to be adjusted, calculated by the capacity loss compensation function based on the current capacity attenuation. Staged adjustment is to avoid sudden and large changes in current that could impact the energy storage device. The adjustment amplitude and duration of each stage are determined based on the device's response characteristics. After each current adjustment, the actual capacity attenuation data of the energy storage unit is collected and the offset is calculated by comparing it with the predicted value of the capacity loss compensation function. The offset is the difference between the actual value and the predicted value; calculating this value helps determine whether the current adjustment direction is correct.

[0034] If the offset continues to converge, meaning the absolute value of the offset gradually decreases, it indicates that the current adjustment direction meets the actual needs and can effectively alleviate capacity decay. In this case, maintain the current adjustment direction until the target capacity retention rate range is reached. The target capacity retention rate range is set according to the equipment's usage requirements and performance standards to ensure stable operation within this range. If the offset shows a divergent trend, meaning the absolute value of the offset gradually increases, it indicates a deviation in the current adjustment direction. The current output amplitude needs to be adjusted in the opposite direction, simultaneously triggering the weighted iteration of the capacity loss compensation function. Weighted iteration allows the capacity loss compensation function to self-correct based on new operating data, improving the accuracy of subsequent predictions and compensations.

[0035] The action threshold of the protection mechanism is dynamically adjusted based on the impedance control parameters in the protection mechanism trigger command. The impedance control parameters are determined by the impedance fluctuation mapping table based on the current impedance abrupt changes and are used to guide the adjustment of the protection mechanism's action threshold. Dynamic adjustment of the action threshold allows the protection mechanism to better adapt to the current equipment operating state, avoiding untimely or erroneous protection due to a fixed threshold. During threshold adjustment, changes in internal impedance are monitored in real time using electrochemical impedance spectroscopy. Electrochemical impedance spectroscopy provides impedance information at different frequencies, comprehensively reflecting changes in internal impedance. The control coefficients in the impedance fluctuation mapping table are updated based on the monitoring results, enabling the mapping table to adapt to changes in equipment operating state and improving the accuracy of subsequent impedance anomaly identification and handling.

[0036] Example 3: See Figure 4 and Figure 5During the feedback calibration phase following the execution of the early warning command, the final capacity retention rate data and internal impedance stability spectrum of the energy storage device after completing a charge-discharge cycle are collected. The final capacity retention rate data refers to the ratio of the remaining capacity of the energy storage device to its initial capacity after a complete charge-discharge cycle. This data is obtained through a dedicated capacity detection module, which must collect data during the stabilization period after the end of the charge-discharge cycle to avoid the influence of instantaneous fluctuations. The internal impedance stability spectrum is obtained through electrochemical impedance spectroscopy, recording the amplitude and phase information of the internal impedance at different frequencies. The test must be performed after the device temperature has recovered to ambient temperature and the internal chemical reaction has stabilized to ensure that the spectrum accurately reflects the impedance characteristics of the device in a stable state.

[0037] A difference analysis is performed between the final capacity retention rate data and the prediction range of the capacity decay prediction model to generate a capacity prediction error signal. This difference analysis is achieved by calculating the difference between the final capacity retention rate data and the upper and lower limits of the prediction range. If the actual data is within the prediction range, the error signal mainly reflects the degree of deviation from the predicted center value; if the actual data exceeds the prediction range, the error signal must include both the magnitude and direction of the deviation. The capacity prediction error signal is presented in the form of time-series data, encompassing the error changes throughout the analysis process, providing detailed information for subsequent model correction.

[0038] The impedance identification error signal is generated by calculating the overlap between the internal impedance stability spectrum and the expected impedance template of the voltage anomaly identification model. The overlap calculation is performed by comparing the impedance amplitude and phase of the two at the same frequency points using a point-by-point matching method. After calculating the difference at each frequency point, the overall overlap index is obtained. The impedance identification error signal not only contains overall overlap information but also records frequency ranges with significant differences. These ranges often correspond to weak points in the model's identification and are important areas of focus for model optimization.

[0039] Based on the systematic deviation component in the capacity prediction error signal, the attenuation compensation rule in the capacity attenuation prediction model is modified. The systematic deviation component refers to the persistent and regular deviation exhibited in the capacity prediction error signal. This deviation is usually caused by the structure or parameter settings of the model itself. By analyzing the systematic deviation component, the parts of the attenuation compensation rule that need adjustment are determined. For example, if the error signal shows that the predicted value is consistently lower than the actual value, certain coefficients in the compensation rule need to be increased to improve the accuracy of the prediction.

[0040] Based on the random fluctuation components in the impedance identification error signal, the impedance control parameters in the voltage anomaly identification model are optimized. Random fluctuation components refer to irregular, sudden deviations in the error signal, mainly caused by random factors such as measurement noise and environmental interference. For these components, their distribution characteristics are determined through statistical analysis, and then the impedance control parameters are adjusted to enable the model to better filter random interference and improve its ability to identify true impedance changes.

[0041] The revised attenuation compensation rules and impedance control parameters are synchronously updated to the historical parameter database of the multi-source operating parameter set. The update process must ensure data integrity and consistency, adding a timestamp and corresponding operating condition label to each correction item for subsequent traceability and analysis. Updating the historical parameter database enables the new model parameters to play a role in subsequent early warning and monitoring processes, achieving continuous model optimization.

[0042] In the process of revising the attenuation compensation rule and optimizing the impedance control parameters, the steady-state offset component in the capacity prediction error signal is identified, and the steady-state compensation amount is calculated using a sliding time window algorithm. The steady-state offset component is the stable part of the system deviation component, which does not change significantly over time. The sliding time window algorithm calculates the steady-state offset estimate corresponding to the window by averaging the error data within the fixed-length time window. As the window slides, the estimate is continuously updated to finally determine the steady-state compensation amount.

[0043] The reference current output value in the capacity loss compensation function should be adjusted based on the steady-state compensation amount. The reference current output value is a fundamental parameter in the capacity loss compensation function and directly affects the compensation effect. During adjustment, the reference current output value needs to be corrected proportionally according to the magnitude and direction of the steady-state compensation amount so that the compensation function can more accurately reflect the actual capacity loss.

[0044] High-frequency noise components in impedance identification error signals are identified, and effective correction values ​​are extracted using wavelet denoising algorithms. High-frequency noise components typically manifest as rapid fluctuations in the error signal, with frequencies higher than the normal impedance change frequency. Wavelet denoising algorithms decompose the error signal into different frequency scales, isolate and suppress the high-frequency noise components, and then reconstruct the signal to obtain the noise-removed effective components. From these, effective correction values ​​for adjusting impedance control parameters are extracted.

[0045] The action threshold weights in the impedance fluctuation mapping table are adjusted based on the effective correction amount. These action threshold weights determine the sensitivity of the protection mechanism under different impedance changes. By adjusting these weights, the protection mechanism can better adapt to the actual impedance change characteristics. For example, if the effective correction amount indicates that impedance changes within a certain frequency range are easily misjudged, the action threshold weights within that range are increased accordingly to reduce the possibility of false tripping.

[0046] The updated capacity loss compensation function and impedance fluctuation mapping table will be integrated into the early warning model parameter library. During integration, the new functions and mapping tables need to be formatted and standardized to ensure compatibility with other data in the parameter library. Simultaneously, an index linking new and historical parameters will be established to facilitate rapid retrieval and loading of appropriate parameters during model calls, ensuring that the early warning model can promptly apply the latest optimization results.

[0047] In the above process, the following formula can be used to calculate the steady-state compensation:

[0048] in: This represents the steady-state compensation amount. This indicates the number of data points contained within the sliding time window. Indicates the first Actual capacity retention rate of each data point Indicates the first Predictive capacity retention rate for each data point.

[0049] Example 4: When performing preprocessing operations before device startup, it is necessary to parse the cycle count code and health status decay segment from the energy storage device's charge / discharge history to generate a health feature vector. The cycle count code is a digital identifier of the number of charge / discharge cycles the device has undergone. For example, if a device has accumulated 1200 charge / discharge cycles, its code might be "CYC-1200". The health status decay segment divides the device's health status into different stages as the number of cycles decreases, such as "initial stable segment (0-300 cycles)", "slow decay segment (301-800 cycles)", and "rapid decay segment (801 cycles and above)". By extracting the decay segment features corresponding to the cycle count, such as the average capacity loss rate and maximum voltage fluctuation value within the decay segment, these features are quantified and combined to form a health feature vector. For example, a device's vector might contain information such as "cycle count 1200", "current decay segment is rapid decay segment", "average capacity loss rate of this segment 0.02% / cycle", and "maximum voltage fluctuation 0.3V".

[0050] The health feature vector is input into a pre-built operating mode database for similarity retrieval, filtering out candidate baseline parameter sets whose matching degree with the current health feature vector exceeds a preset threshold. The operating mode database stores a large amount of historical operating data for similar devices, with each data entry containing a corresponding health feature vector and a corresponding baseline parameter set. During retrieval, the similarity between the current vector and each vector in the database is calculated, for example, by comparing feature values ​​one by one. When the differences between a historical vector and the current vector in features such as the number of cycles, decay range, and capacity loss rate are all within a set range, it is determined to be a high match. If the preset threshold is 85%, then the baseline parameter sets corresponding to all historical data with a matching degree of 85% or higher are included in the candidate range.

[0051] Extract the historical early warning accuracy and false alarm suppression rate for each candidate baseline parameter from the candidate baseline parameter set, and calculate the comprehensive early warning effectiveness score. Historical early warning accuracy refers to the proportion of times the baseline parameter correctly issued an early warning out of the total number of early warnings in past use, while false alarm suppression rate refers to the proportion of times a false alarm was successfully avoided out of the possible false alarms. The comprehensive score can be calculated by combining the weights of both. For example, if a candidate parameter has a historical early warning accuracy of 90% and a false alarm suppression rate of 85%, and both have a weight of 0.5, then the comprehensive score is (90% × 0.5 + 85% × 0.5) = 87.5 points.

[0052] The candidate benchmark parameter sets are sorted according to the comprehensive early warning effectiveness score, and the candidate benchmark parameter with the highest score is selected as the optimal charge and discharge benchmark template. For example, if there are three parameter sets A, B, and C in the candidate set, with comprehensive scores of 87.5, 82, and 85 respectively, then A is selected as the optimal charge and discharge benchmark template. This template includes specific parameters such as the upper and lower limits of charge and discharge current, voltage protection value, and temperature monitoring range.

[0053] Extract the set of impedance regulation reference curves associated with the optimal charge / discharge reference template from the operating mode database. Each curve records the change of the device's internal impedance over time under the corresponding charge / discharge reference. For example, the set of curves associated with template A may contain 5 curves, each corresponding to the impedance change trajectory under different ambient temperatures.

[0054] The synergy between the charging / discharging strategy and impedance control is verified for each curve in the impedance control reference curve set. Abnormal curves with strategy conflicts or impedance abrupt changes are eliminated, and an optimized impedance control benchmark set is generated. Taking a certain curve to be verified as an example, its impedance control data is extracted, such as the impedance value recorded every 10 minutes. The time-aligned charging / discharging strategy sequence in the optimal charging / discharging benchmark template is obtained simultaneously, such as the current adjustment command within the same time period.

[0055] Based on the phase nodes of the charge / discharge strategy sequence, cooperative timestamp identifiers are marked on the curve to be verified to generate an impedance control curve with time sequence markings. For example, if the charge / discharge strategy has current adjustment actions at the 30-minute and 60-minute marks, these two time points are the phase nodes, and "T30" and "T60" are marked on the curve at the corresponding positions as cooperative timestamps.

[0056] The algorithm iterates through each coordinated timestamp identifier in the impedance control curve with time-series markers, detecting whether the control change gradient between adjacent time intervals exceeds the abrupt change tolerance, and identifying abnormal intervals with strategy conflicts. For example, if the impedance value abruptly changes from 2Ω to 5Ω between T30 and T60, with a change gradient of 3Ω / 30 minutes, and the abrupt change tolerance is 2Ω / 30 minutes, then this interval is marked as an abnormal interval.

[0057] If an abnormal interval is detected, the charging and discharging strategy value of the corresponding node in the optimal charging and discharging reference template is traced back to determine whether the direction of the charging and discharging strategy and the direction of impedance control produce reverse interference. For example, at time T30, the charging and discharging strategy command is to increase the current, while the simultaneous impedance control requires to reduce the current to suppress the rise in impedance. The two are in opposite directions, which means reverse interference occurs.

[0058] When the intensity of reverse interference exceeds the conflict threshold, the abnormal interval is marked as a strategy conflict domain, and its starting position and duration are calculated. If the conflict threshold is set to "continuous reverse interference for more than 15 minutes", and the above abnormal interval lasts from T30 to T50, a duration of 20 minutes, then this interval is marked as a strategy conflict domain.

[0059] Based on the spatiotemporal attributes of the strategy conflict domain, a control correction interval is defined on the curve to be verified, and an alternative control smoothing segment is generated based on historical conflict solutions. For example, if the strategy conflict domain is T30-T50, the control correction interval can be set to T25-T55. Referring to the solutions to similar conflicts in history, an alternative curve segment that gradually rises from 2Ω to 3.5Ω is generated to avoid abrupt changes.

[0060] The alternative control smoothing segment is inserted into the control correction interval to generate an optimized impedance control curve, and abnormal data points in the original curve that overlap with the strategy conflict domain are deleted. The optimized impedance control curve after the substitution insertion operation is performed on its integrity, such as checking whether the curve is continuous and whether the data points are complete. Curves with uncorrected conflict domains are removed. For example, if a curve still has a 10-minute conflict interval that has not been processed after correction, it is removed.

[0061] The curves that pass the merge verification are used to generate an optimized impedance control benchmark set. Based on the historical stability index of each curve in the optimized impedance control benchmark set, the curve with the smallest variance is selected as the final impedance control benchmark curve. For example, if the variances of the three curves in the set are 0.8, 1.2, and 0.6 respectively, then the curve with a variance of 0.6 is selected.

[0062] The optimal charge / discharge reference template and the final impedance control reference curve are time-aligned to generate an initial configuration for a multi-source operating parameter set. For example, it is ensured that the T30 time point in the template corresponds to the T30 time point in the curve, so that the charge / discharge parameters and impedance change data are consistent in the time dimension, providing an initial parameter basis for real-time monitoring after the equipment starts up.

[0063] Example 5: The acquisition process of multi-dimensional operating parameter sequences begins with capturing the transient response waveforms during the charging and discharging process of the energy storage device using a high-frequency sampling module. The operation of the high-frequency sampling module is not fixed; its sampling frequency is dynamically adjusted according to the charging and discharging rate characteristics of the energy storage device. When the energy storage device is in the rapid charging and discharging phase, the sampling frequency is increased to accurately capture every minute instantaneous change due to the rapid changes in the device's internal state. Conversely, when the device is in the stable operating phase, the changes in the internal state tend to be gradual, and the sampling frequency is appropriately reduced to avoid generating too much unnecessary data and causing data redundancy.

[0064] After capturing the transient response waveform, several key features need to be extracted, including the rising edge slope, oscillation decay coefficient, and steady-state duration. The rising edge slope is calculated using a piecewise linear fitting method. This method effectively eliminates noise interference that may exist in the initial stage of the waveform. By linearly fitting different segments of the rising edge and then comprehensively calculating, a relatively accurate rising edge slope is obtained. The oscillation decay coefficient is obtained by exponentially fitting the oscillation portion of the waveform. Exponential fitting can better fit the trend of oscillation decay, thus obtaining a coefficient that reflects the oscillation decay characteristics. The steady-state duration is determined by setting an amplitude fluctuation threshold and statistically analyzing the duration of the waveform within this threshold range to judge the waveform's maintenance in the steady-state phase.

[0065] After extracting the transient response waveform features, these features are fused with current fluctuation time-series data, voltage offset distribution curves, and temperature gradient evolution maps to generate an enhanced operational feature tensor. The current fluctuation time-series data records the changes in current over time during equipment operation, the voltage offset distribution curve reflects the distribution characteristics of voltage deviations from the normal range, and the temperature gradient evolution map shows the evolution of temperature gradients at different parts of the equipment over time. Fusing the transient response waveform features with this data integrates multiple aspects of information to form a richer, more comprehensive enhanced operational feature tensor that reflects the operational status of the energy storage device.

[0066] An enhanced operational feature tensor is input into the feature selection layer, where redundant dimensions are filtered out using an importance weighting algorithm. The feature selection layer's role is to select feature dimensions that are more important in reflecting the equipment's operational status. The importance weighting algorithm calculates the corresponding weights based on the correlation between each dimension's parameters and capacity decay and impedance changes. The correlation is evaluated using a mutual information entropy algorithm. This algorithm measures the dependency between two variables by calculating the mutual information entropy between each dimension's parameters and capacity decay and impedance changes. Dimensions with weights below a set threshold are considered redundant and are therefore filtered out.

[0067] Following the filtering process described above, a dimensionality-reduced core parameter matrix is ​​generated and updated into the multi-source operating parameter set. This dimensionality-reduced core parameter matrix retains crucial information for assessing equipment operating status while removing redundant data, making the multi-source operating parameter set more refined and representative. This provides more effective data support for real-time early warning and monitoring of energy storage equipment operating status. The entire process is interconnected, from data collection and feature extraction to data fusion and filtering; each step aims to obtain key parameters that accurately reflect the equipment's operating status.

[0068] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0069] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for real-time early warning and monitoring of the operating status of energy storage devices based on an LSTM hybrid model, characterized in that, include: Collect multi-dimensional operating parameter sequences of energy storage devices to generate a multi-source operating parameter set, which includes current fluctuation time series data, voltage offset distribution curves, and temperature gradient evolution maps; Based on the current fluctuation time series data in the multi-source operating parameter set, a capacity decay prediction model dominated by a long short-term memory network is constructed. The capacity decay prediction model is used to capture the dynamic mapping relationship between current fluctuation and capacity loss. Based on the voltage offset distribution curve in the multi-source operating parameter set, a time-series convolution-assisted voltage anomaly identification model is constructed. The voltage anomaly identification model is used to extract the nonlinear correlation rules between voltage offset and internal impedance change. Real-time monitoring of the capacity decay trajectory and internal impedance fluctuation data of energy storage devices; dynamic activation of the target model in the capacity decay prediction model or voltage anomaly identification model; and generation of risk warning instructions through the target model. The risk warning command is input to the control unit of the energy storage device to adjust the charging and discharging strategy or activate the protection mechanism in real time.

2. The real-time early warning and monitoring method for the operating status of energy storage devices based on an LSTM hybrid model according to claim 1, characterized in that, The multi-dimensional operating parameter sequence of the energy storage device is collected to generate a multi-source operating parameter set, including: Electrochemical characteristic data of the energy storage unit during charge-discharge cycles are collected synchronously by distributed sensors. The electrochemical characteristic data includes polarization impedance spectrum and ion diffusion rate parameters. Based on the peak distribution range in the polarization impedance spectrum, the energy storage unit is divided into at least two operating state clusters, and an initial charge and discharge threshold range and a temperature control reference value are assigned to each operating state cluster. Extract the capacity attenuation compensation coefficient and voltage offset correction parameter that match each operating state cluster from the historical parameter database; The capacity attenuation compensation coefficient is normalized in the time domain to generate an optimized capacity attenuation prediction sequence for each operating state cluster. The voltage offset correction parameters are subjected to frequency domain feature enhancement processing to generate optimized voltage anomaly identification curves corresponding to each operating state cluster; By integrating the optimized capacity decay prediction sequence and optimized voltage anomaly identification curve of each operating state cluster, the multi-source operating parameter set is generated, and the historical operating mode migration trajectory is associated with it.

3. The real-time early warning and monitoring method for the operating status of energy storage devices based on an LSTM hybrid model according to claim 2, characterized in that, The construction of the capacity decay prediction model and voltage anomaly identification model includes: The optimized capacity decay prediction sequence is input into a bidirectional long short-term memory network, and the hidden state weights are updated through a gated recurrent unit to generate the capacity loss compensation function in the capacity decay prediction model. The optimized voltage anomaly identification curve is input into a temporal convolutional network, and the feature extraction kernel is adjusted through a dilated convolutional layer to generate an impedance fluctuation mapping table in the voltage anomaly identification model. After the bidirectional long short-term memory network and the temporal convolutional network converge, the core feature tensors of the capacity loss compensation function and the impedance fluctuation mapping table are extracted respectively. The similarity between the core feature tensor and the real-time acquired electrochemical feature data is compared to verify the adaptability of the capacity decay prediction model and the voltage anomaly identification model. If the similarity comparison result is lower than the preset tolerance threshold, the hidden state vector of the bidirectional long short-term memory network is reinitialized until the core feature tensor satisfies the adaptability constraint.

4. The real-time early warning and monitoring method for the operating status of energy storage devices based on the LSTM hybrid model according to claim 3, characterized in that, The system monitors the capacity decay trajectory and internal impedance fluctuation data of the energy storage device in real time, dynamically activates the target model, and generates risk warning instructions, including: Continuously track the real-time capacity decay rate curve and internal impedance abrupt change amplitude of energy storage devices; When the real-time capacity decay rate curve exceeds the preset decay threshold and the internal impedance change amplitude is in a stable fluctuation range, the capacity loss compensation function in the capacity decay prediction model is activated. Based on the attenuation compensation rule in the capacity loss compensation function, a dynamic adjustment command for the charging and discharging strategy is generated. When the amplitude of the internal impedance change exceeds the preset impedance tolerance and the real-time capacity decay rate curve is in the controllable fluctuation range, the impedance fluctuation mapping table in the voltage anomaly identification model is activated. Based on the impedance control rules in the impedance fluctuation mapping table, a protection mechanism trigger command is generated; If the real-time capacity decay rate curve and the internal impedance abrupt change amplitude both exceed the critical threshold, the dynamic adjustment instruction generated by the capacity loss compensation function will be executed first, and the trigger instruction of the impedance fluctuation mapping table will be delayed until the charging and discharging strategy completes the correction operation.

5. The real-time early warning and monitoring method for the operating status of energy storage devices based on the LSTM hybrid model according to claim 4, characterized in that, The step of inputting the risk warning command to the control unit of the energy storage device to adjust the charging and discharging strategy or activate the protection mechanism in real time includes: Based on the attenuation compensation amount in the dynamic adjustment command, the current output amplitude in the charging and discharging strategy is adjusted in stages. After each current adjustment, the actual capacity decay data of the energy storage unit is collected and the offset is calculated with the predicted value of the capacity loss compensation function. If the offset continues to converge, maintain the current adjustment direction until the target capacity retention range is reached; If the offset shows a diverging trend, the current output amplitude is adjusted in the opposite direction and the weight iteration of the capacity loss compensation function is triggered. The action threshold of the protection mechanism is dynamically adjusted according to the impedance control parameter in the protection mechanism trigger command. During the threshold adjustment process, the internal impedance change is monitored in real time by electrochemical impedance spectroscopy, and the control coefficient in the impedance fluctuation mapping table is updated according to the monitoring results.

6. The real-time early warning and monitoring method for the operating status of energy storage devices based on the LSTM hybrid model according to claim 1, characterized in that, It also includes a feedback calibration phase after the early warning command is executed: Collect the final capacity retention rate data and internal impedance stability spectrum of the energy storage device after it has completed a charge-discharge cycle; The final capacity retention rate data is compared with the prediction range of the capacity decay prediction model to generate a capacity prediction error signal. The overlap between the internal impedance stability spectrum and the expected impedance template of the voltage anomaly identification model is calculated to generate an impedance identification error signal. Based on the system deviation component in the capacity prediction error signal, the attenuation compensation rule in the capacity attenuation prediction model is corrected; Based on the random fluctuation component in the impedance identification error signal, optimize the impedance control parameters in the voltage anomaly identification model; The revised attenuation compensation rules and impedance control parameters are synchronously updated to the historical parameter library of the multi-source operating parameter set.

7. The real-time early warning and monitoring method for the operating status of energy storage devices based on an LSTM hybrid model according to claim 6, characterized in that, The process of revising the attenuation compensation rule and optimizing the impedance control parameters includes: Identify the steady-state offset component in the capacity prediction error signal and calculate the steady-state compensation amount using a sliding time window algorithm; Adjust the reference current output value in the capacity loss compensation function according to the steady-state compensation amount; The high-frequency noise component in the impedance identification error signal is identified, and the effective correction amount is extracted using a wavelet denoising algorithm. Adjust the action threshold weights in the impedance fluctuation mapping table according to the effective correction amount; The updated capacity loss compensation function and impedance fluctuation mapping table are integrated into the early warning model parameter library.

8. The method for real-time early warning and monitoring of the operating status of energy storage devices based on an LSTM hybrid model according to claim 1, characterized in that, This also includes performing preprocessing operations before the device starts up: Analyze the cycle count encoding and health status decay segment in the charge and discharge history of energy storage devices to generate a health feature vector; The health feature vector is input into a pre-built operating mode database for similarity retrieval, and a set of candidate benchmark parameters with a matching degree exceeding a preset threshold is selected. Extract the historical early warning accuracy and false alarm suppression rate of each candidate benchmark parameter in the candidate benchmark parameter set, and calculate the comprehensive early warning effectiveness score; The candidate benchmark parameter set is sorted according to the comprehensive early warning effectiveness score, and the candidate benchmark parameter with the highest score is selected as the optimal charging and discharging benchmark template. Extract the set of impedance regulation reference curves associated with the optimal charge / discharge reference template from the operating mode database; The synergy between the charging / discharging strategy and impedance control is verified for each curve in the impedance control reference curve set. Abnormal curves with strategy conflicts or impedance abrupt changes are eliminated to generate an optimized impedance control benchmark set. Based on the historical stability index of each curve in the optimized impedance control reference set, the curve with the smallest variance is selected as the final impedance control reference curve. The optimal charge / discharge reference template and the final impedance control reference curve are time-aligned to generate an initial configuration for a multi-source operating parameter set.

9. The real-time early warning and monitoring method for the operating status of energy storage devices based on an LSTM hybrid model according to claim 8, characterized in that, The verification of the synergy between the charge / discharge strategy and impedance control for each curve in the impedance control reference curve set includes: Extract the impedance control data of a single curve to be verified, and simultaneously obtain the time-aligned charge and discharge strategy sequence in the optimal charge and discharge reference template; Based on the phase nodes of the charge-discharge strategy sequence, a cooperative timestamp identifier is marked on the curve to be verified to generate an impedance control curve with timing markers. Traverse each cooperative timestamp identifier in the impedance control curve with time-series markers, detect whether the control change gradient between adjacent time intervals exceeds the mutation tolerance, and identify abnormal intervals with strategy conflicts. If an abnormal interval is detected, the charging and discharging strategy value of the corresponding node in the optimal charging and discharging reference template is traced back to determine whether the charging and discharging strategy direction and the impedance control direction produce reverse interference. When the reverse interference intensity exceeds the conflict threshold, the abnormal interval is marked as a strategy conflict domain, and the starting position and duration of the strategy conflict domain are calculated. Based on the spatiotemporal attributes of the strategy conflict domain, a control correction interval is defined on the curve to be verified, and an alternative control smoothing segment is generated based on historical conflict solutions. The alternative control smoothing segment is inserted into the control correction interval to generate an optimized impedance control curve, and abnormal data points that overlap with the strategy conflict domain in the original curve are deleted. The integrity of the optimized impedance control curve after the substitution insertion operation is verified, and curves with residual uncorrected conflict regions are removed. The curves that pass the merge verification are used to generate the optimized impedance control reference set.

10. The method for real-time early warning and monitoring of the operating status of energy storage devices based on an LSTM hybrid model according to claim 1, characterized in that, The acquisition process of the multidimensional operating parameter sequence includes: capturing the transient response waveform during the charging and discharging process of the energy storage device through a high-frequency sampling module, extracting the rising slope, oscillation attenuation coefficient and steady-state duration of the transient response waveform; and fusing the transient response waveform features with current fluctuation time series data, voltage offset distribution curve and temperature gradient evolution map to generate an enhanced operating feature tensor. The enhanced runtime feature tensor is input into the feature selection layer, and redundant dimensions are filtered out through an importance weight allocation algorithm to generate a dimensionality-reduced core parameter matrix, which is then updated to the multi-source runtime parameter set.

Images