An unmanned ship battery safety management early warning method

Abstract

The present invention provides an unmanned vessel battery safety management and early warning method, which relates to the field of data processing technology. The method comprises: obtaining original battery parameter data during operation; performing parameter integration processing on the original battery parameter data; constructing a difference feature sequence based on stability analysis data; extracting trend offset data according to a set nonlinear feature combination rule; extracting feature change indicators within a continuous detection period and constructing an abnormal period candidate set; converting the abnormal period candidate set into an abnormal period vector, and performing a similarity comparison with a preset risk evolution template library; if the similarity exceeds a preset matching threshold, marking the abnormal period vector as a high-risk type; counting the number of abnormal periods of the high-risk type, and if the number reaches a preset abnormal threshold for triggering, generating early warning result data and outputting a safety risk prompt. The present invention improves the autonomy and accuracy of the unmanned vessel battery safety management and early warning.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an unmanned ship battery safety management and early warning method. Background Art

[0002] Currently, the most current technologies typically implement safe battery management for unmanned vessels by monitoring basic parameters such as battery voltage, current, and temperature. Some systems utilize embedded controllers to analyze collected data, identifying abnormal conditions such as low voltage or high temperature based on predefined thresholds and issuing warnings. To enhance monitoring effectiveness, some technologies also integrate simplified model prediction algorithms to provide a rough estimate of future battery conditions, enabling proactive intervention in potential risks.

[0003] During offshore survey operations, unmanned vessels must operate autonomously for extended periods, often in high-humidity, high-salinity environments. In such scenarios, traditional warning mechanisms based on fixed thresholds may not accurately identify abnormal changes caused by seawater erosion or lithium battery aging. For example, a battery may experience a nonlinear temperature rise after micro-infiltration of seawater, but because the magnitude of the change does not exceed the threshold, the system fails to trigger an early warning, leading to the continued accumulation of potential faults. This demonstrates that current technology is insufficiently capable of identifying progressive faults in complex environments, resulting in delayed early warnings. Summary of the Invention

[0004] The purpose of the present invention is to provide a battery safety management and early warning method for an unmanned ship, aiming to solve the problems mentioned in the background technology.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] A battery safety management and early warning method for an unmanned vessel, the method comprising:

[0007] Acquiring raw battery parameter data during operation, wherein the raw battery parameter data includes voltage data, current data, and temperature data;

[0008] Perform parameter integration processing on the original battery parameter data, classify and summarize the data according to the battery cell identification, and perform sliding average calculation on the parameters at different time nodes to obtain stability analysis data;

[0009] Constructing a difference feature sequence based on the stability analysis data, wherein the difference feature sequence is formed by calculating the temperature change gradient, voltage fluctuation frequency, and current imbalance degree in adjacent time periods;

[0010] Performing pattern recognition processing on the difference feature sequence and extracting trend offset data based on the set nonlinear feature combination rules. The trend offset data is used to identify potential progressive failure modes, which are distinguished from transient features of sudden abnormal conditions.

[0011] Based on the trend deviation data, the characteristic change indicators within the continuous detection period are extracted, and the candidate set of abnormal periods is constructed;

[0012] The abnormal cycle candidate set is converted into an abnormal cycle vector and compared with the preset risk evolution template library for similarity. If the similarity exceeds the preset matching threshold, the abnormal cycle vector is marked as a high-risk type;

[0013] The number of abnormal cycles of high-risk types is counted. If the number reaches the preset abnormal threshold, early warning result data is generated and a security risk prompt is output.

[0014] Preferably, the original battery parameter data is subjected to parameter integration processing, the data is classified and summarized according to the battery cell identification, and the parameters at different time nodes are subjected to sliding average calculation to obtain stability analysis data, including:

[0015] According to the battery cell identification, the voltage data, current data, temperature data and ambient humidity data in the original battery parameter data are grouped according to the battery cell to obtain a grouped original parameter set;

[0016] For each group of original parameter sets, the sliding average algorithm is applied to the voltage data, current data, temperature data and ambient humidity data in chronological order to obtain a smoothed parameter sequence;

[0017] Based on the smooth parameter sequence of each battery cell, the statistical characteristics of each time node are extracted to form preliminary stability data;

[0018] Based on the preliminary stability data, the parameter change amplitude and fluctuation trend of each battery cell during continuous operation are evaluated to generate stability analysis data for subsequent analysis.

[0019] Preferably, constructing a differential signature sequence based on the stability analysis data includes:

[0020] Based on the stability analysis data, the temperature data of each time node in the continuous detection cycle is extracted, the temperature change between adjacent time nodes is calculated, and the temperature change gradient is obtained by dividing the temperature change by the time interval;

[0021] Based on the stability analysis data, the voltage data of each time node in the continuous detection cycle is extracted, the voltage data is subjected to Fourier transform or zero-crossing statistics, the number of voltage changes per unit time is calculated, and the voltage fluctuation frequency is obtained;

[0022] Based on the stability analysis data, the current data at each time point within the continuous detection cycle is extracted. The current values ​​of each battery cell are averaged and the absolute deviation of each cell current from the average is calculated. All absolute deviations are summed and normalized to obtain the degree of current imbalance.

[0023] The calculation results of temperature change gradient, voltage fluctuation frequency and current imbalance degree are arranged in sequence according to the detection time to construct a difference feature sequence that reflects the dynamic changes of battery operation.

[0024] Preferably, pattern recognition processing is performed on the difference feature sequence, and trend deviation data is extracted according to a set nonlinear feature combination rule, including:

[0025] Based on the difference feature sequence, for each detection cycle, the temperature change gradient sequence, voltage fluctuation frequency sequence and current imbalance degree sequence are extracted as input features respectively;

[0026] Using the multi-dimensional feature fusion method, each input feature is combined and processed to obtain a fused feature sequence;

[0027] According to the pre-set nonlinear feature combination rules, the fused feature sequence is pattern recognized to identify the nonlinear trend offset in the feature sequence and calculate the trend offset intensity data;

[0028] Based on the trend shift intensity data, the time periods with abnormal changes within multiple consecutive detection cycles are screened out, and the trend shift data within each time period are extracted and summarized.

[0029] Preferably, the abnormal cycle candidate set is converted into an abnormal cycle vector and compared with a preset risk evolution template library for similarity. If the similarity exceeds a preset matching threshold, the abnormal cycle vector is marked as a high-risk type, including:

[0030] According to the abnormal period candidate set, characteristic statistics are calculated for each abnormal period to form an abnormal period feature set;

[0031] Merge the feature data in the abnormal period feature set into an abnormal period vector, each of which contains the mean value of the temperature gradient, the extreme value of the voltage fluctuation frequency, and the maximum value of the current imbalance degree;

[0032] Calculate the similarity between the abnormal cycle vector and the standard template vector in the risk evolution template library to obtain a similarity score;

[0033] If the similarity score of the abnormal cycle vector exceeds the matching threshold, the abnormal cycle vector is marked as a high-risk type and the high-risk abnormal cycle is recorded.

[0034] Preferably, based on the preliminary stability data, the parameter variation and fluctuation trend of each battery cell during continuous operation are evaluated to generate stability analysis data for subsequent analysis, including:

[0035] Based on the preliminary stability data, the voltage variation, temperature variation, and current variation of each battery cell in a continuous time window are calculated to obtain parameter variation data;

[0036] Based on the parameter variation data, the voltage, temperature and current of each battery cell are analyzed for fluctuations. Statistical methods are used to calculate the fluctuation trend data of each parameter within a continuous operation cycle.

[0037] The parameter change amplitude data and fluctuation trend data are integrated to generate comprehensive stability analysis data for each battery cell within the corresponding detection cycle.

[0038] Preferably, according to a pre-set nonlinear feature combination rule, pattern recognition is performed on the fused feature sequence to identify the nonlinear trend offset in the feature sequence and calculate the trend offset intensity data, including:

[0039] Based on the fusion feature sequence, a multivariable nonlinear function is used to jointly model the temperature gradient sequence, voltage fluctuation frequency sequence and current imbalance degree sequence to obtain nonlinear combined feature data.

[0040] Cluster analysis or support vector machine algorithm is applied to the nonlinear combination feature data to identify the representative trend shift feature intervals, and the trend shift intensity is numerically calculated to obtain the trend shift intensity data.

[0041] Preferably, based on the abnormal period candidate set, characteristic statistics are calculated for each abnormal period to form an abnormal period feature set, including:

[0042] According to the temperature change gradient sequence in each abnormal period, the temperature gradient mean and temperature gradient standard deviation of the period are calculated to obtain the temperature gradient statistics;

[0043] According to the voltage fluctuation frequency sequence in each abnormal period, the voltage fluctuation frequency extreme value and the voltage fluctuation frequency mean value of the period are calculated to obtain the voltage fluctuation frequency statistical value;

[0044] According to the current imbalance degree sequence in each abnormal cycle, the maximum value and the average value of the current imbalance degree in the cycle are calculated to obtain the current imbalance degree statistical value;

[0045] The temperature gradient statistics, voltage fluctuation frequency statistics, and current imbalance degree statistics are aggregated to form an abnormal period feature set.

[0046] Preferably, each feature data in the abnormal period feature set is merged into an abnormal period vector, each abnormal period vector including the mean value of the temperature gradient, the extreme value of the voltage fluctuation frequency, and the maximum value of the current imbalance degree, including:

[0047] Based on each abnormal period feature set, perform the following operations:

[0048] Extract the mean value of temperature gradient as the first component of the abnormal period vector;

[0049] Extract the extreme value of voltage fluctuation frequency as the second component of abnormal period vector;

[0050] Extracting the maximum value of the current imbalance degree as the third component of the abnormal period vector;

[0051] The first component, the first component, and the third component are combined and normalized to form an abnormal periodic vector.

[0052] Preferably, similarity calculation is performed between the abnormal period vector and the standard template vector in the risk evolution template library to obtain a similarity score, including:

[0053] Match the abnormal cycle vector to be compared with each standard template vector in the risk evolution template library one by one;

[0054] The similarity algorithm is used to calculate the similarity scores between the abnormal period vector and each standard template vector;

[0055] According to the similarity scores of all standard templates, the template closest to the abnormal period vector and its highest similarity score are screened out as the final similarity score result.

[0056] The above solution of the present invention includes at least the following beneficial effects:

[0057] By continuously collecting and integrating voltage, current, and temperature data during the operation of the unmanned vessel's batteries and classifying and summarizing these parameters based on the battery cell identifiers, we can achieve refined management of the health status of each battery cell. Dynamically smoothing the parameter data through algorithms such as sliding averages yields more realistic and noise-resistant stability analysis data, effectively avoiding false positives and missed positives caused by single-point anomalies or short-term disturbances.

[0058] This invention further constructs a differential feature sequence from stability analysis data, integrating indicators such as temperature gradient, voltage fluctuation frequency, and current imbalance to form a high-dimensional signature reflecting the dynamic changes in the battery system. This can accurately capture subtle abnormal trends before progressive failures occur. By combining nonlinear features with pattern recognition methods, unlike traditional solutions that rely on fixed thresholds, this method can more keenly identify potential risks of nonlinear evolution. It is particularly suitable for intelligent monitoring of hidden risks such as external corrosion and cell degradation in unmanned vessels operating in complex environments such as offshore, humid, and high-salt environments.

[0059] Furthermore, the present invention proposes a high-risk cycle identification mechanism based on comparing a candidate set of abnormal cycles with a risk evolution template. This mechanism can intelligently aggregate and accurately identify risk cycles, significantly improving the ability to provide early warning of progressive battery system failures. Compared to existing technologies that only address sudden anomalies or drastic changes, this invention enables early diagnosis of failure risks in complex scenarios such as slow battery degradation and long-term environmental impacts, significantly enhancing the safety and system reliability of unmanned vessel missions. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flowchart of an unmanned vessel battery safety management and early warning method provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0061] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0062] like Figure 1 As shown, an embodiment of the present invention provides an unmanned ship battery safety management and early warning method, the method comprising:

[0063] S100, acquiring raw battery parameter data during operation, wherein the raw battery parameter data includes voltage data, current data, temperature data, and ambient humidity data;

[0064] S200, performing parameter integration processing on the original battery parameter data, classifying and summarizing the data according to the battery cell identification, and performing sliding average calculation on the parameters at different time nodes to obtain stability analysis data;

[0065] S300, constructing a difference feature sequence based on the stability analysis data, wherein the difference feature sequence is formed by calculating the temperature change gradient, voltage fluctuation frequency, and current imbalance degree in adjacent time periods;

[0066] S400, performing pattern recognition processing on the difference feature sequence, and extracting trend offset data according to a set nonlinear feature combination rule, wherein the trend offset data is used to identify potential progressive failure modes, as distinguished from transient features of sudden abnormal conditions;

[0067] S500, extracting characteristic change indicators within a continuous detection period based on trend deviation data, and constructing an abnormal period candidate set;

[0068] S600: Convert the abnormal cycle candidate set into an abnormal cycle vector and compare the similarity with the preset risk evolution template library. If the similarity exceeds the preset matching threshold, mark the abnormal cycle vector as a high-risk type.

[0069] S700: Count the number of abnormal cycles of high-risk types. If the number reaches a preset abnormal threshold, generate early warning result data and output a security risk prompt.

[0070] In an embodiment of the present invention, a battery safety management and early warning method for unmanned vessels is provided. By continuously collecting raw battery parameter data, including voltage, current, temperature, and ambient humidity data, during the operation of the unmanned vessel, this method enables comprehensive, real-time information on the battery's operating status. This real-time collection of these parameters dynamically reflects the battery's operating status under different operating and environmental conditions, ensuring the comprehensiveness and timeliness of subsequent analysis.

[0071] The collected raw battery parameter data undergoes further parameter integration processing. First, the collected data is grouped and categorized according to each battery cell's unique identifier, ensuring that subsequent data analysis can extract targeted features based on the individual battery cells. Then, for each grouped data, a sliding average algorithm is applied to the voltage, current, temperature, and ambient humidity data in chronological order to produce a continuous, smooth parameter sequence. This effectively filters out data noise and abnormal jumps caused by short-term disturbances. This processing method enhances data reliability and the accuracy of subsequent analysis.

[0072] Based on the smoothed data, a stability analysis is performed. This data reflects the operating trends and fluctuations of each battery cell during the continuous operation of the unmanned vessel. This method can promptly detect unusual fluctuations in battery cells and facilitate proactive countermeasures.

[0073] Furthermore, based on the obtained stability analysis data, a difference feature sequence is constructed. The construction of the difference feature sequence includes three core indicators: temperature change gradient, voltage fluctuation frequency and current imbalance. By calculating the temperature change between adjacent time periods and combining the time interval, the rate of change of the battery temperature can be accurately reflected. By performing frequency statistics on voltage data through methods such as Fourier transform or zero crossing, the frequency of battery terminal voltage fluctuations can be reflected, which helps to reveal the occurrence of load shocks or power supply anomalies. By counting the absolute deviation of each single cell current from the mean and normalizing it, the uniformity of current distribution between different battery cells can be quantified, and the aging or abnormality of certain cells can be sensitively captured. Through the dynamic sequence arrangement of these features, the system can more deeply reflect the working dynamics of the battery pack and lay a data foundation for subsequent fault trend identification.

[0074] The above method not only achieves continuous monitoring of the working status of the unmanned ship's battery, but also combines the dynamic changes of multiple characteristics to achieve timely warning of potential battery risks. Compared with traditional monitoring methods that rely only on a single parameter or static threshold, this method can effectively overcome misjudgments and missed judgments caused by factors such as environmental changes or battery aging, and improve the scientific nature and foresight of battery safety management. For example, during a long-distance autonomous operation at sea, this method can detect in real time the abnormal temperature change rate of a battery cell and issue an early alarm, avoiding the serious consequences caused by further failure.

[0075] In a preferred embodiment of the present invention, the original battery parameter data is subjected to parameter integration processing, the data is classified and summarized according to the battery cell identification, and the parameters at different time nodes are subjected to sliding average calculation to obtain stability analysis data, including:

[0076] According to the battery cell identification, the voltage data, current data, temperature data and ambient humidity data in the original battery parameter data are grouped according to the battery cell to obtain a grouped original parameter set;

[0077] For each group of original parameter sets, the sliding average algorithm is applied to the voltage data, current data, temperature data and ambient humidity data in chronological order to obtain a smoothed parameter sequence;

[0078] Based on the smooth parameter sequence of each battery cell, the statistical characteristics of each time node are extracted to form preliminary stability data;

[0079] Based on the preliminary stability data, the parameter change amplitude and fluctuation trend of each battery cell during continuous operation are evaluated to generate stability analysis data for subsequent analysis.

[0080] In this embodiment of the present invention, the raw battery parameter data acquired during the operation of the unmanned vessel is first grouped and categorized based on the unique identifiers of the battery cells. This creates an independent raw parameter set for each battery cell. This provides accurate data support for the subsequent operation status of each battery cell.

[0081] For each grouped set of raw parameters, a sliding average algorithm is applied to the voltage, current, temperature, and ambient humidity data in chronological order. The sliding average window length can be flexibly adjusted based on the actual operating conditions of the UAV mission, sampling frequency, and environmental fluctuations. This effectively reduces the impact of occasional disturbances and measurement noise, resulting in a smoother and more realistic data curve reflecting the actual battery status. Applying the sliding average yields a smoothed parameter sequence at each time point, providing a solid foundation for further data analysis.

[0082] Next, for each battery cell's smoothed parameter sequence, statistical features such as mean, extreme values, standard deviation, and coefficient of variation are extracted to form preliminary stability data. For example, by analyzing the mean and standard deviation of temperature data, it is possible to assess whether the current battery thermal state is stable. Furthermore, by comparing these statistical features within continuous time windows, the magnitude of changes in voltage, temperature, and current of the battery cells during long-term operation, as well as the fluctuation trends of each parameter over time, are evaluated. These analysis results can comprehensively reflect the health level of the battery and the stability of its operating environment.

[0083] Ultimately, all parameter variation and fluctuation trend data are aggregated to generate comprehensive stability analysis data for each battery cell within a specified testing cycle, providing a high-quality data foundation for subsequent fault prediction and feature extraction. In practical applications, if the stability analysis data of a single battery cell exhibits high volatility or frequent anomalies over a long period of time, it can be identified as a potential risk point, triggering early maintenance or replacement recommendations, thereby improving the overall safety and reliability of the unmanned vessel's operation.

[0084] For each group of original parameter sets, the sliding average algorithm is applied to the voltage data, current data, temperature data, and ambient humidity data in chronological order to obtain a smoothed parameter sequence, specifically including:

[0085] For the original parameter set grouped by battery cell identifier, the parameter data points are first arranged in chronological order of sampling time. For voltage data, for example, several consecutive sampling time points are selected (e.g., a 5-minute window). The voltage data within each window is averaged, and the smoothed voltage value at the final time point in the window is calculated. This method transforms each voltage data series into a new series consisting of multiple sliding averages by shifting the window. A similar approach is used for current, temperature, and ambient humidity data, ensuring that each parameter series is synchronously smoothed across the time dimension.

[0086] The sliding average process described above effectively reduces spikes and outliers caused by short-term interference, environmental fluctuations, or occasional equipment failures, improving the stability of the parameter sequence. For example, when operating an unmanned vessel in strong winds and waves and with sudden changes in environmental parameters, the temperature data collected in a single shot may occasionally rise. However, after processing with the sliding average algorithm, the resulting temperature curve will more accurately reflect the actual operating trend, making subsequent analysis more reliable. The sliding window size can be flexibly set based on the actual application. Common settings include 3-point, 5-point, or 10-point windows, which can be optimized based on the data sampling frequency and system response speed.

[0087] Among them, according to the smooth parameter sequence of each battery cell, the statistical characteristics of each time node are extracted to form preliminary stability data, including:

[0088] For each battery cell, based on the obtained smoothing parameter sequence, key statistical features within that node and adjacent windows are calculated sequentially at each time point. Common features include the current window's mean, extreme values ​​(maximum and minimum), amplitude of change (the difference between the maximum and minimum values), and standard deviation (reflecting the intensity of fluctuation). For example, for a temperature smoothing sequence at a specific point in time, the mean, maximum, and minimum values ​​within the past hour can be calculated, and the temperature fluctuation range and average rate of change can be further calculated.

[0089] The same feature extraction method is used for voltage, current, and ambient humidity data, ensuring that each time node forms a multidimensional statistical feature set reflecting the local trend at that node. These statistical features can intuitively reflect the health and stability of the battery cells during that time period, providing a data foundation for subsequent fault trend analysis. For example, if the standard deviation of a battery cell temperature continues to increase, it may indicate that the battery is experiencing abnormal operating conditions and should attract system attention.

[0090] In practical applications, the selection of the above statistical features can be expanded according to project requirements, such as adding high-order statistical indicators such as skewness and kurtosis to enhance the ability to perceive abnormal trends under complex working conditions.

[0091] In a preferred embodiment of the present invention, constructing a differential signature sequence based on stability analysis data includes:

[0092] Based on the stability analysis data, the temperature data of each time node in the continuous detection cycle is extracted, the temperature change between adjacent time nodes is calculated, and the temperature change gradient is obtained by dividing the temperature change by the time interval;

[0093] Based on the stability analysis data, the voltage data of each time node in the continuous detection cycle is extracted, the voltage data is subjected to Fourier transform or zero-crossing statistics, the number of voltage changes per unit time is calculated, and the voltage fluctuation frequency is obtained;

[0094] Based on the stability analysis data, the current data at each time point within the continuous detection cycle is extracted. The current values ​​of each battery cell are averaged and the absolute deviation of each cell current from the average is calculated. All absolute deviations are summed and normalized to obtain the degree of current imbalance.

[0095] The calculation results of temperature change gradient, voltage fluctuation frequency and current imbalance degree are arranged in sequence according to the detection time to construct a difference feature sequence that reflects the dynamic changes of battery operation.

[0096] In this embodiment of the present invention, based on the stability analysis data generated above, the system further extracts dynamic features of the unmanned vessel's battery pack's operating status. Specifically, within each detection cycle, the temperature data of all battery cells is sequentially extracted. The temperature change between adjacent time nodes is calculated and then divided by the time interval to obtain the temperature change gradient. The temperature change gradient reflects not only the absolute value of the temperature change, but also its rate of change, which is directly helpful in identifying thermal runaway risks or local hot spots.

[0097] The voltage data is then analyzed using a Fourier transform to analyze the signal's frequency domain characteristics. Alternatively, zero-crossing statistics are used to quantify the number of voltage changes per unit time, thereby determining the voltage fluctuation frequency. This voltage fluctuation frequency can reveal the battery's adaptability to high-frequency loads and can also be used to identify the impact of unstable power supply or load shocks.

[0098] For current data, the current value of each battery cell during each test cycle is counted, the overall mean is calculated, and the absolute deviation of each cell from the mean is further analyzed. All absolute deviations are normalized and aggregated to determine the degree of current imbalance. This feature can effectively identify potential risks of cell performance degradation or mismatch within the battery pack, preventing cell failures from escalating into systemic failures.

[0099] Finally, the temperature gradient, voltage fluctuation frequency, and current imbalance are arranged in chronological order according to the detection time to construct a differential feature sequence. This feature sequence comprehensively reflects the healthy changes in the battery system of the unmanned vessel at each stage of operation, providing solid data support for subsequent pattern recognition and anomaly detection. For example, in a certain actual mission, the system successfully warned of the impending thermal runaway of a single battery by detecting an abnormal increase in the temperature gradient over multiple consecutive cycles, arranging the return trip in advance to avoid an accident, demonstrating the significant advantages of this embodiment in practical applications.

[0100] In a preferred embodiment of the present invention, pattern recognition processing is performed on the difference feature sequence, and trend deviation data is extracted according to a set nonlinear feature combination rule, including:

[0101] Based on the difference feature sequence, for each detection cycle, the temperature change gradient sequence, voltage fluctuation frequency sequence and current imbalance degree sequence are extracted as input features respectively;

[0102] Using the multi-dimensional feature fusion method, each input feature is combined and processed to obtain a fused feature sequence;

[0103] According to the pre-set nonlinear feature combination rules, the fused feature sequence is pattern recognized to identify the nonlinear trend offset in the feature sequence and calculate the trend offset intensity data;

[0104] Based on the trend shift intensity data, the time periods with abnormal changes within multiple consecutive detection cycles are screened out, and the trend shift data within each time period are extracted and summarized.

[0105] In this embodiment of the present invention, based on the constructed differential feature sequence, further pattern recognition processing is performed on the battery operating status. First, for each detection cycle, the system extracts a temperature gradient sequence, a voltage fluctuation frequency sequence, and a current imbalance degree sequence. These three sequences are used as input features to fully describe the dynamic characteristics of the battery during that cycle. To improve the accuracy of pattern recognition, a multidimensional feature fusion method is used to combine these input features to produce a fused feature sequence. This fused sequence achieves the coordinated utilization of multi-source information based on a single feature, enhancing the discernibility of fault trends.

[0106] Next, the system conducts an in-depth analysis of the fused feature sequence based on pre-set nonlinear feature combination rules, and uses methods such as nonlinear transformation and multivariable function combination to model and synthesize the complex relationships between different features. This nonlinear pattern recognition method can effectively distinguish the nonlinear differences between normal operating conditions and potential fault conditions. For the detected abnormal patterns, trend offset intensity data is further calculated to quantify the significance of feature changes. Finally, based on the trend offset intensity data, the system screens out time periods with abnormal changes within multiple consecutive detection cycles, extracts and summarizes the trend offset data for each time period, and ensures high sensitivity and low false alarm rate for subsequent fault warnings. For example, during a long-term navigation mission, through this fusion and identification process, the system can accurately capture early fault trends caused by multi-dimensional parameter coupling, realizing intelligent warning and scientific operation and maintenance of unmanned ship batteries.

[0107] Among them, based on the trend deviation intensity data, the time period with abnormal changes in multiple consecutive detection cycles is screened out, and the trend deviation data in each time period is extracted and summarized, including:

[0108] After performing pattern recognition on the fused feature sequence, the system obtains trend shift intensity data for each detection cycle. Trend shift intensity can be understood as reflecting the degree of abnormal evolution or rapid change in the feature sequence. To identify abnormal changes, the system pre-defines a set of criteria or reference intervals. For example, it compares the trend shift intensity to the normal range of variation under different operating conditions, as determined by historical data.

[0109] When the trend deviation intensity within a certain detection period consistently exceeds the normal reference value, and this phenomenon recurs over several consecutive detection periods, the system automatically identifies these consecutive periods as periods of abnormal change. To ensure the effectiveness of the screening results, the system can make a comprehensive assessment based on conditions such as the minimum number of continuous periods and the intensity threshold. For example, if the system sets the deviation intensity above the threshold for three consecutive periods, it will be marked as an abnormal period.

[0110] After filtering out abnormal time periods, the system further extracts and aggregates trend deviation data from those time periods, facilitating subsequent generation of candidate sets of abnormal periods or risk template comparison. In practical applications, for example, when an offshore unmanned vessel encounters continuous extreme weather, the trend deviation intensity of various battery characteristic parameters increases continuously within a short period of time. This method allows the system to quickly locate risky time periods and accurately monitor and respond to potential faults in the early stages.

[0111] In a preferred embodiment of the present invention, the abnormal cycle candidate set is converted into an abnormal cycle vector and compared with a preset risk evolution template library for similarity. If the similarity exceeds a preset matching threshold, the abnormal cycle vector is marked as a high-risk type, including:

[0112] According to the abnormal period candidate set, characteristic statistics are calculated for each abnormal period to form an abnormal period feature set;

[0113] Merge the feature data in the abnormal period feature set into an abnormal period vector, each of which contains the mean value of the temperature gradient, the extreme value of the voltage fluctuation frequency, and the maximum value of the current imbalance degree;

[0114] Calculate the similarity between the abnormal cycle vector and the standard template vector in the risk evolution template library to obtain a similarity score;

[0115] If the similarity score of the abnormal cycle vector exceeds the matching threshold, the abnormal cycle vector is marked as a high-risk type and the high-risk abnormal cycle is recorded.

[0116] In an embodiment of the present invention, the system is able to determine the abnormal changes in characteristics within the continuous detection cycle through the aforementioned trend offset data, and construct a candidate set of abnormal cycles based on this. For each abnormal cycle, the system first calculates the mean and standard deviation of the temperature change gradient series, the extreme value and mean of the voltage fluctuation frequency series, and the maximum value and mean of the current imbalance degree series to form a rich set of characteristic statistical values. Subsequently, the statistical data in the abnormal cycle feature set are integrated and merged into abnormal cycle vectors. Each abnormal cycle vector contains core components such as the temperature gradient mean, the voltage fluctuation frequency extreme value, and the current imbalance degree maximum value. Through this structured representation, the multi-dimensional health status of each abnormal cycle can be quantified, normalized, and compared.

[0117] The system further matches the abnormal cycle vector with the standard template vectors in the preset risk evolution template library one by one, and calculates the similarity score between the abnormal cycle vector and each template through similarity algorithms such as cosine similarity and Euclidean distance. By comparing all standard templates, the template closest to the current abnormal cycle and its highest similarity score are screened out and used as the final similarity score result. If the similarity score exceeds the risk threshold preset by the system, the abnormal cycle vector is marked as a high-risk type, and the high-risk abnormal cycle and related parameters are recorded and archived. This method can greatly improve the accuracy of anomaly detection, effectively reduce misjudgments and missed judgments, and thus significantly improve the level of intelligent safety management of unmanned ship battery packs. For example, in actual operation, the system can discover cycle characteristics that are highly similar to certain typical failure cases by comparing historical templates, thereby achieving early intervention and intelligent alarms.

[0118] According to the abnormal period candidate set, characteristic statistics are calculated for each abnormal period to form an abnormal period feature set, which specifically includes:

[0119] For the set of candidate abnormal cycles identified through trend shift analysis, the system first performs a statistical analysis of the data contained in each abnormal cycle. Taking the temperature gradient as an example, the system calculates the temperature gradient values ​​at all time points within the abnormal cycle, and then calculates the average, maximum, minimum, and standard deviation of the temperature gradient for that cycle. These statistical values ​​accurately reflect the overall trend and fluctuation of battery temperature evolution during the abnormal cycle.

[0120] Similarly, statistical analysis is performed on parameters such as voltage fluctuation frequency and current imbalance. For example, within a specific abnormal period, the extreme values ​​and mean values ​​of voltage fluctuation frequency, as well as the maximum and mean values ​​of current imbalance, are counted at all time points. This allows for a multi-dimensional characterization of the core characteristics of each abnormal period.

[0121] Ultimately, the system organizes and categorizes the statistical values ​​calculated for each abnormal cycle to form a structured abnormal cycle feature set. For example, a feature set for an abnormal cycle might include the mean and standard deviation of the temperature gradient, the extreme values ​​of the voltage fluctuation frequency, and the maximum value of the current imbalance. This feature set not only facilitates subsequent vectorization of abnormal cycles and comparison with a template library, but also provides experts with detailed data to analyze battery operation anomalies. In actual engineering, if the statistical values ​​of various characteristic parameters within a specific abnormal cycle are significantly higher than historical normal values, it indicates a potential risk in that cycle and requires focused monitoring or early warning.

[0122] In a preferred embodiment of the present invention, based on the preliminary stability data, the parameter variation and fluctuation trend of each battery cell during continuous operation are evaluated to generate stability analysis data for subsequent analysis, including:

[0123] Based on the preliminary stability data, the voltage variation, temperature variation, and current variation of each battery cell in a continuous time window are calculated to obtain parameter variation data;

[0124] Based on the parameter variation data, the voltage, temperature and current of each battery cell are analyzed for fluctuations. Statistical methods are used to calculate the fluctuation trend data of each parameter within a continuous operation cycle.

[0125] The parameter change amplitude data and fluctuation trend data are integrated to generate comprehensive stability analysis data for each battery cell within the corresponding detection cycle.

[0126] In this embodiment of the present invention, after obtaining preliminary stability data for each battery cell, the system further evaluates the magnitude of parameter changes and fluctuation trends during continuous operation. First, for each battery cell, the magnitude of voltage, temperature, and current changes within a continuous time window is calculated. Specifically, detailed parameter variation data can be obtained using methods such as the range and average difference between adjacent time nodes. Subsequently, based on this variation data, a fluctuation analysis is performed on the voltage, temperature, and current of each battery cell. Statistical methods (such as standard deviation and variance) are used to analyze the fluctuation trends of each parameter within the continuous operation cycle and quantify the degree of health fluctuation of the battery cell.

[0127] By comprehensively integrating the above parameter change amplitude data and fluctuation trend data, the system generates comprehensive stability analysis data for each battery cell within the corresponding detection cycle. This data can comprehensively reflect the dynamic health level of the battery cell at different operating stages, helping to detect short-term anomalies and long-term trend changes. For example, if the voltage of a single battery cell fluctuates abnormally over a continuous cycle, combined with a sharp increase in temperature and increased current fluctuations, the system can output an alarm indicating decreased stability, providing strong support for maintenance and management decisions. By refining the health analysis process, this solution significantly improves the reliability and scientific nature of unmanned ship battery management.

[0128] Among them, based on the parameter variation data, the voltage, temperature and current of each battery cell are subjected to fluctuation analysis. The fluctuation trend data of each parameter in the continuous operation cycle is calculated using statistical methods, including:

[0129] First, the system calculates the maximum and minimum voltage, temperature, and current values ​​of each battery cell within a set time period based on a sliding window. The difference between the two represents the magnitude of the parameter change during that period. This magnitude reflects the degree of short-term and drastic changes in the battery's operating state within each sampling window.

[0130] To further analyze parameter fluctuation trends, the system uses statistical methods such as variance, standard deviation, and root mean square (RMS) to quantify the fluctuations of the aforementioned amplitude data for each battery cell over multiple consecutive operating cycles. Specifically, the standard deviation of the voltage, temperature, and current fluctuation amplitudes over several consecutive cycles can be calculated to obtain the fluctuation trends of these parameters over time.

[0131] For example, if the temperature variation of a battery cell over five consecutive sampling cycles is A, B, C, D, and E, the standard deviation of this series can be calculated to quantify the temperature fluctuation trend. If the standard deviation continues to increase, it indicates that the battery temperature fluctuations are becoming more severe, indicating that the operating stability of the battery cell is declining. This type of fluctuation trend analysis can detect abnormal battery status in advance and guide operation and maintenance in actual projects.

[0132] The parameter variation data and fluctuation trend data are integrated to generate comprehensive stability analysis data for each battery cell within the corresponding test cycle, including:

[0133] After obtaining the variation and fluctuation trend data for each battery cell's parameters, the system organically integrates this data to form comprehensive stability analysis data reflecting the battery's overall health status. Specifically, the variation of voltage, temperature, and current within a certain detection cycle is weighted or normalized, combining the variation and fluctuation trend values ​​of each parameter to generate a unified data vector that characterizes the operational stability and risk level of the battery cell within that cycle.

[0134] For example, the system can assign different weights to the magnitude and fluctuation trends of voltage, temperature, and current based on actual application needs, such as appropriately increasing the weight of temperature indicators under high-temperature conditions. The resulting comprehensive stability analysis data can be used for both automated machine detection of anomalies and for manual interpretation and comparison of the health status of different battery cells.

[0135] This integrated approach effectively identifies battery cells that exhibit both significant temperature variations and strong fluctuations, enabling early detection of high-risk targets and providing robust data support for safe operation of unmanned vessels. For example, if both the temperature variation and fluctuation trend of a battery cell are abnormally high, the system will identify its comprehensive stability analysis data as high risk and recommend prompt repair or replacement.

[0136] In a preferred embodiment of the present invention, pattern recognition is performed on the fused feature sequence according to a pre-set nonlinear feature combination rule, the nonlinear trend offset in the feature sequence is identified, and the trend offset intensity data is calculated, including:

[0137] Based on the fusion feature sequence, a multivariable nonlinear function is used to jointly model the temperature gradient sequence, voltage fluctuation frequency sequence and current imbalance degree sequence to obtain nonlinear combined feature data.

[0138] Cluster analysis or support vector machine algorithm is applied to the nonlinear combination feature data to identify the representative trend shift feature intervals, and the trend shift intensity is numerically calculated to obtain the trend shift intensity data.

[0139] In this embodiment of the present invention, the system uses multivariate nonlinear functions to jointly model the temperature gradient sequence, voltage fluctuation frequency sequence, and current imbalance degree sequence for the fused feature sequence, generating nonlinear combined feature data. For example, methods such as multivariate polynomial functions and neural network mapping can be used to nonlinearly couple the changing relationships between different physical quantities, thereby fully exploring the potential connections between multiple features.

[0140] Subsequently, the system applies algorithms such as cluster analysis or support vector machines to the nonlinear combination feature data to identify the trend offset feature intervals in the data. For example, by automatically dividing different patterns through clustering algorithms, or by using support vector machines to distinguish the boundaries of normal and abnormal intervals, representative trend offsets can be efficiently extracted. For each trend offset interval, numerical calculations are further performed to obtain trend offset intensity data. The system can dynamically classify and track the trend offset intensity within all detection cycles, and issue risk warnings for periods of continuous high-intensity offsets. In practical applications, this method can effectively identify early signs of battery failure caused by multi-parameter coupling and nonlinear evolution, greatly improving the sensitivity and accuracy of fault warnings, and providing solid technical support for the intelligent and safe operation and maintenance of unmanned ship battery systems.

[0141] Among them, based on the fusion feature sequence, a multivariable nonlinear function is used to jointly model the temperature change gradient sequence, voltage fluctuation frequency sequence, and current imbalance degree sequence to obtain nonlinear combined feature data, including:

[0142] Based on the temperature gradient sequence, voltage fluctuation frequency sequence, and current imbalance degree sequence collected and processed by the system, these three features are first synchronously arranged according to the same detection cycle to form a multidimensional feature vector for each cycle. To capture the complex and nonlinear relationships between multiple physical features, the system selects multivariate nonlinear modeling methods such as polynomial regression, radial basis function neural network, decision tree regression, or ensemble learning models to jointly model these feature vectors.

[0143] Specifically, for each detection cycle, the system inputs three values: the temperature gradient, voltage fluctuation frequency, and current imbalance. These values ​​are combined and processed by a nonlinear function to output a comprehensive feature result. The form of the nonlinear function can be flexibly set, such as a three-variable high-order polynomial, a nested sigmoid function, or a neural network hidden layer mapping. In this way, the model can capture the coupled change trends between multiple parameters and distinguish between abnormal fluctuations in a single parameter and coordinated changes in multiple parameters.

[0144] In practical applications, the use of nonlinear joint modeling helps address blind spots in anomaly detection in a variety of complex scenarios. For example, when temperature and current are individually within critical normal ranges, but their combined relationship indicates potential system failure, traditional linear or single-parameter analysis may miss the risk. However, nonlinear combined features can effectively improve the sensitivity and accuracy of anomaly identification.

[0145] Among them, cluster analysis or support vector machine algorithm is applied to the nonlinear combination feature data to identify representative trend shift feature intervals, and the trend shift intensity is numerically calculated to obtain trend shift intensity data, which specifically includes:

[0146] After acquiring the nonlinear combination feature data for each detection cycle, the system uses unsupervised cluster analysis (such as K-means clustering and Gaussian mixture models) or supervised classification algorithms (such as support vector machines (SVMs) and random forests) to comprehensively analyze and classify the combined feature data for all cycles. Cluster analysis automatically divides data into multiple categories or intervals, each representing a specific operating trend or anomaly type. Classification algorithms such as support vector machines (SVMs) utilize labeled normal and abnormal data samples to train the discrimination boundaries and automatically classify unknown data.

[0147] Through this analysis, the system can identify a series of representative trend deviation feature intervals. For example, if a certain category of characteristic data shows a coordinated increase in temperature, voltage, and current parameters, the system will identify this interval as an abnormal segment with a significant trend deviation.

[0148] For each identified trend shift feature interval, the system further calculates the trend shift intensity by calculating the distance between the mean of the interval's combined features and the mean of the reference normal interval, or directly using methods such as the model's classification confidence and cluster center distance. A higher value indicates a stronger abnormal evolution trend in that interval. For example, if the combined features of a trend shift segment are far from the center value of the historical normal segment, or if the model determines it to be highly abnormal, its trend shift intensity score will be correspondingly higher.

[0149] Through the above analysis and quantification, the system can use trend offset intensity data for subsequent abnormal cycle screening and early warning, effectively improving the ability to automatically identify hidden risks under the influence of progressive failures and complex environments.

[0150] In a preferred embodiment of the present invention, based on the abnormal period candidate set, characteristic statistics are calculated for each abnormal period to form an abnormal period feature set, including:

[0151] According to the temperature change gradient sequence in each abnormal period, the temperature gradient mean and temperature gradient standard deviation of the period are calculated to obtain the temperature gradient statistics;

[0152] According to the voltage fluctuation frequency sequence in each abnormal period, the voltage fluctuation frequency extreme value and the voltage fluctuation frequency mean value of the period are calculated to obtain the voltage fluctuation frequency statistical value;

[0153] According to the current imbalance degree sequence in each abnormal cycle, the maximum value and the average value of the current imbalance degree in the cycle are calculated to obtain the current imbalance degree statistical value;

[0154] The temperature gradient statistics, voltage fluctuation frequency statistics, and current imbalance degree statistics are aggregated to form an abnormal period feature set.

[0155] In an embodiment of the present invention, for the identified candidate set of abnormal cycles, the system performs detailed feature statistical analysis on each abnormal cycle. First, the system processes the temperature change gradient sequence within the abnormal cycle and calculates the mean and standard deviation of the temperature gradient within the cycle. This not only reflects the overall rate of temperature change within the cycle, but also reveals its fluctuation amplitude, which helps to judge the risk of temperature out of control. Secondly, the system processes the voltage fluctuation frequency sequence within the abnormal cycle, calculates its extreme value and mean value respectively, accurately depicts the maximum frequency and overall level of voltage fluctuation in the cycle, and provides data support for identifying high-frequency anomalies. At the same time, the system processes the current imbalance degree sequence to obtain the maximum value and mean value of the cycle, so that it can sensitively identify abnormal current distribution and extreme imbalance between single cells.

[0156] All of the above statistical analysis results are systematically summarized to form a set of abnormal cycle features. This feature set, in the form of structured data, provides a comprehensive and detailed foundation for the subsequent construction of abnormal cycle vectors. This multi-index, multi-level feature statistical method ensures a comprehensive characterization of the abnormal cycle operating status, effectively improving the scientific nature and accuracy of abnormal cycle identification and classification. For example, during actual testing, the system can promptly identify the evolution trend of local faults within the battery pack by comparing the dynamic changes in various statistical values, enabling early intervention.

[0157] In a preferred embodiment of the present invention, each feature data in the abnormal period feature set is merged into an abnormal period vector. Each abnormal period vector includes the mean value of the temperature gradient, the extreme value of the voltage fluctuation frequency, and the maximum value of the current imbalance degree, including:

[0158] Based on each abnormal period feature set, perform the following operations:

[0159] Extract the mean value of temperature gradient as the first component of the abnormal period vector;

[0160] Extract the extreme value of voltage fluctuation frequency as the second component of abnormal period vector;

[0161] Extracting the maximum value of the current imbalance degree as the third component of the abnormal period vector;

[0162] The first component, the first component, and the third component are combined and normalized to form an abnormal periodic vector.

[0163] In an embodiment of the present invention, the system, based on obtaining the feature set of each abnormal cycle, further merges the various feature data to construct an abnormal cycle vector. For each abnormal cycle, the system first extracts the mean temperature gradient as the first component of the cycle vector, which is used to describe the average rate of temperature change. Subsequently, the system extracts the extreme value of the voltage fluctuation frequency as the second component. This indicator can effectively reflect the maximum activity of voltage changes in the entire abnormal cycle. The maximum value of the current imbalance is then extracted as the third component, which is used to reflect the maximum imbalance level between single cells in the cycle. The combination of the above three core components as the abnormal cycle vector realizes the quantitative and standardized description of the multi-dimensional characteristics of the abnormal cycle.

[0164] To ensure data comparability across cycles and algorithmic processing stability, the system also normalizes all abnormal cycle vectors. This normalization eliminates the impact of varying physical quantities or cycle data dimensions and scales, ensuring that all abnormal cycle vectors fall within the same standard range. These abnormal cycle vectors can not only be used directly for subsequent template comparison and risk assessment, but also provide the data foundation for advanced functions such as automatic grading of abnormal cycles and multi-dimensional cluster analysis. For example, by performing multi-sample clustering on the normalized vectors, the system can identify regular trends in the health evolution of a battery population and implement intelligent health management.

[0165] In a preferred embodiment of the present invention, similarity calculation is performed between the abnormal period vector and the standard template vector in the risk evolution template library to obtain a similarity score, including:

[0166] Match the abnormal cycle vector to be compared with each standard template vector in the risk evolution template library one by one;

[0167] The similarity algorithm is used to calculate the similarity scores between the abnormal period vector and each standard template vector;

[0168] According to the similarity scores of all standard templates, the template closest to the abnormal period vector and its highest similarity score are screened out as the final similarity score result.

[0169] In an embodiment of the present invention, after obtaining the abnormal cycle vector to be compared, the system pairs it one-to-one with each standard template vector in the risk evolution template library. The template library stores the characteristic vectors of different typical battery failure evolution processes in history, covering a variety of failure types and evolution paths. The system uses multiple similarity algorithms, including cosine similarity, Euclidean distance, and Mahalanobis distance, to calculate the similarity score between the abnormal cycle vector and each template vector. This multi-algorithm scoring mechanism further improves the scientific nature and fault tolerance of the similarity calculation, reducing the distortion risk caused by a single algorithm.

[0170] Based on the similarity scores of all standard templates, the system selects the template that is closest to the current abnormal cycle vector and records the highest similarity score of the template. The final similarity score result can be used as a basis for high-risk judgment of abnormal cycles, and can also be used for subsequent risk classification and failure mode tracing. When the similarity score exceeds the risk threshold preset by the system, the system will automatically mark the abnormal cycle as a high-risk type and give targeted early warning prompts based on the corresponding risk evolution path. In practical applications, this intelligent comparison and scoring mechanism can effectively reduce the uncertainty of human judgment and realize automated and scientific fault trend identification and risk classification management. For example, in the unmanned ship battery pack operation mission, the system successfully identified cycle characteristics that were extremely similar to historical failure cases, issued a high-risk warning in advance, and greatly ensured the safety and continuity of mission execution.

[0171] Among them, the abnormal cycle vector to be compared is matched one by one with each standard template vector in the risk evolution template library, specifically including:

[0172] For each abnormal cycle vector to be compared, the system first selects all stored standard template vectors from the risk evolution template library. Each standard template vector represents a different historical battery failure type, evolution process, or typical operating abnormality scenario. Its characteristics are composed of multiple dimensions such as the mean temperature gradient, the extreme voltage fluctuation frequency, and the maximum current imbalance.

[0173] During implementation, the system pairs the current abnormal cycle vector with each standard template vector in the template library in sequence according to the predetermined template number order, matches the two sets of data dimension by dimension, and unifies the normalization scale of the data to ensure that each component has the same weight and meaning when compared.

[0174] This one-to-one matching ensures that each anomaly cycle vector is fully compared with all known risk evolution patterns, comprehensively covering all possible failure scenarios. For example, if the template library contains ten different standard templates, the system will match the current anomaly cycle vector with each of the ten templates one by one, preparing for the next similarity scoring step.

[0175] Among them, the similarity algorithm is used to calculate the similarity scores between the abnormal period vector and each standard template vector, specifically including:

[0176] After pairing, the system uses a similarity scoring algorithm to compare and analyze each set of abnormal periodic vectors with the standard template vector. The choice of similarity algorithm can be customized based on actual needs and data type. Common methods include the Euclidean distance method based on spatial distance, the cosine similarity method for measuring vector angles, and the Mahalanobis distance method, which considers the weights of different dimensions.

[0177] In practice, the system subtracts each component of the abnormal periodic vector from the template vector, or calculates the distance or angle between them in multidimensional space based on the dot product and module length relationship between the vectors. For example, using Euclidean distance, a smaller distance indicates closer proximity between the two vectors, and a closer fit between the risk patterns. For example, using cosine similarity, a score closer to 1 indicates a greater similarity in direction and a higher degree of feature fit.

[0178] To improve the reliability of the results, the system can combine the results of multiple similarity algorithms and set thresholds for judgment. For example, if the Euclidean distance is lower than a preset standard, or the cosine similarity is higher than a specified value, the abnormal periodic vector is considered to be highly matched with the standard template vector, and the highest similarity score and matching template number are output accordingly.

[0179] In a real-world case, if an unmanned vessel's battery monitoring system detects an abnormal cycle and its feature vector has a similarity score greater than 0.95 (e.g., using a cosine similarity algorithm) with a historical battery overheating failure template in a template library, the system will identify the cycle as high-risk and trigger the corresponding early warning process. This enables intelligent identification and response to potential battery failure trends under complex and changing operating conditions.

[0180] An embodiment of the present invention further provides an unmanned vessel battery safety management and warning system, the system comprising:

[0181] The original parameter acquisition module is used to obtain the original battery parameter data during the operation of the unmanned ship, and the original battery parameter data includes voltage data, current data, temperature data and ambient humidity data;

[0182] The parameter integration processing module is used to perform parameter integration processing on the original battery parameter data, classify and summarize the data according to the battery cell identification, and perform sliding average calculation on the parameters at different time nodes to obtain stability analysis data;

[0183] A difference feature construction module is used to construct a difference feature sequence based on the stability analysis data, wherein the difference feature sequence is formed by calculating the temperature change gradient, voltage fluctuation frequency and current imbalance degree in adjacent time periods;

[0184] A pattern recognition module is used to perform pattern recognition processing on the difference feature sequence and extract trend offset data according to the set nonlinear feature combination rules. The trend offset data is used to identify potential progressive failure modes, which are distinguished from transient features of sudden abnormal conditions;

[0185] The abnormal cycle analysis module is used to extract characteristic change indicators within the continuous detection period based on trend deviation data and construct an abnormal cycle candidate set;

[0186] The risk assessment module is used to convert the abnormal cycle candidate set into an abnormal cycle vector and compare the similarity with the preset risk evolution template library. If the similarity exceeds the preset matching threshold, the abnormal cycle vector is marked as a high-risk type;

[0187] The early warning output module is used to count the number of high-risk abnormal cycles. If the number reaches the preset abnormal threshold, it will generate early warning result data and output a security risk prompt.

[0188] It should be noted that this system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0189] An embodiment of the present invention further provides a computing device comprising: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the above-described method. All implementations in the above-described method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0190] The embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, causes the computer to execute the above-described method. All implementations in the above-described method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0191] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A battery safety management and early warning method for unmanned vessels, characterized in that: The method comprises: Acquiring raw battery parameter data during operation, wherein the raw battery parameter data includes voltage data, current data, and temperature data; Perform parameter integration processing on the original battery parameter data, classify and summarize the data according to the battery cell identification, and perform sliding average calculation on the parameters at different time nodes to obtain stability analysis data; Constructing a difference feature sequence based on the stability analysis data, wherein the difference feature sequence is formed by calculating the temperature change gradient, voltage fluctuation frequency, and current imbalance degree in adjacent time periods; Based on the difference feature sequence, for each detection cycle, the temperature change gradient sequence, voltage fluctuation frequency sequence and current imbalance degree sequence are extracted as input features respectively; Using the multi-dimensional feature fusion method, each input feature is combined and processed to obtain a fused feature sequence; According to the pre-set nonlinear feature combination rules, the fused feature sequence is pattern recognized to identify the nonlinear trend offset in the feature sequence and calculate the trend offset intensity data; Based on the trend shift intensity data, time periods with abnormal changes within multiple consecutive detection cycles are screened out, and the trend shift data within each time period is extracted and summarized. The trend shift data is used to identify potential progressive failure modes, as opposed to transient characteristics of sudden abnormal conditions. Based on the trend deviation data, the characteristic change indicators within the continuous detection period are extracted, and the candidate set of abnormal periods is constructed; The abnormal cycle candidate set is converted into an abnormal cycle vector and compared with the preset risk evolution template library for similarity. If the similarity exceeds the preset matching threshold, the abnormal cycle vector is marked as a high-risk type; The number of abnormal cycles of high-risk types is counted. If the number reaches the preset abnormal threshold, early warning result data is generated and a security risk prompt is output.

2. The unmanned vessel battery safety management and early warning method according to claim 1 is characterized in that: The original battery parameter data is processed by parameter integration, the data is classified and summarized according to the battery cell identification, and the parameters at different time nodes are calculated by sliding average to obtain stability analysis data, including: According to the battery cell identification, the voltage data, current data, temperature data and ambient humidity data in the original battery parameter data are grouped according to the battery cell to obtain a grouped original parameter set; For each group of original parameter sets, the sliding average algorithm is applied to the voltage data, current data, temperature data and ambient humidity data in chronological order to obtain a smoothed parameter sequence; Based on the smooth parameter sequence of each battery cell, the statistical characteristics of each time node are extracted to form preliminary stability data; Based on the preliminary stability data, the parameter change amplitude and fluctuation trend of each battery cell during continuous operation are evaluated to generate stability analysis data for subsequent analysis.

3. The unmanned vessel battery safety management and early warning method according to claim 1 is characterized in that: Based on the stability analysis data, a differential feature sequence is constructed, including: Based on the stability analysis data, the temperature data of each time node in the continuous detection cycle is extracted, the temperature change between adjacent time nodes is calculated, and the temperature change gradient is obtained by dividing the temperature change by the time interval; Based on the stability analysis data, the voltage data of each time node in the continuous detection cycle is extracted, the voltage data is subjected to Fourier transform or zero-crossing statistics, the number of voltage changes per unit time is calculated, and the voltage fluctuation frequency is obtained; Based on the stability analysis data, the current data at each time point within the continuous detection cycle is extracted. The current values ​​of each battery cell are averaged and the absolute deviation of each cell current from the average is calculated. All absolute deviations are summed and normalized to obtain the degree of current imbalance. The calculation results of temperature change gradient, voltage fluctuation frequency and current imbalance degree are arranged in sequence according to the detection time to construct a difference feature sequence that reflects the dynamic changes of battery operation.

4. The unmanned vessel battery safety management and early warning method according to claim 1, characterized in that: The abnormal cycle candidate set is converted into an abnormal cycle vector and compared with the preset risk evolution template library for similarity. If the similarity exceeds the preset matching threshold, the abnormal cycle vector is marked as a high-risk type, including: According to the abnormal period candidate set, characteristic statistics are calculated for each abnormal period to form an abnormal period feature set; Merge the feature data in the abnormal period feature set into an abnormal period vector, each of which contains the mean value of the temperature gradient, the extreme value of the voltage fluctuation frequency, and the maximum value of the current imbalance degree; Calculate the similarity between the abnormal cycle vector and the standard template vector in the risk evolution template library to obtain a similarity score; If the similarity score of the abnormal cycle vector exceeds the matching threshold, the abnormal cycle vector is marked as a high-risk type and the high-risk abnormal cycle is recorded.

5. The unmanned vessel battery safety management and early warning method according to claim 2 is characterized in that: Based on the preliminary stability data, evaluate the parameter variation and fluctuation trend of each battery cell during continuous operation, and generate stability analysis data for subsequent analysis, including: Based on the preliminary stability data, the voltage variation, temperature variation, and current variation of each battery cell in a continuous time window are calculated to obtain parameter variation data; Based on the parameter variation data, the voltage, temperature and current of each battery cell are analyzed for fluctuations. Statistical methods are used to calculate the fluctuation trend data of each parameter within a continuous operation cycle. The parameter change amplitude data and fluctuation trend data are integrated to generate comprehensive stability analysis data for each battery cell within the corresponding detection cycle.

6. The unmanned vessel battery safety management and early warning method according to claim 1 is characterized in that: According to the pre-set nonlinear feature combination rules, the fused feature sequence is pattern recognized to identify the nonlinear trend offset in the feature sequence and calculate the trend offset intensity data, including: Based on the fusion feature sequence, a multivariable nonlinear function is used to jointly model the temperature gradient sequence, voltage fluctuation frequency sequence and current imbalance degree sequence to obtain nonlinear combined feature data. Cluster analysis or support vector machine algorithm is applied to the nonlinear combination feature data to identify the representative trend shift feature intervals, and the trend shift intensity is numerically calculated to obtain the trend shift intensity data.

7. The unmanned vessel battery safety management and early warning method according to claim 4 is characterized in that: Based on the abnormal period candidate set, characteristic statistics are calculated for each abnormal period to form an abnormal period feature set, including: According to the temperature change gradient sequence in each abnormal period, the temperature gradient mean and temperature gradient standard deviation of the period are calculated to obtain the temperature gradient statistics; According to the voltage fluctuation frequency sequence in each abnormal period, the voltage fluctuation frequency extreme value and the voltage fluctuation frequency mean value of the period are calculated to obtain the voltage fluctuation frequency statistical value; According to the current imbalance degree sequence in each abnormal cycle, the maximum value and the average value of the current imbalance degree in the cycle are calculated to obtain the current imbalance degree statistical value; The temperature gradient statistics, voltage fluctuation frequency statistics, and current imbalance degree statistics are aggregated to form an abnormal period feature set.

8. The unmanned vessel battery safety management and early warning method according to claim 7 is characterized in that: Each feature data in the abnormal period feature set is merged into an abnormal period vector. Each abnormal period vector contains the mean value of the temperature gradient, the extreme value of the voltage fluctuation frequency, and the maximum value of the current imbalance degree, including: Based on each abnormal period feature set, perform the following operations: Extract the mean value of temperature gradient as the first component of the abnormal period vector; Extract the extreme value of voltage fluctuation frequency as the second component of abnormal period vector; Extracting the maximum value of the current imbalance degree as the third component of the abnormal period vector; The first component, the first component, and the third component are combined and normalized to form an abnormal periodic vector.

9. The unmanned vessel battery safety management and early warning method according to claim 8, characterized in that: Calculate the similarity between the abnormal cycle vector and the standard template vector in the risk evolution template library to obtain a similarity score, including: Match the abnormal cycle vector to be compared with each standard template vector in the risk evolution template library one by one; The similarity algorithm is used to calculate the similarity scores between the abnormal period vector and each standard template vector; According to the similarity scores of all standard templates, the template closest to the abnormal period vector and its highest similarity score are screened out as the final similarity score result.

Images