Online monitoring method and system applied to storage battery

Abstract

The embodiment of the invention discloses an online monitoring method and system applied to a storage battery. The method comprises the following steps: acquiring a multi-source time sequence monitoring data set of a target storage battery; calling a pre-trained feature extraction network to perform time sequence feature extraction processing on the multi-source time sequence monitoring data set, and generating an internal resistance state feature vector, an electrolyte activity feature vector and a polar plate corrosion feature vector; performing state fusion analysis processing on the internal resistance state feature vector, the electrolyte activity feature vector and the polar plate corrosion feature vector to generate a comprehensive state evaluation vector of the target storage battery; and generating an online monitoring result of the target storage battery based on the comprehensive state evaluation vector, wherein the online monitoring result comprises an abnormal state early warning instruction and a maintenance optimization strategy instruction. Therefore, the defects in the aspect of storage battery state monitoring in the prior art are effectively overcome.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of data monitoring technology, and specifically to an online monitoring method and system for batteries. Background Art

[0002] In the field of battery monitoring, accurate monitoring of battery status is crucial, as batteries are key energy storage components in numerous devices and systems. As industries increasingly rely on batteries, the need for battery status monitoring is becoming increasingly urgent. Accurately understanding battery status helps identify potential problems promptly and plan maintenance accordingly, thereby ensuring the stable operation of equipment and systems.

[0003] Investigations and research have revealed that most existing monitoring methods focus solely on a single dimension of data, making it difficult to fully reflect the battery's true condition. For example, monitoring only internal resistance changes overlooks other important factors, such as electrolyte activity and plate corrosion, leading to inability to promptly detect potential faults. Furthermore, existing technologies lack precision in data processing and feature extraction, making it difficult to effectively extract key information from complex monitoring data. Furthermore, there is a lack of effective fusion analysis methods for feature data from different dimensions, hindering a comprehensive assessment of the battery's overall condition. Summary of the Invention

[0004] The embodiments of the present application provide an online monitoring method and system for batteries, which is used to obtain a multi-source time-series monitoring data set, use a pre-trained feature extraction network to accurately extract key feature vectors and perform fusion analysis, and ultimately generate comprehensive and practical online monitoring results, effectively overcoming the shortcomings of existing technologies in battery status monitoring.

[0005] In a first aspect, an embodiment of the present application provides an online monitoring method for a battery, which is applied to an online monitoring system, and the method includes: obtaining a multi-source time series monitoring data set of a target battery; calling a pre-trained feature extraction network to perform time series feature extraction processing on the multi-source time series monitoring data set to generate an internal resistance state feature vector, an electrolyte activity feature vector and a plate corrosion feature vector; performing state fusion analysis processing on the internal resistance state feature vector, the electrolyte activity feature vector and the plate corrosion feature vector to generate a comprehensive state evaluation vector of the target battery; generating an online monitoring result of the target battery based on the comprehensive state evaluation vector, and the online monitoring result includes an abnormal state warning instruction and a maintenance optimization strategy instruction.

[0006] In a second aspect, an embodiment of the present application provides an online monitoring system, comprising: processor; a storage device having a computer program stored thereon, When the computer program is executed by the processor, the processor implements any of the online monitoring methods for batteries.

[0007] An embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the online monitoring method applied to a battery are implemented.

[0008] It can be seen that the embodiments of the present application have the following beneficial effects: first, a multi-source time series monitoring data set of the target battery is obtained; secondly, by calling a pre-trained feature extraction network to perform time series feature extraction processing on the multi-source time series monitoring data set, the internal resistance state feature vector, the electrolyte activity feature vector and the plate corrosion feature vector can be accurately extracted from the complex data to characterize the state of the battery from different key angles; then, through state fusion analysis and processing, the internal resistance state feature vector, the electrolyte activity feature vector and the plate corrosion feature vector are organically integrated to generate a comprehensive state evaluation vector that comprehensively and systematically reflects the overall condition of the target battery; finally, based on the comprehensive state evaluation vector, online monitoring results are generated, which can not only issue abnormal state warning instructions in time to provide advance notification, but also give maintenance optimization strategy instructions to guide the reasonable planning of maintenance work, improve the service life and performance of the battery, and ensure its stable and reliable operation.

[0009] In summary, the embodiments of the present application obtain a multi-source time-series monitoring data set, use a pre-trained feature extraction network to accurately extract key feature vectors, and perform fusion analysis to ultimately generate comprehensive and practical online monitoring results, effectively overcoming the shortcomings of existing technologies in battery status monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flowchart of an online monitoring method for a battery provided in an embodiment of the present application.

[0011] Figure 2 A schematic diagram of the basic structure of an online monitoring system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the embodiments of the present application are further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0013] See also Figure 1 As shown in FIG, this figure is a flow chart of an online monitoring method for a battery provided by an embodiment of the present application, which can be applied to an online monitoring system. Figure 1 As shown, the method includes steps 110 to 140.

[0014] Step 110: Acquire a multi-source time series monitoring data set of the target battery.

[0015] It can be understood that this step aims to obtain a multi-source time-series monitoring data set that can comprehensively reflect the state of the target battery. This data set is crucial for the subsequent accurate assessment of the battery state and provides basic data support for in-depth analysis of the battery's internal resistance changes, electrolyte activity, and plate corrosion.

[0016] In one example, the target battery is a lead-acid battery, and the multi-source time series monitoring data set includes internal resistance change time series data, electrolyte activity time series data, and plate corrosion time series data. Then, obtaining the multi-source time series monitoring data set of the target battery includes: Step 111: Periodically collect a first raw data stream from the internal resistance sensor of the target battery, and generate internal resistance change time series data based on the first raw data stream; the first raw data stream includes a sequence of internal resistance measurement values of the target battery during different charge and discharge cycles.

[0017] In this step, the internal resistance sensor measures the internal resistance of the lead-acid battery in different charge and discharge cycles at certain time intervals, for example, every t1 minutes. In detail, in the nth charge and discharge cycle, the internal resistance measured by the internal resistance sensor at the mth measurement moment after the start of the cycle is {Rn, m}. As time goes by and the charge and discharge cycles continue, a series of internal resistance measurement values will be obtained, which constitute the first original data stream. Then, these internal resistance measurement values arranged in chronological order are organized into internal resistance change time series data, which contains the internal resistance information at each time point under different charge and discharge cycles. For example, it can be presented in the form of a list, and each element contains the charge and discharge cycle number, the measurement time, and the corresponding internal resistance measurement value.

[0018] Step 112: Periodically collect a second raw data stream from the electrolyte sensor of the target battery, and generate electrolyte activity time series data based on the second raw data stream; the second raw data stream includes a sequence of changes in electrolyte ion concentration of the target battery at different ambient temperatures.

[0019] In an embodiment of the present application, the electrolyte sensor measures the electrolyte ion concentration of the lead-acid battery at different ambient temperatures with a period of t2 minutes. When the ambient temperature is T1, the electrolyte ion concentration measured at the pth measurement moment is {C1, p}, and when the ambient temperature is T2, the electrolyte ion concentration at the qth measurement moment is {C2, q}. With multiple measurements under different ambient temperature conditions, a large number of ion concentration measurement values will be accumulated to form a second raw data stream. Afterwards, these measurement values are sorted according to the time sequence and the corresponding ambient temperature to generate electrolyte activity time series data. The data records the changes in electrolyte ion concentration at each time point under different ambient temperatures in an ordered form, and can be stored in a list form similar to the internal resistance change time series data, where each element contains the ambient temperature value, the measurement time and the corresponding electrolyte ion concentration.

[0020] Step 113: Periodically collect a third raw data stream from the plate sensor of the target battery, and generate plate corrosion time series data based on the third raw data stream; the third raw data stream includes a plate thickness variation sequence of the target battery under multiple working load conditions.

[0021] The plate sensor measures the plate thickness of the lead-acid battery under various workload conditions every t3 minutes. For example, when the workload is W1, the plate thickness measured at the rth measurement moment is {D1, r}, and when the workload is W2, the plate thickness at the sth measurement moment is {D2, s}. After a long period of measurement under different workload conditions, a series of plate thickness measurements are obtained, forming the third raw data stream. These measurements are then organized chronologically and by workload conditions to form plate corrosion time series data. This data records the changes in plate thickness at various time points under different workload conditions using a suitable data structure, facilitating subsequent analysis of plate corrosion.

[0022] Step 114: Perform data alignment processing on the internal resistance change time series data, the electrolyte activity time series data and the plate corrosion time series data to synchronize the internal resistance measurement value sequence, the electrolyte ion concentration change sequence and the plate thickness change sequence in the timestamp dimension to generate the multi-source time series monitoring data set.

[0023] When performing data alignment processing, first determine a common time reference standard. Taking the set starting time point as the benchmark, for the internal resistance change time series data, electrolyte activity time series data, and plate corrosion time series data, find the timestamp corresponding to the starting time point in each data set. Then, match the data in the three data sets at the same time interval, such as in minutes. If there is no measurement value in one of the data sets at a certain moment, use a suitable interpolation method, such as linear interpolation, to supplement the missing value. In this way, the data from three different sources are synchronized in the timestamp dimension, so that each time point can correspond to the internal resistance measurement value, electrolyte ion concentration value, and plate thickness value at the same time, and finally generate a multi-source time series monitoring data set, which integrates data from different sensors and provides a unified and synchronized data basis for subsequent feature extraction and status assessment.

[0024] Step 120: Calling a pre-trained feature extraction network to perform time series feature extraction processing on the multi-source time series monitoring data set to generate an internal resistance state feature vector, an electrolyte activity feature vector, and a plate corrosion feature vector.

[0025] In an embodiment of the present application, the pre-trained feature extraction network is obtained through training with a large amount of data and can effectively extract valuable features from a multi-source time series monitoring data set. The function of the network is to convert the original time series data into feature vectors that can represent different aspects of the battery state. These feature vectors will provide key information for subsequent state fusion and evaluation.

[0026] In an optional embodiment, the calling of a pre-trained feature extraction network to perform time series feature extraction processing on the multi-source time series monitoring data set to generate an internal resistance state feature vector, an electrolyte activity feature vector, and a plate corrosion feature vector includes: Step 121: Input the internal resistance change time series data into the first long short-term memory unit of the feature extraction network, capture the steady-state dependence pattern and transient fluctuation pattern in the internal resistance measurement value sequence through the first long short-term memory unit, and generate the internal resistance state feature vector.

[0027] Optionally, the internal resistance change time series data is input into the first long short-term memory unit of the feature extraction network. The first long short-term memory unit will analyze the input internal resistance measurement value sequence. It will observe the relative stability of the internal resistance measurement value over a period of time, that is, the steady-state dependence mode. For example, in several consecutive charge and discharge cycles, the internal resistance measurement value remains within a relatively fixed range, which can be a steady-state mode. At the same time, attention will also be paid to sudden changes or short-term fluctuations in the internal resistance measurement value, that is, transient fluctuation modes. For example, in one of the charge and discharge cycles, the internal resistance suddenly rises or falls significantly. By learning and capturing these patterns, the first long short-term memory unit will encode this information into a vector, which can be an internal resistance state feature vector. It contains the key feature information of the internal resistance change and can reflect the long-term and short-term changes in the internal resistance.

[0028] Step 122: Input the electrolyte activity time series data into the second long short-term memory unit of the feature extraction network, and capture the periodic decay pattern and ambient temperature correlation pattern in the electrolyte ion concentration change sequence through the second long short-term memory unit to generate the electrolyte activity feature vector.

[0029] Optionally, the electrolyte activity time series data is input into the second long short-term memory unit of the feature extraction network, which conducts an in-depth analysis of the electrolyte ion concentration change sequence. As time goes by, the electrolyte ion concentration may have a periodic decay pattern. For example, the ion concentration will drop regularly at regular intervals. At the same time, attention will also be paid to the correlation pattern between ion concentration and ambient temperature. For example, when the ambient temperature rises, the ion concentration may have a corresponding change trend. By capturing these patterns, the second long short-term memory unit integrates the relevant information into a vector to form an electrolyte activity feature vector, which can reflect the changing characteristics of electrolyte activity over time and ambient temperature.

[0030] Step 123: Input the plate corrosion time series data into the third long short-term memory unit of the feature extraction network, and capture the load stress correlation pattern and the corrosion rate nonlinear change pattern in the plate thickness change sequence through the third long short-term memory unit to generate the plate corrosion feature vector.

[0031] Optionally, the plate corrosion time series data is fed into the third long short-term memory unit of the feature extraction network. The third long short-term memory unit studies the plate thickness change sequence and discovers the correlation pattern between the plate thickness change and the workload stress. For example, under high workload, the plate thickness may decrease faster. At the same time, it also notices the nonlinear change pattern of the corrosion rate, that is, the corrosion rate is not uniform and may change faster in some stages and slower in others. By learning these patterns, the third long short-term memory unit converts the relevant information into a vector, namely the plate corrosion feature vector, which reflects the key characteristics of the plate corrosion.

[0032] Step 130: Perform state fusion analysis on the internal resistance state feature vector, the electrolyte activity feature vector, and the plate corrosion feature vector to generate a comprehensive state assessment vector of the target battery.

[0033] In the embodiment of the present application, the three feature vectors are subjected to state fusion analysis processing with the aim of integrating feature information from different aspects to form an evaluation vector that can comprehensively reflect the comprehensive state of the target battery, thereby more accurately evaluating the overall health status of the battery.

[0034] In an optional technical solution, the state fusion analysis processing of the internal resistance state feature vector, the electrolyte activity feature vector, and the plate corrosion feature vector to generate a comprehensive state assessment vector of the target battery includes: Step 131: Perform a first cross-attention calculation on the internal resistance state feature vector and the electrolyte activity feature vector to obtain an internal resistance-electrolyte interaction feature vector; perform a second cross-attention calculation on the internal resistance state feature vector and the plate corrosion feature vector to obtain an internal resistance-plate interaction feature vector; perform a third cross-attention calculation on the electrolyte activity feature vector and the plate corrosion feature vector to obtain an electrolyte-plate interaction feature vector.

[0035] In an embodiment of the present application, a first cross-attention calculation is first performed, taking the internal resistance state eigenvector and the electrolyte activity eigenvector as input. During the calculation process, the correlation between the two vectors is analyzed. For example, it is seen how the change in internal resistance affects the electrolyte activity, and the reaction of the change in electrolyte activity to the internal resistance. Through this two-way attention and calculation, a new vector, the internal resistance-electrolyte interaction eigenvector, is obtained, which contains the characteristic information of the interaction between the internal resistance and the electrolyte activity.

[0036] Next, the second cross-attention calculation is performed, taking the internal resistance state feature vector and the plate corrosion feature vector as input. The correlation between the two is also analyzed, such as the impact of internal resistance changes on plate corrosion and the feedback of plate corrosion status on internal resistance. This calculation results in the internal resistance-plate interaction feature vector, which reflects the interaction between internal resistance and plate corrosion.

[0037] Finally, a third cross-attention calculation is performed, taking the electrolyte activity feature vector and the plate corrosion feature vector as input. This calculation analyzes the relationship between electrolyte activity and plate corrosion, such as how changes in electrolyte activity accelerate or slow plate corrosion, and the impact of plate corrosion on electrolyte activity. This calculation yields an electrolyte-plate interaction feature vector, which captures the interaction between these two aspects.

[0038] Step 132: Perform weighted fusion processing on the internal resistance-electrolyte interaction feature vector, the internal resistance-plate interaction feature vector, and the electrolyte-plate interaction feature vector to generate the comprehensive state evaluation vector.

[0039] It can be understood that after obtaining the three interactive feature vectors, a weighted fusion process is performed. A weight is assigned to each interactive feature vector, for example, a weight w1 is assigned to the internal resistance-electrolyte interactive feature vector, a weight w2 is assigned to the internal resistance-plate interactive feature vector, and a weight w3 is assigned to the electrolyte-plate interactive feature vector. The weights can be determined based on actual conditions and experience, or through some training algorithms to reflect the importance of each interactive feature vector in the comprehensive state assessment. Then, the three vectors are fused according to their respective weights, for example, the internal resistance-electrolyte interactive feature vector is multiplied by the weight w1, the internal resistance-plate interactive feature vector is multiplied by the weight w2, and the electrolyte-plate interactive feature vector is multiplied by the weight w3, and then they are spliced together to form a comprehensive state assessment vector, which integrates the information of the three interactive feature vectors and can comprehensively reflect the comprehensive state of the battery.

[0040] Step 140: Generate an online monitoring result of the target battery based on the comprehensive state evaluation vector, wherein the online monitoring result includes an abnormal state warning instruction and a maintenance optimization strategy instruction.

[0041] In an embodiment of the present application, online monitoring results are generated based on a comprehensive state evaluation vector, providing users with intuitive information about the target battery state and strategies for dealing with different states, helping users to understand the battery state in a timely manner and take appropriate measures.

[0042] As an implementation manner, generating the online monitoring result of the target battery based on the comprehensive state assessment vector includes: Step 141: Input the comprehensive state evaluation vector into a pre-trained state classifier, and determine the current state category of the target battery through the state classifier, where the current state category includes a normal attenuation state, a local abnormal state, and a global failure state.

[0043] Optionally, the comprehensive state assessment vector is input into a pre-trained state classifier. Having been trained with a large amount of historical data, the state classifier has learned the characteristic patterns of comprehensive state assessment vectors under different states. It then compares the input comprehensive state assessment vector with these learned patterns. For example, when the characteristics of the comprehensive state assessment vector closely match the pattern of a normal decay state, the state classifier will determine that the target battery is in a normal decay state. If the vector features indicate local abnormal changes in certain aspects, consistent with the pattern of a local abnormal state, the target battery is judged to be in a local abnormal state. If the vector features indicate serious problems in multiple key aspects, consistent with the pattern of a global failure state, the target battery is judged to be in a global failure state. In this way, the current state category of the target battery is determined.

[0044] Step 142: When the current state category is the local abnormal state, the abnormal source direction is determined according to the contribution weights of each interactive feature vector in the comprehensive state evaluation vector, and the abnormal state warning instruction containing the abnormal source direction is generated; when the current state category is the global failure state, the maintenance optimization strategy instruction is generated according to the attenuation rate of each interactive feature vector in the comprehensive state evaluation vector, and the maintenance optimization strategy instruction includes electrolyte replenishment cycle recommendations and plate replacement priority recommendations.

[0045] When a local abnormal state is determined, the various interactive eigenvectors in the comprehensive state assessment vector play a role in determining the source of the abnormality. For example, a high weight for the internal resistance-electrolyte interaction eigenvector indicates that the abnormality may stem from the interaction between the internal resistance and the electrolyte. By analyzing the contribution weights of this interaction eigenvector and other interaction eigenvectors, the source of the abnormality is determined. For example, abnormal changes in electrolyte activity may be the cause of the abnormal internal resistance. This abnormality source direction information is then integrated into the abnormal state warning instructions to inform the user of the possible problem direction.

[0046] When a global failure state is determined, maintenance optimization strategy instructions are generated based on the decay rates of the interactive eigenvectors within the comprehensive state evaluation vector. For example, if the plate corrosion eigenvector decays rapidly, indicating severe corrosion, the electrolyte replenishment cycle may be shortened to maximize electrolyte protection. Similarly, the plate replacement priority recommendation may prioritize plate replacements to resolve the plate failure as quickly as possible. By comprehensively considering the decay rates of the interactive eigenvectors, appropriate maintenance optimization strategy instructions are generated.

[0047] In an alternative embodiment, before inputting the integrated state evaluation vector into a pre-trained state classifier, the method further comprises: Step 210: Obtain a sample comprehensive state evaluation vector set and a corresponding sample state label set of historical battery samples, where each sample state label in the sample state label set is used to indicate an actual state category of the corresponding historical battery sample.

[0048] This step involves collecting a large amount of battery sample data from historical data. For each sample, a comprehensive sample state assessment vector is obtained using the same monitoring and feature extraction methods as the current target battery. At the same time, the actual state category of each sample is recorded to form a set of sample state labels. For example, if one of the historical battery samples is determined to be in a normal decay state after actual testing and analysis, the corresponding sample state label can be "normal decay state." By collecting multiple such sample data, a set of sample comprehensive state assessment vectors and a corresponding set of sample state labels are formed, which will be used to train the state classifier.

[0049] Step 220: Divide the sample comprehensive state evaluation vector set into a training set and a validation set, iteratively train the initial state classifier using the training set, and calculate the classification accuracy index using the validation set after each iteration.

[0050] Optionally, the sample comprehensive state evaluation vector set is divided into a certain proportion, for example, 70% as a training set and 30% as a validation set. The training set is used to train the initial state classifier. The initial state classifier has a series of parameters. During the training process, the sample comprehensive state evaluation vector in the training set is input into the classifier, and the classifier outputs a predicted state category. The predicted result is compared with the actual state category in the sample state label set, and the difference between the two is calculated, for example, by using the cross entropy loss function to measure the difference. The parameters of the classifier are then adjusted based on the difference, and multiple iterations of training are performed. After each iteration, the validation set is used to calculate the classification accuracy index. The sample comprehensive state evaluation vector in the validation set is input into the classifier, and the ratio of the number of samples correctly classified by the classifier to the total number of samples in the validation set is counted to obtain the classification accuracy index. The accuracy of the classifier is improved by continuously adjusting parameters and training.

[0051] As a preferred embodiment, the configuration process of the initial state classifier includes: Step 221: Configure a first fully connected layer to receive the comprehensive state evaluation vector and perform nonlinear transformation processing to generate a first hidden feature vector; configure a second fully connected layer to perform dimensionality compression processing on the first hidden feature vector to generate a second hidden feature vector; configure a third fully connected layer to perform probability mapping processing on the second hidden feature vector to generate a probability distribution vector of the current state category.

[0052] First, configure the first fully connected layer. Its input is the comprehensive state evaluation vector. Each neuron in the fully connected layer is connected to all elements of the input vector. The first fully connected layer applies a nonlinear transformation to the input vector, such as using the ReLU function. This nonlinear transformation enables the classifier to learn more complex feature relationships. After this transformation, the first hidden feature vector is generated.

[0053] Next, configure the second fully connected layer, which takes the first hidden feature vector as input. This layer performs dimensionality compression on the first hidden feature vector, reducing its dimensions while retaining key information. This dimensionality compression reduces computational complexity and prevents overfitting. After processing, the second hidden feature vector is generated.

[0054] Finally, the third fully connected layer is configured to receive the second hidden feature vector. It performs probability mapping on the second hidden feature vector, such as using a softmax function. This process converts the second hidden feature vector into a probability distribution vector for the current state category, where each element represents the probability of the target battery being in a different state category.

[0055] Step 222: Parameter optimization is performed on the first fully connected layer, the second fully connected layer, and the third fully connected layer according to the cross entropy loss between the probability distribution vector and the sample state label set; and the initial state classifier is obtained based on the optimized first fully connected layer, the second fully connected layer, and the third fully connected layer.

[0056] Next, the cross-entropy loss is calculated between the probability distribution vector and the set of sample state labels. This cross-entropy loss measures the degree of discrepancy between the predicted probability distribution and the actual labels. Based on this loss value, an optimization algorithm, such as stochastic gradient descent, is used to adjust the parameters of the first, second, and third fully connected layers. This process is repeated until the loss value reaches a smaller value or stops decreasing. At this point, the optimized first, second, and third fully connected layers form the initial state classifier, which can more accurately classify the input comprehensive state evaluation vector.

[0057] Step 230: When the classification accuracy index reaches a preset accuracy, the training is stopped to obtain the pre-trained state classifier.

[0058] It can be understood that during the continuous training process, the classification accuracy index calculated by the validation set after each iteration will be compared with the preset accuracy. The preset accuracy is a threshold set based on actual needs and experience, for example, it is set to 90%. As the training continues, the parameters of the classifier are constantly adjusted and optimized, and its classification ability of the sample comprehensive state evaluation vector is gradually improved. When the classification accuracy index reaches or exceeds the preset 90% for the first time, it is considered that the classifier has learned a sufficiently accurate classification pattern, and the training is stopped at this time. The state classifier obtained at this time can be a pre-trained state classifier, which can be used for subsequent state classification tasks of the target battery comprehensive state evaluation vector, and can more accurately judge whether the target battery is in a normal attenuation state, a local abnormal state, or a global failure state.

[0059] In a preferred design approach, the method further includes: Step 310: monitoring the execution status of the abnormal state warning instruction and the maintenance optimization strategy instruction in the online monitoring result of the target battery in real time.

[0060] During the entire operation of the monitoring system, it is necessary to monitor in real time whether the abnormal status warning instructions and maintenance optimization strategy instructions generated for the target battery are effectively executed. The system will continuously track the execution of these instructions, for example, by obtaining feedback information through the interactive interface with the executing equipment or personnel. For example, the abnormal status warning instruction is sent to the relevant monitoring equipment or the terminal device of the maintenance personnel. The system will check whether the terminal device has received the instruction and whether the corresponding measures have been taken in accordance with the instruction requirements. For maintenance optimization strategy instructions, such as electrolyte replenishment cycle recommendations and plate replacement priority recommendations, the system will monitor whether the relevant maintenance operations are carried out according to the time and requirements specified in the instructions. If there is no feedback information from the executing device, it can also be recorded through manual regular inspections to ensure that the execution status of the instructions can be understood in a timely manner so that further decisions and adjustments can be made based on the execution status.

[0061] Step 320: When it is detected that the abnormal state warning instruction has not been responded to for more than a preset time window, the forced load reduction operation of the target battery is activated and an emergency maintenance notification is generated: the output current is gradually reduced by the load controller of the target battery until it reaches the safety threshold range, and a load reduction process log is generated; the key event sequence and the corresponding timestamp information in the load reduction process log are extracted, the key event sequence is matched with a preset safety protocol library, and a protocol annotation report is generated; the protocol annotation report is associated with the emergency maintenance notification and sent to a preset maintenance terminal, and a visual chart of the load reduction process log is displayed on the maintenance terminal.

[0062] If the system detects that no response has been received within a preset time window (e.g., 30 minutes) after issuing an abnormality warning command, a series of actions are triggered. First, the load controller of the target battery is activated, causing it to gradually reduce the output current. The load controller follows a specific strategy, such as reducing the current by a fixed value or by a specific ratio, until the output current reaches a safe threshold. During this load reduction process, the system records a detailed log of the load reduction process, including information such as the moment of each current adjustment and the corresponding current value.

[0063] Next, key event sequences are extracted from the load shedding process log, such as key points of current reduction and changes in the magnitude of current adjustment. The corresponding time stamp information for these key events is also extracted. These key event sequences are then compared and matched against a pre-set safety protocol library. The safety protocol library stores various current adjustment patterns and event sequences that comply with safety standards. If any event sequences during the load shedding process are found to not conform to the standards in the safety protocol library, a protocol annotation report is generated, identifying the non-compliant portions and the corresponding safety protocol clauses.

[0064] The protocol annotation report is then linked to an emergency maintenance notice. This notice explains the unresponsiveness of the abnormal status warning and the forced load reduction action currently being taken, emphasizing the need for maintenance personnel to address the issue as quickly as possible. Both files are then sent to a pre-defined maintenance terminal, which can be a dedicated maintenance device or a maintenance personnel's work computer.

[0065] The maintenance terminal displays a visual chart of the load shedding process log. This chart intuitively illustrates how current changes over time during the load shedding process. For example, a current drop curve is plotted with time on the horizontal axis and current on the vertical axis. This allows maintenance personnel to monitor current trends in real time during the load shedding process, quickly understanding whether the load shedding operation meets safety requirements. Combined with protocol annotation reports and emergency maintenance notifications, they can accurately assess the situation and take appropriate maintenance measures.

[0066] In an optional embodiment, associating the protocol annotation report with the emergency maintenance notification and sending it to a preset maintenance terminal, and displaying a visual chart of the load shedding process log on the maintenance terminal, includes: Step 321: extract the event type identifier, current change rate exceeding amplitude parameter and corresponding second time stamp set of the abnormal adjustment event from the protocol annotation report, and extract the event triggering reason code and emergency processing level parameter from the emergency maintenance notification.

[0067] Carefully search and extract key information about abnormal regulation events in the protocol annotation report. The event type identifier identifies the specific type of anomaly, such as whether the current dropped too quickly or by too much. The current rate of change exceeding the standard amplitude parameter quantifies the degree of deviation from the normal current change. Additionally, the corresponding secondary timestamps for these abnormal events are recorded. These timestamps clearly indicate the exact time when the abnormal event occurred.

[0068] The emergency maintenance notification extracts the event trigger code, which explains why the forced load reduction operation was triggered, such as because the abnormal state warning instruction was not responded to. The emergency level parameter indicates the urgency of the current situation, such as high, medium, and low, so that maintenance personnel can understand the severity of the situation.

[0069] Step 322: Map and match the event type identifier with the event triggering cause code to generate an association mapping table containing event association index key-value pairs, wherein each key-value pair includes a corresponding relationship between a current change rate exceeding standard amplitude parameter and an emergency processing level parameter.

[0070] Optionally, the event type identifier extracted from the protocol annotation report and the event triggering reason code extracted from the emergency maintenance notification are subjected to correlation analysis. By establishing a mapping relationship, an association mapping table is generated. In the mapping table, each key-value pair not only contains the correspondence between the event type and the triggering reason, but also further associates the current change rate exceeding the standard amplitude parameter with the emergency processing level parameter. For example, when the event type identifier is "current drops too fast" and the event triggering reason code is "abnormal state warning is not responded", the corresponding key-value pair will record the current current change rate exceeding the standard amplitude, and whether the emergency processing level set according to this situation is high, medium or low. The above mapping table can help maintenance personnel quickly understand the full picture of the abnormal event and the corresponding degree of urgency.

[0071] Step 323: Based on the key-value pairs in the association mapping table, the second time stamp set and the current change rate exceeding amplitude parameter are data-encapsulated in chronological order to generate a protocol-notification association data packet with timing coding, in which each data unit contains a normalized time offset parameter and a standardized exceeding amplitude ratio value.

[0072] According to the key-value pair information in the associated mapping table, the second time stamp set and the current change rate exceeding the standard amplitude parameter are integrated. In chronological order, each time stamp and the corresponding exceeding the standard amplitude parameter are data encapsulated. During the encapsulation process, the time offset parameter is normalized and converted to a unified range, such as between 0 and 1, to facilitate subsequent processing and comparison. At the same time, the exceeding amplitude ratio value is standardized so that it is also within a suitable range, such as between -1 and 1. Through the above processing, a protocol-notification association data packet with timing coding is generated, and each data unit contains the processed time and exceeding amplitude information, which is convenient for data transmission and subsequent analysis.

[0073] Step 324: Send the protocol-notification associated data packet to the maintenance terminal through a preset encryption channel, perform a decryption verification operation on the data receiving interface of the maintenance terminal, and extract the decrypted time-series coded data unit set.

[0074] As you can understand, to ensure the security of data transmission, the generated protocol-notification-related data packets are sent to the maintenance terminal via a preset encrypted channel. The encrypted channel uses an encryption algorithm to encrypt the data packets to prevent the data from being stolen or tampered with during transmission. After receiving the data packet, the data receiving interface of the maintenance terminal performs decryption verification operations. First, the data packet is decrypted using the corresponding decryption key to restore the original data content. The decrypted data is then verified, for example, to check the data integrity and accuracy to ensure that no errors or corruption occurred during transmission. Once verification is successful, the decrypted set of time-series coded data units is extracted and used for subsequent visualization processing.

[0075] Step 325: calling the visualization engine of the maintenance terminal, converting the time offset parameter in the time-series coded data unit into a sequence of coordinate points on a unified time axis, and mapping the excess amplitude ratio value into a color level value under a preset color gradient.

[0076] On the maintenance terminal, the visualization engine is invoked to process the extracted time-coded data units. First, the time offset parameters in the data units are converted according to a unified time axis. The normalized time offsets are then converted into actual time coordinate points, forming a coordinate point sequence. This time axis can be used to map each time offset to a specific time point, starting from the start of the load shedding operation.

[0077] At the same time, the excess magnitude percentage values are mapped to a preset color gradient. For example, a value between -1 and -0.5 is set to red, indicating a severe excess; between -0.5 and 0 is set to orange, indicating a moderate excess; between 0 and 0.5 is set to yellow, indicating a mild excess; and between 0.5 and 1 is set to green, indicating a near-normal condition. This color mapping converts the excess magnitude percentage values into color scale values, allowing for intuitive visualization of the excess situation in the visualization chart.

[0078] Step 326: Generate a dynamic scatter distribution diagram based on the coordinate point sequence and the color scale value, and superimpose the current value curve of each adjustment stage in the load reduction process log on the scatter distribution diagram, wherein the current value curve is generated by linear interpolation based on the second time stamp and the corresponding adjustment stage current value.

[0079] Optionally, a dynamic scatter plot can be generated using the coordinate point sequence and color scale values. In this distribution plot, the horizontal axis represents the time coordinate point, and the vertical axis can be a general axis representing the exceedance or other related information. Each scatter point is plotted based on its corresponding time and exceedance magnitude ratio, and the color is determined by the color scale value. This allows for an intuitive visualization of the exceedance distribution at different time points.

[0080] To more comprehensively illustrate the load shedding process, the current value curve for each adjustment stage in the load shedding process log is overlaid on the scatter plot. Linear interpolation is used to generate the current value curve based on the second timestamp and the corresponding adjustment stage current value. For example, if the current value at time 1 is I1 and the current value at time 2 is I2, the corresponding current values at other time points between time 1 and time 2 are calculated using a linear interpolation formula, resulting in a smooth current value curve. This allows both the current trend and the distribution of exceeding the standard to be visualized in a single chart.

[0081] Step 327: Render the dynamic scatter distribution graph and the current value curve in parallel on the display interface of the maintenance terminal, and mark the emergency level parameter in the event association index key-value pair to generate a multi-layer visualization chart with time axis alignment.

[0082] On the display interface of the maintenance terminal, the dynamic scatter distribution graph and the current value curve are rendered in parallel, which means that the two graphics will be displayed simultaneously on the same interface, and the time axis is aligned, which is convenient for users to conduct comparative analysis. At the same time, the emergency level parameters in the event-related index key-value pair are marked on the chart. For example, high, medium, and low emergency levels are marked with different colors or symbols at the corresponding positions of the chart, so that users can quickly understand the degree of urgency at different time points and under different exceeding standards. In this way, a multi-layer visualization chart with time axis alignment is generated, providing maintenance personnel with a comprehensive and intuitive information display.

[0083] Step 328: Parse the text content of the protocol annotation report into a structured summary, which includes the statistical number of event types, the maximum exceedance amplitude parameter and the associated emergency processing suggestion items, and embed the structured summary into the sidebar of the visual chart for interactive display.

[0084] This step thoroughly analyzes and parses the text content of the protocol annotation report, extracting key information and generating a structured summary. It counts the number of occurrences of different event types, records the largest excess parameters, and summarizes emergency response recommendations for different situations. For example, a structured summary might display "Event Type: Current Drop Excessively Rapid, Occurred 3 times; Maximum Exceedance: 50%; Emergency Response Recommendation: Immediately Check and Adjust Load Controller Settings."

[0085] Then, the structured summary is embedded in the sidebar of the visual chart. In this way, when maintenance personnel view the visual chart, they can easily obtain key information of the protocol annotation report through the sidebar, and can obtain more detailed information through interactive operations, such as clicking on the items in the sidebar. This method combines visual charts and text information, providing maintenance personnel with a more convenient and comprehensive way to view information, helping them make decisions and take maintenance measures faster.

[0086] Step 330: When it is detected that the electrolyte replenishment cycle suggestion in the maintenance optimization strategy instruction is not applied, recalculate the replenishment frequency of the electrolyte replenishment cycle suggestion and update the maintenance optimization strategy instruction.

[0087] Optionally, the system continuously monitors the execution of maintenance optimization strategy instructions. If it finds that the electrolyte replenishment cycle recommendations are not being implemented as required, it will initiate a recalculation process. First, the various parameters of the current target battery are analyzed, such as the changes in electrolyte ion concentration reflected by the electrolyte activity eigenvector and the plate corrosion status reflected by the plate corrosion eigenvector. Combining this information with historical data, the electrolyte consumption rate and demand are reassessed. For example, if the electrolyte ion concentration is found to be decreasing at an accelerated rate recently, it means that the electrolyte is being consumed more quickly, and the replenishment cycle may need to be shortened.

[0088] Based on the new assessment results, the recommended electrolyte replenishment frequency is recalculated. This calculation can be performed using algorithms or models, such as those based on factors such as the current electrolyte ion concentration, target ion concentration, and historical replenishment data, to determine a more appropriate replenishment frequency. Once the calculation is complete, the electrolyte replenishment frequency recommendation section of the maintenance optimization strategy instructions is updated to ensure that maintenance personnel receive the latest and most appropriate maintenance recommendations to ensure proper battery operation.

[0089] As a non-limiting embodiment, after generating the online monitoring result of the target battery based on the comprehensive state evaluation vector, the method further includes: adjusting the acquisition period of the multi-source time series monitoring data set according to the triggering frequency of the abnormal state warning instruction and the execution feedback of the maintenance optimization strategy instruction; calculating the ratio of the cumulative triggering times of each abnormal source direction in the abnormal state warning instruction to the preset safety threshold to generate an abnormal level assessment parameter; inputting the abnormal level assessment parameter into a preset period adjustment function, and the period adjustment function outputs an updated data acquisition period according to the mapping relationship between the abnormal level assessment parameter and the preset level interval; generating a period adjustment instruction according to the updated data acquisition period, and sending the period adjustment instruction to the internal resistance sensor, the electrolyte sensor and the plate sensor to synchronously adjust the acquisition frequency of the first original data stream, the second original data stream and the third original data stream.

[0090] Optionally, the system will continuously record the triggering frequency of abnormal state warning instructions and the execution feedback information of maintenance optimization strategy instructions. For abnormal state warning instructions, the cumulative number of triggers for each abnormal source direction is counted. For example, over a period of time, it was found that the abnormal state warning instruction caused by electrolyte activity problems was triggered 5 times, and due to plate corrosion problems it was triggered 3 times. Compare the cumulative number of triggers for each abnormal source direction with the preset safety threshold and calculate the ratio. For example, the preset safety threshold is 10 times, and the abnormal level evaluation parameter for the electrolyte activity problem can be 5 / 10=0.5, and the abnormal level evaluation parameter for the plate corrosion problem can be 3 / 10=0.3.

[0091] These anomaly level assessment parameters are input into a preset cycle adjustment function. The cycle adjustment function pre-defines a mapping between the anomaly level assessment parameters and preset level intervals. For example, when the anomaly level assessment parameter is between 0 and 0.3, the anomaly is considered minor and the data collection cycle can be appropriately extended; when it is between 0.3 and 0.6, the current collection cycle is maintained; and when it is between 0.6 and 1, the anomaly is considered severe and the collection cycle needs to be shortened. Based on this mapping, the cycle adjustment function outputs the updated data collection cycle.

[0092] Based on the updated collection cycle, a cycle adjustment instruction is generated. This instruction specifies the new collection cycle requirement, for example, adjusting the internal resistance sensor's collection cycle from every 10 minutes to every 5 minutes. The cycle adjustment instruction is then sent to the internal resistance sensor, electrolyte sensor, and plate sensor. Upon receiving the instruction, the sensors synchronously adjust the collection frequency of the first, second, and third raw data streams according to the new collection cycle, ensuring timely acquisition of more accurate data for better monitoring of the target battery's status.

[0093] As a non-limiting embodiment, after generating the online monitoring result of the target battery based on the comprehensive state assessment vector, the method further includes: performing time alignment on the comprehensive state assessment vector and a set of historical comprehensive state assessment vectors to generate a state decay trajectory sequence; extracting the characteristic vector change rate of each time point in the state decay trajectory sequence, and calculating the change rate difference value between adjacent time points; comparing the change rate difference value with a preset decay rate threshold, and generating a capacity sudden drop warning signal when multiple consecutive change rate difference values exceed the decay rate threshold; activating the backup power supply switching operation of the target battery according to the capacity sudden drop warning signal, and recording the voltage fluctuation data during the switching process; associating the voltage fluctuation data with the capacity sudden drop warning signal and storing it in a fault log database, and updating the plate replacement priority recommendation in the maintenance optimization strategy instruction.

[0094] First, the currently generated comprehensive state assessment vector is time-series aligned with a set of historical comprehensive state assessment vectors. This set of historical comprehensive state assessment vectors records the comprehensive state information of the target battery at different points in the past. This time-series alignment arranges the vectors at different times in chronological order, forming a sequence of state decay trajectories. This sequence can intuitively demonstrate the changing trend of the battery's comprehensive state over time.

[0095] Next, we conduct an in-depth analysis of the state decay trajectory sequence. We extract the rate of change of the feature vector at each time point. For example, we calculate the difference between the comprehensive state evaluation vectors at two adjacent time points to quantify the speed of state change. We then calculate the difference in the rate of change between adjacent time points, comparing the difference in the rate of change of the feature vectors at two adjacent time points.

[0096] These rate-of-change differences are compared with a preset decay rate threshold. This threshold is a limit set based on battery performance standards and experience. If multiple consecutive rate-of-change differences exceed this threshold, it indicates that the battery's state is decaying at an abnormally rapid rate, potentially leading to a sudden capacity drop. A sudden capacity drop warning signal is generated.

[0097] Once a capacity drop warning signal is generated, the system automatically switches to the backup power source of the target battery. During the switchover process, specialized monitoring equipment is used to record voltage fluctuations, such as the voltage change at the moment of switching and the time it takes for the voltage to return to a stable state.

[0098] The recorded voltage fluctuation data is associated with the capacity sudden drop warning signal and stored in the fault log database. The fault log database is used to record various abnormal conditions and related data during the operation of the battery to facilitate subsequent analysis and troubleshooting. At the same time, based on the capacity sudden drop warning signal and related analysis results, the plate replacement priority recommendation in the maintenance optimization strategy instruction is updated. If the capacity sudden drop warning signal indicates that problems such as plate corrosion have a serious impact on the battery capacity, then the priority of plate replacement needs to be increased so that timely measures can be taken to prevent further damage to the battery and ensure the normal operation of the system. For example, the original plate replacement priority was at a lower level. After the capacity sudden drop warning signal appears, the priority is raised to a higher level and clearly marked in the maintenance optimization strategy instruction to prompt maintenance personnel to arrange plate replacement as soon as possible.

[0099] As a non-limiting embodiment, after generating the online monitoring result of the target battery based on the comprehensive state evaluation vector, the method further includes: generating a time node sequence of electrolyte replenishment operations based on the electrolyte replenishment cycle recommendation in the maintenance optimization strategy instruction; after each electrolyte replenishment operation is completed, collecting electrolyte activity time series data within a preset time window after replenishment to generate a replenishment effect verification data set; calling the feature extraction network to perform feature extraction on the replenishment effect verification data set to generate a post-replenishment electrolyte activity feature vector; calculating the cosine similarity between the post-replenishment electrolyte activity feature vector and the standard electrolyte activity feature vector, and determining a replenishment effect score based on the cosine similarity; when the replenishment effect score is lower than the set score, recalculating the replenishment dosage parameters in the electrolyte replenishment cycle recommendation, and generating a dosage adjustment instruction to update the maintenance optimization strategy instruction.

[0100] First, based on the electrolyte replenishment cycle recommendations given in the maintenance optimization strategy instructions, for example, it is recommended to perform electrolyte replenishment operations every T days, combined with the current time information, a time node sequence for the electrolyte replenishment operations is generated. This sequence clarifies the specific time when each electrolyte replenishment operation should be performed, providing accurate guidance for actual operations.

[0101] After each electrolyte replenishment operation, the data collection process is initiated. Within a preset time window after the replenishment operation is completed, for example, within 24 hours after replenishment, the electrolyte sensor periodically collects electrolyte activity time series data. This data reflects the change in electrolyte activity over time after the replenishment. This data is then compiled to generate a replenishment effect verification dataset.

[0102] Next, a pre-trained feature extraction network is used to extract features from the supplementation validation dataset. The feature extraction network analyzes and processes the information in the dataset, extracting key features that represent the post-supplement electrolyte activity state. Ultimately, it generates a post-supplement electrolyte activity feature vector, which contains various characteristic information about the post-supplement electrolyte activity, facilitating subsequent comparison and evaluation.

[0103] In order to evaluate the effect of electrolyte supplementation, the cosine similarity between the activity feature vector of the electrolyte after supplementation and the activity feature vector of the standard electrolyte is calculated. The activity feature vector of the standard electrolyte represents the activity characteristics of the electrolyte under ideal conditions. The calculation of cosine similarity can measure the degree of similarity between two vectors in direction, and the value range is between -1 and 1. The supplementation effect score is determined by combining the calculated cosine similarity value with certain scoring rules. For example, a cosine similarity value of 0.8 or above is set as excellent, corresponding to a higher supplementation effect score; between 0.6 and 0.8 is good, with a moderate score; and below 0.6 is poor, with a low score.

[0104] When the calculated supplementation effect score is lower than the set score, it means that the current electrolyte supplementation plan may not be effective and the supplementation dosage parameters need to be adjusted. The electrolyte activity characteristic vector after supplementation and related historical data are analyzed to recalculate the supplementation dosage parameters in the electrolyte supplementation cycle recommendation. For example, if it is found that the electrolyte activity is not significantly improved under the current supplementation dosage, it may be necessary to increase the supplementation dosage appropriately. Based on the recalculated results, a dosage adjustment instruction is generated. The instruction clarifies the new supplementation dosage requirements, which are used to update the maintenance optimization strategy instructions to ensure that maintenance personnel can follow the new dosage requirements in subsequent electrolyte supplementation operations to improve the effect of electrolyte supplementation and maintain good battery performance. Through such a series of operations, the battery electrolyte supplementation process can be effectively monitored and optimized to ensure the normal operation and stable performance of the battery.

[0105] In practical applications, adaptive improvements can be made based on existing multi-sensor data fusion methods, LSTM time series feature extraction models, and attention mechanisms. First, for data acquisition cycle parameters, the optimal interval can be determined through experimental verification, combining the sensor sampling frequency range recommended in battery industry standards (such as the requirements for energy storage battery monitoring in GB / T 34131-2017).

[0106] Secondly, for feature extraction networks, a multi-branch network can be expanded based on the classic LSTM architecture (such as the basic structure proposed by Hochreiter). Steady-state and transient modes can be captured separately through gating mechanisms. Transfer learning can be performed using public battery datasets (such as the NASA battery aging dataset) during pre-training. Cross-attention calculations can reference the multi-head attention mechanism in the Transformer model to achieve feature interaction, and weight parameters can be dynamically optimized through backpropagation. The state classifier can be constructed using the PyTorch framework with three fully connected layers, trained using a cross-entropy loss function and the Adam optimizer. The preset accuracy threshold is dynamically adjusted based on the F1-score of the confusion matrix.

[0107] Furthermore, the interpolation method used in data alignment can be implemented using the linear interpolation function of the Pandas library, and dimensionality issues can be addressed through Z-score normalization preprocessing. The construction of the security protocol library can define event sequence rules based on the IEC 62443 standard, and visual rendering can achieve multi-layer overlay using Matplotlib or ECharts libraries.

[0108] In this way, the whole process of multi-source data synchronous collection, time series feature extraction, interactive feature fusion and status assessment can be realized completely and clearly.

[0109] The embodiment of the present application first obtains a multi-source time series monitoring data set of the target battery, and then performs time series feature extraction processing on the multi-source time series monitoring data set by calling a pre-trained feature extraction network, which can accurately extract the internal resistance state feature vector, electrolyte activity feature vector and plate corrosion feature vector from complex data to characterize the state of the battery from different key angles; then, through state fusion analysis and processing, the internal resistance state feature vector, electrolyte activity feature vector and plate corrosion feature vector are organically integrated to generate a comprehensive state evaluation vector that comprehensively and systematically reflects the overall condition of the target battery; finally, based on the comprehensive state evaluation vector, online monitoring results are generated, which can not only issue abnormal state warning instructions in time to provide advance notification, but also give maintenance optimization strategy instructions to guide the reasonable planning of maintenance work, improve the service life and performance of the battery, and ensure its stable and reliable operation.

[0110] In summary, the embodiments of the present application obtain a multi-source time-series monitoring data set, use a pre-trained feature extraction network to accurately extract key feature vectors, and perform fusion analysis to ultimately generate comprehensive and practical online monitoring results, effectively overcoming the shortcomings of existing technologies in battery status monitoring.

[0111] See also Figure 2 As shown, this figure is a schematic diagram of the basic structure of an online monitoring system 200 provided in an embodiment of the present application. The online monitoring system 200 includes: Processor 201; a storage device 202 having a computer program 2020 stored thereon; When the computer program 2020 is executed by the processor 201, the processor 201 implements any of the online monitoring methods for batteries.

[0112] Based on the above, a readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0113] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems or devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

Claims

1. An online monitoring method for a battery, characterized in that: include: Obtain a multi-source time series monitoring data set of a target battery; Calling a pre-trained feature extraction network to perform time series feature extraction processing on the multi-source time series monitoring data set to generate an internal resistance state feature vector, an electrolyte activity feature vector, and a plate corrosion feature vector; Performing state fusion analysis on the internal resistance state feature vector, the electrolyte activity feature vector, and the plate corrosion feature vector to generate a comprehensive state assessment vector of the target battery; An online monitoring result of the target battery is generated based on the comprehensive state evaluation vector, wherein the online monitoring result includes an abnormal state warning instruction and a maintenance optimization strategy instruction.

2. The online monitoring method for batteries according to claim 1, characterized in that: The multi-source time series monitoring data set includes internal resistance change time series data, electrolyte activity time series data, and plate corrosion time series data. The multi-source time series monitoring data set of the target battery is obtained, including: Periodically collecting a first raw data stream from an internal resistance sensor of the target battery, and generating internal resistance variation time series data based on the first raw data stream; the first raw data stream includes a sequence of internal resistance measurement values of the target battery during different charge and discharge cycles; Periodically collecting a second raw data stream from the electrolyte sensor of the target battery, and generating electrolyte activity time series data based on the second raw data stream; the second raw data stream includes a sequence of changes in electrolyte ion concentration of the target battery at different ambient temperatures; Periodically collecting a third raw data stream from a plate sensor of the target battery, and generating plate corrosion time series data based on the third raw data stream; the third raw data stream includes a sequence of plate thickness changes of the target battery under multiple working load conditions; Data alignment processing is performed on the internal resistance change time series data, the electrolyte activity time series data and the plate corrosion time series data to synchronize the internal resistance measurement value sequence, the electrolyte ion concentration change sequence and the plate thickness change sequence in the timestamp dimension to generate the multi-source time series monitoring data set.

3. The online monitoring method for batteries according to claim 2, characterized in that: The calling of the pre-trained feature extraction network to perform time series feature extraction processing on the multi-source time series monitoring data set to generate an internal resistance state feature vector, an electrolyte activity feature vector, and a plate corrosion feature vector includes: Inputting the internal resistance change time series data into a first long short-term memory unit of the feature extraction network, capturing the steady-state dependency pattern and transient fluctuation pattern in the internal resistance measurement value sequence through the first long short-term memory unit, and generating the internal resistance state feature vector; Inputting the electrolyte activity time series data into the second long short-term memory unit of the feature extraction network, capturing the periodic decay pattern and the ambient temperature correlation pattern in the electrolyte ion concentration change sequence through the second long short-term memory unit to generate the electrolyte activity feature vector; The plate corrosion time series data is input into the third long short-term memory unit of the feature extraction network, and the load stress correlation pattern and the corrosion rate nonlinear change pattern in the plate thickness change sequence are captured by the third long short-term memory unit to generate the plate corrosion feature vector.

4. The online monitoring method for batteries according to claim 1, characterized in that: The performing state fusion analysis on the internal resistance state feature vector, the electrolyte activity feature vector, and the plate corrosion feature vector to generate a comprehensive state evaluation vector of the target battery includes: Performing a first cross-attention calculation on the internal resistance state feature vector and the electrolyte activity feature vector to obtain an internal resistance-electrolyte interaction feature vector; Perform a second cross-attention calculation on the internal resistance state feature vector and the plate corrosion feature vector to obtain an internal resistance-plate interaction feature vector; Performing a third cross-attention calculation on the electrolyte activity feature vector and the plate corrosion feature vector to obtain an electrolyte-plate interaction feature vector; The internal resistance-electrolyte interaction feature vector, the internal resistance-plate interaction feature vector and the electrolyte-plate interaction feature vector are weightedly fused to generate the comprehensive state evaluation vector.

5. The online monitoring method for batteries according to claim 4, characterized in that: The generating the online monitoring result of the target battery based on the comprehensive state assessment vector includes: Inputting the comprehensive state evaluation vector into a pre-trained state classifier, and determining the current state category of the target battery through the state classifier, wherein the current state category includes normal attenuation state, local abnormal state and global failure state; When the current state category is the local abnormal state, determining the abnormal source direction according to the contribution weights of each interactive feature vector in the comprehensive state evaluation vector, and generating the abnormal state warning instruction including the abnormal source direction; When the current state category is the global failure state, the maintenance optimization strategy instruction is generated according to the decay rate of each interactive feature vector in the comprehensive state evaluation vector, and the maintenance optimization strategy instruction includes electrolyte replenishment cycle suggestions and plate replacement priority suggestions.

6. The online monitoring method for batteries according to claim 5, characterized in that: Before inputting the integrated state evaluation vector into the pre-trained state classifier, the method further includes: Obtaining a sample comprehensive state evaluation vector set and a corresponding sample state label set of historical battery samples, wherein each sample state label in the sample state label set is used to indicate an actual state category of the corresponding historical battery sample; Dividing the sample comprehensive state evaluation vector set into a training set and a validation set, iteratively training the initial state classifier using the training set, and calculating the classification accuracy index using the validation set after each iteration; When the classification accuracy index reaches a preset accuracy rate, the training is stopped to obtain the pre-trained state classifier.

7. The online monitoring method for batteries according to claim 6, characterized in that: The configuration process of the initial state classifier includes: configuring a first fully connected layer to receive the comprehensive state evaluation vector and perform nonlinear transformation processing to generate a first hidden feature vector; configuring a second fully connected layer to perform dimension compression processing on the first hidden feature vector to generate a second hidden feature vector; configuring a third fully connected layer to perform probability mapping processing on the second hidden feature vector to generate a probability distribution vector of the current state category; Optimizing parameters of the first fully connected layer, the second fully connected layer, and the third fully connected layer according to the cross entropy loss between the probability distribution vector and the sample state label set; The initial state classifier is obtained according to the optimized first fully connected layer, the second fully connected layer and the third fully connected layer.

8. The online monitoring method for batteries according to claim 1, characterized in that: The method further comprises: monitoring in real time the execution status of the abnormal state warning instruction and the maintenance optimization strategy instruction in the online monitoring result of the target battery; When it is detected that the abnormal state warning instruction has not been responded to for more than a preset time window, the forced load reduction operation of the target battery is activated and an emergency maintenance notice is generated: the output current is gradually reduced by the load controller of the target battery until it reaches the safety threshold range, and a load reduction process log is generated; the key event sequence and corresponding timestamp information in the load reduction process log are extracted, the key event sequence is matched with a preset safety protocol library, and a protocol annotation report is generated; the protocol annotation report is associated with the emergency maintenance notice and sent to a preset maintenance terminal, and a visual chart of the load reduction process log is displayed on the maintenance terminal; When it is detected that the electrolyte replenishment cycle suggestion in the maintenance optimization strategy instruction is not applied, the replenishment frequency of the electrolyte replenishment cycle suggestion is recalculated and the maintenance optimization strategy instruction is updated.

9. The online monitoring method for batteries according to claim 8, characterized in that: The step of associating the protocol annotation report with the emergency maintenance notice and sending the report to a preset maintenance terminal, and displaying a visual chart of the load shedding process log on the maintenance terminal, includes: Extracting the event type identifier, the current change rate exceeding standard amplitude parameter and the corresponding second time stamp set of the abnormal adjustment event from the protocol annotation report, and extracting the event triggering cause code and the emergency processing level parameter from the emergency maintenance notification; Mapping and matching the event type identifier with the event triggering cause code to generate an association mapping table containing event association index key-value pairs, wherein each key-value pair includes a corresponding relationship between a current change rate exceeding standard amplitude parameter and an emergency processing level parameter; Based on the key-value pairs in the association mapping table, encapsulate the second time stamp set and the current change rate exceeding amplitude parameter in chronological order to generate a protocol-notification association data packet with time sequence encoding, wherein each data unit in the data packet includes a normalized time offset parameter and a normalized exceeding amplitude ratio value; Sending the protocol-notification associated data packet to the maintenance terminal through a preset encryption channel, performing a decryption verification operation on a data receiving interface of the maintenance terminal, and extracting a decrypted set of time-series coded data units; Invoking the visualization engine of the maintenance terminal to convert the time offset parameter in the time-series coded data unit into a sequence of coordinate points on a unified time axis, and mapping the excess amplitude ratio value into a color scale value under a preset color gradient; generating a dynamic scatter point distribution graph according to the coordinate point sequence and the color scale value, and superimposing a current value curve of each adjustment stage in the load reduction process log on the scatter point distribution graph, wherein the current value curve is generated by linear interpolation according to the second time stamp and the corresponding current value of the adjustment stage; Rendering the dynamic scatter distribution graph and the current value curve in parallel on the display interface of the maintenance terminal, and marking the emergency level parameter in the event association index key-value pair to generate a multi-layer visualization chart with time axis alignment; The text content of the protocol annotation report is parsed into a structured summary, which includes a statistical number of event types, a maximum exceeding range parameter, and associated emergency processing suggestion items, and the structured summary is embedded in a sidebar of the visual chart for interactive display.

10. An online monitoring system, characterized in that: include: processor; A storage device having a computer program stored thereon, wherein when the computer program is executed by the processor, the processor implements the online monitoring method for a battery as claimed in any one of claims 1 to 9.

Images