A battery intelligent fault diagnosis method and system suitable for a power grid energy storage system

Abstract

The present application relates to the technical field of power grid energy storage system, and particularly to a battery intelligent fault diagnosis method and system suitable for power grid energy storage system. The method obtains multi-source monitoring data including battery electrical parameters, power grid characteristics, environmental conditions and historical operation records, aligns the data in a unified time step and extracts features, and obtains statistical features, time sequence features and frequency domain features. A double-layer diagnosis model structure is adopted, in which a random forest model is used to diagnose main fault modes such as overcharging and overdischarging, and a convolutional neural network model is used to diagnose complex fault modes such as short circuit and temperature out of control. The system has a dynamic updating mechanism, which realizes online optimization by calculating the incremental update value of the model parameters; at the same time, a model evaluation mechanism is established, and when the evaluation index is lower than the preset standard, the retraining process is triggered. The present application realizes high-precision and real-time identification and early warning of battery faults, and significantly improves the operation reliability and safety of the energy storage system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid energy storage systems, in particular to a battery intelligent fault diagnosis method and system suitable for power grid energy storage systems. BACKGROUND

[0002] Among numerous energy storage technologies, electrochemical energy storage systems (such as lithium-ion batteries, lead-acid batteries, etc.) have become the mainstream choice for power grid energy storage due to their high efficiency, fast response, modularity, and scalability. Such battery energy storage systems form battery packs by connecting multiple battery cells in series and parallel to meet the power and capacity requirements of the power grid. However, the performance and lifespan of batteries, as the core components of energy storage systems, are affected by various factors, including the number of charge and discharge cycles, operating temperature, depth of discharge, load fluctuations, and usage environment. When batteries exhibit problems such as capacity degradation, increased internal resistance, and abnormal temperature, not only will the output quality and availability of the energy storage system be affected, but equipment damage and safety risks may also occur. In addition, as the scale of the power grid and the capacity of energy storage continue to increase, the number of batteries is large and the distribution is complex, making it difficult for traditional operation and maintenance modes to meet the requirements of large-scale battery health management.

[0003] Currently, battery fault diagnosis technology is not yet mature, and the main problems are as follows:

[0004] Traditional battery management systems (BMS) mainly rely on basic parameters such as voltage, current, and temperature for state monitoring, making it difficult to capture early signs of potential faults such as increased internal resistance and capacity degradation. When the fault is obvious, it often has a negative impact on the operation of the energy storage system, affecting the stability and reliability of the power grid. In actual power grid energy storage operation, once a battery fault occurs, traditional methods often require manual intervention, resulting in longer repair and recovery times, increased operation and maintenance costs, and reduced reliability and flexibility of the power grid.

[0005] Battery fault diagnosis is independent of power grid dispatching mechanisms and lacks immediate linkage, making it impossible to quickly adjust power grid loads or activate backup energy storage devices based on changes in battery status, thereby weakening the response capabilities of energy storage systems in power grid peak shaving and frequency regulation and emergency power supply.

[0006] Current research relies heavily on empirical models and simple algorithms, lacking deep application of big data and intelligent algorithms, resulting in insufficient prediction accuracy, weak model adaptability, and limited intelligence, which cannot meet the urgent needs of large-scale energy storage systems for high precision, real-time, and intelligent management. SUMMARY

[0007] In view of the problems existing in the prior art, the present application is proposed.

[0008] Therefore, the problem to be solved by the present application is how to solve the problems of the traditional battery management system battery fault detection lag, long recovery time after fault, and disconnection between battery fault detection and grid scheduling.

[0009] To solve the above technical problems, the present application provides the following technical solutions.

[0010] In a first aspect, the embodiments of the present application provide a battery intelligent fault diagnosis method suitable for a grid energy storage system, which comprises: acquiring multi-source monitoring data of a target energy storage system, and performing data fusion on the multi-source monitoring data to form first monitoring data;

[0011] A first fault diagnosis model is preset, and the first monitoring data is input into the first fault diagnosis model;

[0012] According to the output of the first fault diagnosis model, the battery fault intelligent diagnosis is performed.

[0013] As a preferred scheme of the battery intelligent fault diagnosis method suitable for the grid energy storage system, the multi-source monitoring data comprises:

[0014] Battery electrical parameter data, the electrical parameter data comprising voltage, current and temperature parameters;

[0015] Grid characteristic data, the grid characteristic data comprising load power and scheduling plan index;

[0016] Environmental condition data, the environmental condition data comprising environmental temperature and humidity;

[0017] Historical operation record data, the historical operation record data comprising cumulative cycle number and charge-discharge curve.

[0018] As a preferred scheme of the battery intelligent fault diagnosis method suitable for the grid energy storage system, the data fusion to form the first monitoring data comprises:

[0019] Aligning the multi-source monitoring data in a unified time step;

[0020] Extracting statistical features of the multi-source monitoring data, the statistical features comprising mean, variance, skewness and kurtosis;

[0021] Extracting time sequence features of the multi-source monitoring data, the time sequence features comprising trend features and autocorrelation coefficients;

[0022] Extracting frequency domain features of the multi-source monitoring data, the frequency domain features comprising power spectral density.

[0023] As a preferred scheme of the battery intelligent fault diagnosis method suitable for power grid energy storage system, wherein: the first fault diagnosis model comprises:

[0024] A random forest model is used to diagnose the main fault modes of overcharge, overdischarge, overheating, internal resistance increase and capacity attenuation.

[0025] A convolutional neural network model is used to diagnose the complex fault modes of short circuit, damage, temperature out of control, unbalanced charging and group failure.

[0026] The loss function of the first fault diagnosis model comprises a positioning loss, a confidence loss and a classification loss.

[0027] As a preferred scheme of the battery intelligent fault diagnosis method suitable for power grid energy storage system, wherein: the battery fault intelligent diagnosis according to the output of the first fault diagnosis model comprises:

[0028] The posterior probability of each fault type is calculated.

[0029] When the posterior probability of a certain fault type is greater than a preset threshold, it is determined that this type of fault has occurred.

[0030] According to the fault determination result, a warning signal is sent to the battery management system.

[0031] As a preferred scheme of the battery intelligent fault diagnosis method suitable for power grid energy storage system, wherein: further comprising a dynamic updating mechanism:

[0032] According to the newly obtained monitoring data and the model prediction error, the incremental update value of the model parameters is calculated.

[0033] The parameters of the first fault diagnosis model are updated online.

[0034] The fault diagnosis is performed again according to the updated model.

[0035] As a preferred scheme of the battery intelligent fault diagnosis method suitable for power grid energy storage system, wherein: further comprising a model evaluation mechanism:

[0036] The precision, recall and F1 score of the model on the validation set are calculated.

[0037] A confusion matrix is generated to analyze the diagnosis accuracy of each type of fault.

[0038] When the model evaluation index is lower than the preset standard, the model retraining process is triggered.

[0039] In a second aspect, the embodiments of the present application provide a battery intelligent fault diagnosis system suitable for a power grid energy storage system, which comprises a data acquisition module, a multi-source monitoring data acquisition module of a target energy storage system, and a data fusion module for forming first monitoring data from the multi-source monitoring data;

[0040] a preset module for presetting a first fault diagnosis model and inputting the first monitoring data into the first fault diagnosis model;

[0041] an output module for performing battery fault intelligent diagnosis according to the output of the first fault diagnosis model.

[0042] In a third aspect, the embodiments of the present application provide a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program instructions are executed by the processor, the steps of the battery intelligent fault diagnosis method suitable for a power grid energy storage system according to the first aspect of the present application are implemented.

[0043] In a fourth aspect, the embodiments of the present application provide a computer readable storage medium having a computer program stored thereon, wherein when the computer program instructions are executed by a processor, the steps of the battery intelligent fault diagnosis method suitable for a power grid energy storage system according to the first aspect of the present application are implemented.

[0044] The method of the present application has higher accuracy and faster response time in multiple application scenarios compared to traditional fault diagnosis methods such as simple threshold method and traditional algorithm. Through a unique feature extraction strategy, combined with the classification models of Random Forest and Convolutional Neural Network (CNN), the method can accurately identify the multi-dimensional features of battery faults. The feature extraction process not only includes basic statistical features, time series features, but also integrates frequency domain feature analysis, so as to capture the subtle changes and complex fault modes of the battery system. The multi-dimensional feature extraction strategy greatly improves the model's ability to identify different types of faults and effectively reduces the occurrence of false positives and false negatives. In addition, the method ensures

[0045] The method of the present application is outstanding in improving the stability and safety of the power grid energy storage system, especially in the application of early fault warning. Through real-time monitoring and accurate diagnosis, the system can identify and issue warning signals in a timely manner to prevent the further spread and deterioration of faults. This mechanism effectively improves the overall operational stability of the power grid energy storage system, ensuring that the system can maintain balance and stability in the face of renewable energy fluctuations. In addition, early identification of potential battery faults reduces the risk of major accidents such as fires and explosions caused by battery faults, ensuring the safe operation of the energy storage system and surrounding equipment, and improving the safety of the entire power grid system.

[0046] After implementing the intelligent fault diagnosis method of the present application, the power grid energy storage system shows significant improvement in multiple economic indicators. Precise fault diagnosis reduces unnecessary maintenance and replacement costs caused by false positives and false negatives, while extending the service life of the battery and reducing the frequency of battery replacement, thereby significantly reducing operating and maintenance costs. Through real-time monitoring and automated fault diagnosis, maintenance and management personnel can more efficiently perform maintenance and management, reducing the frequency and workload of manual inspections and improving overall maintenance efficiency. Fast and accurate fault diagnosis and timely warning mechanisms reduce system downtime due to faults, improve the availability of the energy storage system and the overall efficiency of the power grid, not only reducing economic losses due to downtime, but also improving the service quality of the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0048] Fig. 1 Flowchart of the battery intelligent fault diagnosis method suitable for the power grid energy storage system;

[0049] Fig. 2 Computer device diagram of the battery intelligent fault diagnosis method suitable for the power grid energy storage system;

[0050] Fig. 3 System architecture diagram of the battery intelligent fault diagnosis method suitable for the power grid energy storage system. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0052] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from the description, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited by the specific embodiments disclosed below.

[0053] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is separate or selectively excluded from other embodiments.

[0054] Embodiment 1

[0055] With reference to Figs. 1-2 For the first embodiment of the present application, the embodiment provides a battery intelligent fault diagnosis method suitable for a power grid energy storage system, comprising,

[0056] S100: Obtain multi-source monitoring data of a target energy storage system, and perform data fusion on the multi-source monitoring data to form first monitoring data;

[0057] In the embodiments of the present application, the multi-source monitoring data mainly comes from a plurality of sensors and monitoring devices deployed in the energy storage system. Specifically, these devices include voltage sensors, current sensors, temperature sensors distributed at key positions of the battery pack, and environment monitors and power grid state monitors. These sensors are designed with industrial standards, have high precision and stability, and ensure the accuracy of data acquisition.

[0058] In an optional embodiment, when the multi-source monitoring data is fused, the sampling frequencies of different data sources can be flexibly configured. For example, the sampling frequency of voltage and current data can be set to 100 Hz to capture rapidly changing electrical characteristics; the sampling frequency of temperature data can be set to 1 Hz because temperature changes relatively slowly; and the sampling frequency of environmental data can be set to 0.1 Hz to meet the needs of environmental monitoring. Through this differentiated sampling strategy, the monitoring accuracy of key parameters is ensured, and the waste of data storage and processing resources is avoided.

[0059] It should be noted that the data acquisition system adopts a hierarchical architecture design, including a field layer, a control layer and a management layer. The field layer is responsible for the acquisition of basic data, and industrial communication protocols such as CAN bus and RS485 are used; the control layer realizes preliminary processing and local storage of data, and PLC or industrial computer is used; the management layer is responsible for comprehensive analysis and storage of data, and is realized based on industrial server. This hierarchical design improves the reliability and scalability of the system.

[0060] S101: The multi-source monitoring data includes:

[0061] Battery electrical parameter data, the electrical parameter data including voltage, current and temperature parameters;

[0062] Power grid characteristic data, the power grid characteristic data including load power and dispatching plan index;

[0063] Environmental condition data, the environmental condition data including environmental temperature and humidity;

[0064] Historical operation record data, the historical operation record data including cumulative cycle number and charge-discharge curve.

[0065] In the embodiments of the present application, for the collection of electrical parameter data: the voltage sensor collects the battery monomer voltage and group voltage, the measurement range is 0-5V (monomer) and 0-1000V (group voltage), and the accuracy is ±0.1%; the current sensor adopts a Hall sensor scheme, the measurement range is ±500A, and the accuracy is ±0.5%; the temperature sensor adopts a PT100 platinum resistance, the measurement range is -40℃-125℃, and the accuracy is ±0.1℃.

[0066] In an optional embodiment, the collection of electrical parameter data can also include internal resistance measurement. By injecting a small signal excitation in the battery charging and discharging process, the internal resistance value of the battery is measured by using an alternating current impedance method. The measurement range of the internal resistance measurement system is 0.1mΩ-100mΩ, the accuracy is ±1%, and the measurement frequency can be adjusted within the range of 0.1Hz-1kHz. This method can find signs of battery performance degradation earlier.

[0067] It should be noted that the collection of electrical parameters adopts a redundant design strategy. Dual sensors are configured at key measurement points, and the two sensors work simultaneously but adopt different measurement principles, such as voltage using a voltage divider method and a differential method, temperature using a thermocouple and a platinum resistance, etc. When one sensor fails or the data is abnormal, the system automatically switches to another sensor, ensuring the continuity and reliability of data collection.

[0068] In the embodiments of the present application, for the collection of grid characteristic data: the load power is monitored in real time by an electric energy analyzer, the sampling rate is 10kHz, and parameters such as active power, reactive power, and power factor can be measured; the scheduling plan index is obtained from the grid scheduling system in real time, including the target power curve, standby capacity requirement, etc.

[0069] In an optional embodiment, the grid characteristic data can also include grid frequency and voltage quality parameters. The frequency measurement accuracy is ±0.01Hz, and the voltage harmonic measurement can reach 50 harmonics. These parameters help to analyze the grid-connected operation state of the energy storage system and discover potential problems in time.

[0070] It should be noted that the system establishes a hierarchical cache mechanism for grid characteristic data. According to the importance and frequency of use of data, the data is divided into three levels of real-time data, minute-level data and hour-level data, which are stored in memory databases, relational databases and time series databases respectively, which not only ensures the data query efficiency, but also saves storage space.

[0071] In the embodiments of the present application, for the monitoring of environmental conditions data: a high-precision temperature and humidity sensor is used, the temperature measurement range is -40℃-85℃, and the accuracy is ±0.3℃; the humidity measurement range is 0-100%RH, and the accuracy is ±2%RH. The sensor is arranged at the air inlet, air outlet and around the key equipment of the energy storage system.

[0072] In an optional embodiment, the environmental monitoring can also include parameters such as dust concentration, flammable gas concentration, etc. For example, PM2.5 and PM10 concentrations are measured using a laser dust sensor, and H2 concentration is monitored using an electrochemical sensor. These parameters are crucial for evaluating the operating environment of the energy storage system.

[0073] It should be noted that the environmental monitoring system adopts a zoning management mode. The energy storage system is divided into battery compartments, PCS compartments, power distribution compartments, and other functional areas, and each area is configured with an independent monitoring unit. The monitoring data is used not only for fault diagnosis, but also for closed-loop control of the environmental control system, such as ventilation, refrigeration, etc.

[0074] In the embodiments of the present application, for the maintenance of historical operation record data: the system automatically records the start and end time, depth, and power of each charge and discharge cycle; the cumulative cycle count is calculated by integration, with an accuracy of 0.1 times; complete charge and discharge curve data, including the time sequence changes of voltage, current, and temperature, are stored.

[0075] In an optional embodiment, the historical record can also include maintenance records and alarm records. The maintenance records contain detailed information of periodic inspections, replacement of components, and other operations; the alarm records contain fault time, type, and handling measures. These records help establish a health record of the equipment and improve diagnostic accuracy.

[0076] It should be noted that the historical data adopts a distributed storage architecture. Recent data (such as the last 3 months) is stored in a local server for fast access; historical data is stored in a cloud data center to support big data analysis. The two-level storage system maintains data consistency through an incremental synchronization mechanism.

[0077] S102: Data fusion to form first monitoring data includes:

[0078] Aligning the time steps of the multi-source monitoring data;

[0079] Extracting statistical features of the multi-source monitoring data, including mean, variance, skewness, and kurtosis;

[0080] Extracting time sequence features of the multi-source monitoring data, including trend features and autocorrelation coefficients;

[0081] Extracting frequency domain features of the multi-source monitoring data, including power spectral density.

[0082] In the embodiments of the present application, the time step alignment adopts a two-level synchronization mechanism. First, a unified clock source is provided for all collection devices through a GPS or Beidou time service module, with a time accuracy better than 1 μs; then, interpolation algorithms are used to align data of different frequencies, supporting linear interpolation, cubic spline interpolation, and other methods.

[0083] In an optional embodiment, the data alignment can also adopt an adaptive step strategy. When the battery state changes sharply (such as sudden load increase or sudden load decrease), the alignment step is automatically reduced to 10 ms to capture the transient characteristics; when in steady state, the step is appropriately increased to 1 s to reduce the data volume. This strategy ensures the accuracy of data at critical moments and improves the efficiency of the system.

[0084] It should be noted that the system establishes a data quality evaluation mechanism. By checking the integrity, accuracy and timeliness of the data, a quality label is assigned to each data point. In subsequent feature extraction, the feature calculation method is dynamically adjusted according to the data quality, such as using robust statistical methods for data with high noise.

[0085] In the embodiments of the present application, the extraction of statistical features adopts a sliding window mechanism. The window length can be configured, and a typical value is 5 minutes; the window overlap rate is 50%, ensuring the continuity of the features; for each window, the mean, variance, skewness, kurtosis and other statistical quantities are calculated to characterize the distribution characteristics of the data.

[0086] In an optional embodiment, more statistical features can also be calculated, such as quantiles, peak factor, waveform factor, etc. These features describe the statistical regularity of the data from different angles, which helps to improve the accuracy of fault diagnosis. In particular, the peak factor plays an important role in detecting sudden faults such as battery short circuit.

[0087] It should be noted that the calculation of statistical features adopts an incremental update algorithm. Whenever new data arrives, only the incremental part related to the old and new data needs to be calculated, greatly reducing the computational load. At the same time, the system periodically performs full calculation to prevent error accumulation in incremental calculation.

[0088] In the embodiments of the present application, the time series feature extraction focuses on the dynamic change law of the data. The trend feature is obtained by least squares fitting, reflecting the change rate of the parameter; the calculation of the autocorrelation coefficient supports multiple delay values (such as 1 minute, 5 minutes, 15 minutes), capturing periodic changes at different time scales.

[0089] In an optional embodiment, the time series features can also include conditional entropy, Lyapunov exponent and other nonlinear features. These features can characterize the dynamic characteristics of the battery system, especially the instability under complex working conditions, providing an important basis for early warning of faults.

[0090] It should be noted that the seasonality and periodicity of the data are considered in the extraction of time series features. The system can automatically identify and compensate for regular changes such as daily and weekly cycles, thereby more accurately extracting abnormal change features. For long-term trends, the sliding baseline method is used to eliminate the drift effect.

[0091] In the embodiments of the present application, the frequency domain feature extraction is implemented by using fast Fourier transform (FFT). A Hanning window function is used to reduce spectral leakage, and the transform point number is 8192 to ensure the frequency resolution. The energy distribution of the signal in each frequency band is analyzed by calculating the power spectral density.

[0092] In an optional embodiment, the frequency domain analysis can also use wavelet transform or Hilbert-Huang transform and other time-frequency analysis methods. These methods can provide joint time-frequency distribution information of the signal, and are particularly suitable for analyzing transient processes and non-stationary signals in the battery system.

[0093] It should be noted that an automatic frequency band division strategy is used in the extraction process of the frequency domain features. According to the characteristic frequencies of different types of faults, the frequency spectrum is divided into multiple monitoring bands. The range of each frequency band can be dynamically adjusted according to actual operation experience to improve the specificity of the features.

[0094] S200: preset a first fault diagnosis model, and input first monitoring data into the first fault diagnosis model;

[0095] In the embodiments of the present application, the first fault diagnosis model adopts a double-layer structure. The first layer uses a random forest model to identify common faults such as overcharging, over-discharging, etc.; and the second layer uses a convolutional neural network to process complex fault modes. The two layers work cooperatively to ensure the diagnosis speed and improve the accuracy.

[0096] In an optional embodiment, an attention mechanism can also be introduced into the model. By learning the importance weights of different features, the most critical indicators for the current fault judgment are automatically focused on. This mechanism improves the interpretability of the model and improves the diagnosis performance.

[0097] It should be noted that the training of the model adopts an online-offline hybrid mode. The basic model is obtained by offline training using historical fault data and expert annotated samples; and after deployment, it is continuously optimized by using an incremental learning method to adapt to the dynamic changes of the system. Cross-validation is used in the training process to ensure the generalization ability of the model.

[0098] S201: the first fault diagnosis model comprises:

[0099] a random forest model for diagnosing main fault modes such as overcharging, over-discharging, overheating, internal resistance increase and capacity attenuation;

[0100] a convolutional neural network model for diagnosing complex fault modes such as short circuit, damage, temperature out of control, unbalanced charging and group fault;

[0101] The loss function of the first fault diagnosis model includes a positioning loss, a confidence loss and a classification loss.

[0102] In the embodiments of the present application, the configuration of the random forest model is as follows: the number of decision trees is set to 100 to balance the calculation efficiency and the performance of the model; the maximum depth of each tree is 10 layers to prevent overfitting; the Gini coefficient evaluation index is used for feature selection; and the random sampling with weight is used for sample sampling, and the sampling weight of rare fault types is appropriately increased.

[0103] In an optional embodiment, the random forest model can also integrate multiple sub-models, each of which is specifically designed to handle a type of fault. For example, the overcharge and over-discharge sub-model focuses on voltage features, the overheat sub-model focuses on temperature features, and the internal resistance and capacity sub-model mainly analyzes impedance spectrum features. This specialized design improves the diagnostic accuracy of the model.

[0104] It should be noted that the random forest model uses a distributed training framework. The training data is sharded by fault type, and the sub-models are trained in parallel on multiple computing nodes, and then the results are integrated through a voting mechanism. The system supports hot updates, and the model parameters can be updated without stopping.

[0105] In the embodiments of the present application, the convolutional neural network model adopts the ResNet architecture and contains 18 convolutional layers. The input layer receives multi-channel time series data, each channel corresponding to a monitoring parameter; the middle layer uses 1D convolution to extract time series features; and the fully connected layer completes the final classification judgment. The model training uses the Adam optimizer with an initial learning rate of 0.001 and uses the cosine annealing strategy.

[0106] In an optional embodiment, the convolutional neural network can adopt a multi-branch structure. The main branch processes the original time series data, and the auxiliary branch processes the statistical features and frequency domain features. The outputs of multiple branches are combined through a feature fusion layer to fully utilize the advantages of different types of features. This design improves the recognition ability of the model for complex faults.

[0107] It should be noted that in order to improve the robustness of the model, a variety of data augmentation techniques are used in the training process. Including adding Gaussian noise, random scaling, time shifting, etc. Dropout and L2 regularization are used to prevent overfitting. The output of the middle layer of the model can be visualized to facilitate understanding of the decision-making process of the network.

[0108] In the embodiments of the present application, the design of the loss function is as follows: the positioning loss adopts mean square error (MSE) to evaluate the prediction accuracy of the fault occurrence time; the confidence loss uses cross-entropy to measure the reliability of the prediction probability; and the classification loss uses Focal Loss to focus on difficult classification samples. The total loss is obtained by weighted summation of the three losses.

[0109] In an optional embodiment, the loss function can also include a temporal consistency loss. This loss term penalizes sharp fluctuations in the prediction results, making the diagnosis results more stable. At the same time, a knowledge distillation loss is introduced to guide model learning using the output of the expert system.

[0110] It should be noted that the weight coefficients of the loss function adopt an adaptive adjustment mechanism. In the early stage of training, the weight of the classification loss is increased to accelerate convergence; as the training progresses, the weights of the positioning loss and the confidence loss are gradually increased to improve the accuracy of the model. The system records the loss distribution of each batch for evaluating the training effect.

[0111] S300: intelligently diagnosing the battery fault according to the output of the first fault diagnosis model.

[0112] S301: intelligently diagnosing the battery fault according to the output of the first fault diagnosis model includes:

[0113] calculating the posterior probability of each fault type;

[0114] when the posterior probability of a certain fault type is greater than a preset threshold, determining that the fault type;

[0115] sending a warning signal to the battery management system according to the fault determination result.

[0116] In the embodiments of the present application, the calculation of the posterior probability adopts a Bayesian framework. First, the probability distribution of the fault type output by the model is obtained, and then the posterior probability is calculated in combination with prior knowledge (such as equipment age, historical faults, etc.). The system maintains historical records of probability calculation for optimizing the warning threshold.

[0117] In an optional embodiment, the posterior probability calculation can also consider the influence of environmental factors and operating conditions. For example, appropriately lowering the warning threshold of the overheat fault in the high-temperature season, and increasing the detection sensitivity of the internal resistance fault in the heavy load working condition. This context-aware diagnosis strategy improves the accuracy of the warning.

[0118] It should be noted that the system adopts a multi-level warning threshold design. For different types of faults, three levels of thresholds of warning, alarm, and fault are set. When the posterior probability exceeds the corresponding threshold, different levels of response measures are triggered. The threshold can be dynamically adjusted according to the operation and maintenance experience and the importance of the system.

[0119] In the embodiments of the present application, the generation of the warning signal includes multiple steps: first, determining the severity and urgency of the fault; then generating a standard format warning message containing information such as fault type, location, and time; finally, sending the warning through multiple channels to ensure that the information is delivered in time.

[0120] In an optional embodiment, the early warning system can also automatically generate treatment recommendations. Based on historical fault treatment experience, the system recommends appropriate treatment plans, including emergency measures and maintenance recommendations. These recommendations can help operations personnel respond quickly to failures.

[0121] It should be noted that the delivery of early warning information adopts a hierarchical push strategy. Different levels of information are pushed to different personnel. Field maintenance personnel receive detailed technical parameters, and management personnel receive fault profiles and impact assessments. The system supports early warning confirmation and treatment progress tracking.

[0122] S302: Also includes a dynamic update mechanism:

[0123] According to the newly acquired monitoring data and model prediction error, calculate the incremental update value of the model parameters;

[0124] Online update of the parameters of the first fault diagnosis model;

[0125] According to the updated model, re-perform fault diagnosis.

[0126] In the embodiments of the present application, the incremental update adopts a sliding window mechanism. The last 30 days of operation data are retained, and the model parameter update is performed every 24 hours. The update process uses a stochastic gradient descent algorithm, and the learning rate is adaptively adjusted according to the prediction error.

[0127] In an optional embodiment, the update mechanism can also include dynamic adjustment of the model structure. For example, when a new fault mode is found, the output layer of the model is automatically expanded; when the discriminative ability of some features decreases, the network structure is appropriately adjusted. This adaptive architecture improves the scalability of the model.

[0128] It should be noted that the model update adopts a gradual strategy to avoid drastic changes in parameters. Before each update, the effect of the new parameters is evaluated on the validation set, and only when the performance improves is the update applied. The system saves the model parameters of the last 5 versions, supporting fast rollback.

[0129] In the embodiments of the present application, the trigger conditions for model update include: the prediction error exceeds a set threshold; a certain number of new samples are accumulated; a new fault type is found; the system operating conditions change significantly, etc. The update process is performed on a standby server, which does not affect the operation of the main system.

[0130] In an optional embodiment, the update process can also incorporate expert knowledge. Through a human-computer interaction interface, experts are allowed to review and correct the diagnosis results of the model, and these correction opinions are used as new training samples for model optimization. This human-machine collaborative approach improves the reliability of the model.

[0131] It should be noted that the system establishes a model performance tracking mechanism. The evaluation indicators before and after each update are recorded, such as accuracy, recall rate, F1 score, etc. Through trend analysis of these indicators, the health status of the model is evaluated, and performance degradation problems are discovered and handled in a timely manner.

[0132] S303: also includes a model evaluation mechanism:

[0133] Calculate the precision, recall rate and F1 score of the model on the validation set;

[0134] Generate a confusion matrix to analyze the diagnostic accuracy of each type of failure;

[0135] When the model evaluation indicators are lower than the preset standards, trigger the model retraining process.

[0136] In the embodiments of the present application, the model evaluation adopts a hierarchical verification mechanism. The validation set is sampled according to the failure type, ensuring that there are enough samples for each type of failure. The evaluation indicators include: precision, recall rate, F1 score, ROC curve and AUC value. The system automatically calculates these indicators every hour and generates an evaluation report.

[0137] In an optional embodiment, the evaluation can also include timeliness analysis of the model. By comparing the prediction accuracy of different time windows, the stability and timeliness of the model are evaluated. For example, calculate the performance indicators in the 24-hour, 7-day and 30-day sliding windows respectively, and timely discover the degradation trend of the model performance. This multi-scale evaluation method helps to fully understand the running status of the model.

[0138] It should be noted that the evaluation process adopts a weighted calculation method. Higher weight is given to high-risk failure types (such as short circuit, thermal runaway), and lower weight is given to low-risk failure types (such as slight capacity attenuation). This differentiated evaluation strategy is more in line with actual application needs.

[0139] In the embodiments of the present application, the confusion matrix analysis adopts an automated tool. The system generates a detailed confusion matrix report, including:

[0140] The accuracy, false negative rate and false positive rate of each type of failure; identification of easily confused failure types; confidence distribution of fault judgment; feature analysis of misdiagnosis cases.

[0141] In an optional embodiment, the confusion matrix analysis can also take into account the severity of the failure. The failure is divided into several levels according to the severity, and the diagnostic accuracy of different levels of failure is calculated respectively. For misdiagnosis of high-level failures, the system will trigger a manual review mechanism. At the same time, by analyzing the misdiagnosis reasons, the fault features and diagnosis rules are continuously optimized.

[0142] It should be noted that the system establishes a perfect evaluation result tracking mechanism. All evaluation data is stored in a special performance database, supporting historical trend analysis and comparison. Evaluation reports are automatically pushed to relevant personnel, and manual intervention is triggered if necessary.

[0143] In the embodiments of the present application, the triggering conditions of model retraining include: overall accuracy is lower than 90%; recall rate of certain key fault is lower than 85%; false positive rate exceeds 5%; performance indicators decrease for 72 consecutive hours; new fault samples exceed 20% of existing samples.

[0144] In an optional embodiment, the retraining process can also include data cleaning and feature reconstruction. Before starting the training, the historical data is evaluated and cleaned to remove noise samples and outdated data. At the same time, based on the feature analysis of the new samples, the feature engineering scheme is optimized. This preprocessing can improve the effect of retraining.

[0145] It should be noted that the retraining adopts a step-by-step iterative strategy: first, use the latest data for small-scale training and evaluate the effect; if the effect is significantly improved, expand the training scale; perform final training on the complete data set; maintain the online service capability of the model during the training process, and use a dual model switching mechanism.

[0146] In the embodiments of the present application, the model deployment process after retraining includes: verifying the performance of the new model in the test environment; performing A / B testing to compare the effects of the new and old models; gradually switching business traffic to the new model; continuously monitoring the running state of the new model; retaining rollback mechanism to ensure system reliability.

[0147] In an optional embodiment, a model version management mechanism can also be established. Each version of the model has a complete configuration record, including: feature distribution of the training data set; model structure and parameters; evaluation results and deployment history; version update description and responsible person.

[0148] It should be noted that the system comprehensively evaluates the retraining results, including: performance improvement degree analysis; resource consumption evaluation; influence on various fault diagnoses; comparison with historical versions; stability and reliability verification.

[0149] Through this complete evaluation and retraining mechanism, it is ensured that the fault diagnosis model always maintains the best state, providing reliable protection for the safe operation of the energy storage system.

[0150] Further, the embodiments also provide a battery intelligent fault diagnosis system suitable for a power grid energy storage system, comprising,

[0151] The data acquisition module acquires multi-source monitoring data of the target energy storage system, and performs data fusion on the multi-source monitoring data to form first monitoring data;

[0152] a preset module, presetting a first fault diagnosis model, and inputting the first monitoring data into the first fault diagnosis model;

[0153] an output module, intelligently diagnosing the battery fault according to the output of the first fault diagnosis model.

[0154] In summary, the present application proposes a battery intelligent fault diagnosis method and system suitable for power grid energy storage systems, which realizes high-precision, real-time identification and early warning of battery faults by combining data-driven machine learning models with the physical mechanism of the battery. Compared with traditional fault diagnosis methods, the method of the present application has significant advantages in accuracy and response speed, can more effectively identify various fault types, reduce false positives and false negatives, and significantly improve the overall operation reliability of the system.

[0155] One of the key innovations of the present application is the unique feature extraction strategy. By deeply mining the multi-dimensional features of the battery and the grid state, including statistical features, time series features and frequency domain features, and combining advanced classification models (random forest and convolutional neural network) to handle main fault modes and complex fault modes respectively, the fault diagnosis model can more accurately identify the health status and fault mode of the battery. This feature extraction method can capture subtle changes and potential faults in battery operation, thereby improving the accuracy of diagnosis.

[0156] In addition, the fault warning method of the present application can continuously monitor the real-time operation of the battery system based on the self-adaptive updating of the classification model, and timely issue a warning when the fault risk is higher than the set threshold. By combining real-time data with historical data, the system not only improves the accuracy of the warning, but also dynamically adjusts the fault diagnosis strategy according to the changing working conditions, ensuring that the power grid energy storage system operates efficiently and safely.

[0157] This method has good scalability and adaptability, and is suitable for different types and scales of power grid energy storage systems, and can meet the development needs of future smart grids and microgrids. Through modular design and standardized interface, the method of the present application can be easily integrated into existing power grid management systems, supporting the expansion and upgrading of the system, and has broad application prospects and market potential.

[0158] In summary, the battery intelligent fault diagnosis method of the present application not only significantly improves the safety and economic benefits of the power grid energy storage system, but also provides solid technical support for intelligent and data-driven power system management. This method has important engineering value and promotion significance in practical application, and can effectively promote the development of the power system towards more efficient and intelligent direction.

[0159] Example 2

[0160] Reference Figs. 2-3For the second embodiment of the present application.

[0161] The battery operating state of the grid energy storage system is affected by multiple factors: the electrical characteristics of the battery (voltage, current, temperature), grid load and dispatch information, environmental conditions (such as temperature, humidity), and the battery's historical operating records (charging and discharging strategy, cycle count, capacity decay, etc.). The present application constructs a stable and rich feature space at the data level, ensuring that key fault signals can be extracted from multi-source, multi-dimensional data.

[0162] Unified representation of multi-source data and time series alignment

[0163] By obtaining multi-source data from the battery, the grid, and environmental monitoring, the data is aligned using a unified time step, and noise is removed through Kalman filtering to ensure data stability and reliability.

[0164] Set the discrete time step as Δt, and at each time t (t = 0, Δt, 2Δt,...), obtain the corresponding observation data from the energy storage system and grid monitoring platform. The basic data includes:

[0165] Battery electrical parameters: voltage V(t), current I(t), temperature T(t);

[0166] Grid feature parameters: load power P load (t), dispatch plan index S schcd (t);

[0167] Environmental conditions: external sensor data for temperature and humidity, represented as E cnv (t) feature set Historical operating records and statistics: cumulative cycle count C cycle (t), specific charging and discharging curve features The above-mentioned dimensional information at time t forms the original data vector:

[0168] X(t) = [V(t), I(t), T(t), P load (t), S schcd (t), E cnv (t), C cycle (t),]←

[0169] Since the sampling frequencies of different sensors and data sources may be different, the present application uses linear interpolation to align the time series. Linear interpolation calculates the interpolation value x(t') at the interpolation point t' through the observation values x(t1) and x(t2) of two adjacent time points:

[0170]

[0171] Ensure that all observation data have consistent resolution at the unified time step Δt.

[0172] For the noise signal that may exist in the sensor data, the present application adopts Kalman filtering technology to denoise the data. Kalman filtering combines prior estimates and observations through a prediction-update iterative process to obtain denoised data values

[0173]

[0174] where K(t) is the Kalman gain, expressed as:

[0175]

[0176] P(t|t-1) is the predicted covariance, and R is the measurement noise covariance. Through Kalman filtering, the data noise is effectively smoothed, which is suitable for dynamic and complex power grid operating environment.

[0177] Feature extraction: To reduce redundant information and improve the recognition value of data to the model, the present application adopts feature extraction and dimension reduction technology to extract key features related to fault signals and life attenuation based on the original data vector X(t). The extracted features are as follows:

[0178] Statistical features: Calculate the mean, variance, skewness, kurtosis and other statistical indicators of each data channel. These statistical features help to capture the global characteristics of the battery and power grid. Specifically, they include:

[0179] Mean: represents the average level of the data channel within a certain time window, which is used to capture long-term trends.

[0180]

[0181] Variance: measures the degree of data fluctuation, reflecting the stability of the battery or power grid.

[0182]

[0183] Skewness: represents the asymmetry of data distribution. For battery health state monitoring, skewness helps to identify abnormal changes.

[0184]

[0185] Kurtosis: measures the "sharpness" of data distribution, which helps to identify high-frequency fluctuation fault features

[0186]

[0187] Maximum and minimum values: reflect the extreme behavior of the battery charging and discharging process, such as the risk of overcharging or overdischarging.

[0188] Max(X) = max{X(t1),X(t2),X(t3)} n )}

[0189] Min(X)=min{X(t1),X(t2),X(t n )}

[0190] (2) Temporal characteristics

[0191] The trend characteristics and autocorrelation coefficients of each data channel are calculated using the sliding window method to further extract dynamic features related to faults. Specifically, this includes:

[0192] Trend: Indicates the trend of data changes within a certain time window.

[0193]

[0194] Autocorrelation coefficient: Measures the correlation of data over time, capturing periodic or recurring patterns. It is of great significance for understanding the charge-discharge behavior during battery cycling.

[0195]

[0196] (3) Frequency domain characteristics

[0197] The Fast Fourier Transform (FFT) is used to convert the signal from the time domain to the frequency domain, and the power spectral density in the frequency domain is extracted. These frequency domain features help to identify high-frequency or low-frequency signals of the battery system under different operating conditions, especially when monitoring the battery health status, it can identify high-frequency noise or abnormal vibration modes.

[0198] Spectrum analysis: Obtaining frequency domain characteristics by calculating the power spectral density of battery voltage or current.

[0199]

[0200] Where P(f) is the spectral power, f is the frequency, and t is the frequency. i These are the sampling points for the time series.

[0201] After extracting effective features, this invention employs Principal Component Analysis (PCA) to reduce the dimensionality of the extracted high-dimensional features, ensuring data compactness and discriminative power. Let the original high-dimensional feature matrix be Z = [z1, z2, ..., z...]. m ], where each z i This represents the extracted i-th feature. The oblique variance matrix C is calculated as follows:

[0202]

[0203] Where n is the number of samples. is the transpose of the feature matrix. Then, by performing eigenvalue decomposition on the covariance matrix C, the principal component z' corresponding to the largest eigenvalue is extracted:

[0204]

[0205] where w is the principal component weight vector, representing the projection of the feature in the principal component space.

[0206] After dimensionality reduction, to ensure the efficiency and accuracy of the model, the present application further screens the most meaningful features for battery state of health diagnosis and failure prediction through information gain evaluation and classification contribution analysis. Information gain is used to evaluate the correlation of each feature with the failure category, from which the key features that can maximize the performance of the classification model are selected.

[0207] In addition, by analyzing the contribution of the features under different failure modes, the feature set is further optimized, redundant features are removed, and features most closely related to the state of health and failure mode are retained. The final reduced feature set Z' will be used as the input of the diagnosis and prediction model.

[0208] Data dynamic updating mechanism

[0209] To adapt to the dynamic operating environment of the grid energy storage system, the present application introduces a sliding window updating mechanism in the feature extraction process. For window length τ, the sliding feature φ win is calculated as:

[0210] φ win (X(t)) = φ(X(t-τ), X(t-τ+1), X(t)),

[0211] where X(t) is the original data vector at the current time, X(t-τ), X(t-τ+1), X(t) are the data at the past τ time. Within the window, the feature function φ calculates the comprehensive characteristics of the data in this time period, ensuring that the extracted features always reflect the latest dynamic changes of the battery under the current operating conditions.

[0212] Through the above data fusion and feature extraction strategy, the present application realizes the multi-source integration, noise processing and feature optimization of the original data, providing a stable, reliable and engineering valuable input for the intelligent fault diagnosis model.

[0213] Intelligent fault diagnosis mechanism

[0214] This invention proposes a mechanism for real-time battery fault diagnosis in grid energy storage systems based on a data-driven intelligent classification model. This mechanism extracts and analyzes multi-source feature data, combined with a posterior probability judgment model, to achieve accurate identification and early warning of various battery fault types. The method of this invention is particularly suitable for grid energy storage systems with complex battery operating environments and variable operating conditions.

[0215] Fault type definition and model input

[0216] Battery malfunctions can involve various types. For ease of diagnosis, the following set of malfunctions is defined:

[0217] F =

[0218] {Overcharge, over-discharge, overheating, increased internal resistance, capacity decay, short circuit, damage, temperature runaway, charging imbalance, group indicates possible fault types and normal states.}

[0219] This set includes common failure types (primary failure modes) and more complex failure types that are difficult to identify using traditional machine learning methods (complex failure modes). Specific failure types are as follows:

[0220] Main failure modes (identified via random forest):

[0221] Overcharge: The battery voltage exceeds the maximum safe voltage, which may cause a fire or explosion.

[0222] Over-discharge: The battery voltage drops below the safe threshold, which may damage the battery.

[0223] Overheating: The battery temperature is too high, which can lead to thermal runaway or performance degradation.

[0224] Increased internal resistance: Increased internal resistance due to battery aging or damage affects charging and discharging efficiency.

[0225] Capacity degradation: The battery capacity gradually decreases, resulting in a reduction in usable energy.

[0226] Complex Failure Modes (identified via Convolutional Neural Networks (CNNs)):

[0227] Short circuit: A short circuit occurs inside the battery or between batteries, causing the battery to overheat rapidly.

[0228] Damage: Physical damage or cracks occur inside the battery, leading to a decrease in battery performance.

[0229] Temperature runaway (thermal runaway): An abnormal increase in battery temperature may trigger a thermal runaway reaction.

[0230] Uneven charging: Uneven charging among multiple battery modules can lead to overcharging or over-discharging of some modules.

[0231] Group failure: the performance of the whole battery pack or system is degraded due to multiple cell failures or system management problems.

[0232] At time t, the feature vector Z'(t) obtained after feature extraction and dimension reduction is input into the fault diagnosis classifier f diag , and the posterior probability of each fault type is calculated:

[0233] p(f|Z'(t)) = f diag (Z'(t)), f F

[0234] The classifier constructs the posterior probability distribution based on the training data and uses different machine learning algorithms for classification learning. Specifically, the Random Forest is used for classification of the main fault mode, and the Convolutional Neural Network (CNN) is used for further analysis and diagnosis of the complex fault mode.

[0235] Fault determination and early warning mechanism

[0236] According to the posterior probability p(f|Z'(t)) output by the classifier, fault determination is performed by maximizing the criterion:

[0237]

[0238] where f * (t) is the fault type with the maximum probability in the diagnosis result. The conditions for fault determination are:

[0239] 1. The determined fault type f * (t) is not normal

[0240] 2. The corresponding posterior probability p(f * (t)|Z'(t)) > eta, where eta is the threshold for fault determination.

[0241] If both conditions are met, the system sends a fault warning signal. The warning result can be directly fed back to the battery management system (BMS) and the grid dispatching center to support timely operation and maintenance response.

[0242] Adaptive fault model updating mechanism

[0243] The invention introduces an online updating mechanism to adapt to the dynamic working conditions of the energy storage system. The parameter set of the classifier is defined as diag , and the classification model is optimized by incremental learning method. The updating rule is:

[0244]

[0245] where is the increment adjusted according to the latest data, which is calculated based on the new observation data Z'(t) and the prediction error of the model.

[0246]

[0247] where α is the learning rate, is the classification error loss function, y t is the true fault label (e.g., obtained through historical validation).

[0248] Model Evaluation and Validation

[0249] To ensure the effectiveness and reliability of the intelligent fault diagnosis mechanism proposed in the present invention, the following evaluation methods are used to validate the classification model:

[0250] Evaluation Indicators:

[0251] (1) Precision

[0252] Precision is used to measure the proportion of actual faults among the samples predicted as faults by the model, focusing on the control of false positive rate. The calculation formula is:

[0253]

[0254] where TP is the true positive, and FP is the false positive.

[0255] (2) Recall

[0256] Recall is used to measure the ability of the model to identify actual fault samples, focusing on the control of false negative rate. The calculation formula is:

[0257] where TP is the true positive, and FN is the false negative.

[0258] (3) F1 Score

[0259] F1 Score is the harmonic mean of precision and recall, considering both false positives and false negatives, and is suitable for scenarios with unbalanced data categories. The calculation formula is:

[0260]

[0261] (4) Confusion Matrix

[0262] Confusion Matrix shows the prediction results of the model for each fault mode, helping to analyze the performance of the model on different fault types. The composition of the confusion matrix includes:

[0263] True Positive (TP): The number of samples predicted as faults and actually as faults.

[0264] True Negative (TN): The number of samples predicted as normal and actually normal.

[0265] False Positive (FP): The number of samples predicted as fault and actually normal (false alarm).

[0266] False Negative (FN): The number of samples predicted as normal and actually fault (missed alarm).

[0267] The confusion matrix helps developers identify the performance of the model under different fault types, facilitating adjustment and optimization.

[0268] Through the above evaluation indicators, the accuracy, stability and reliability of the intelligent fault diagnosis method of the application in actual application can be comprehensively evaluated, so as to ensure that the model can effectively identify battery faults and reduce false alarms and missed alarms.

[0269] Embodiment 3

[0270] The embodiment also provides a computer device suitable for a battery intelligent fault diagnosis method suitable for a power grid energy storage system, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the power grid forced oscillation detection and positioning method proposed in the above embodiment.

[0271] The embodiment also provides a storage medium having a computer program stored thereon, and the program is executed by a processor to realize the power grid forced oscillation detection and positioning method proposed in the above embodiment.

[0272] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. In addition, the input device can be an external keyboard, touchpad or mouse, etc.

[0273] If the functions are implemented in software, the functions can be stored in or implemented as one or more instructions or code on a computer-readable medium. Computer-readable media include both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage medium can be any available medium that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, or twisted pair, then the coaxial cable, fiber optic cable, or twisted pair are included in the definition of medium. Disk and disc, as used herein, include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), and Blu-Ray® disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0274] In other words, like a human driver of a vehicle, the autonomous vehicle 100 can be programmed to follow traffic laws and rules of the road, and to make decisions based on its programming and the information it receives from its sensors and other sources. The autonomous vehicle 100 can also be programmed to make decisions based on its programming and the information it receives from its sensors and other sources, even if those decisions are not in accordance with traffic laws and rules of the road. For example, the autonomous vehicle 100 can be programmed to avoid a collision with another vehicle, even if doing so would violate a traffic law or rule of the road.

[0275] In other words, like a human driver of a vehicle, the autonomous vehicle 100 can be programmed to follow traffic laws and rules of the road, and to make decisions based on its programming and the information it receives from its sensors and other sources. The autonomous vehicle 100 can also be programmed to make decisions based on its programming and the information it receives from its sensors and other sources, even if those decisions are not in accordance with traffic laws and rules of the road. For example, the autonomous vehicle 100 can be programmed to avoid a collision with another vehicle, even if doing so would violate a traffic law or rule of the road.

[0276] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technology, known in the art, or combinations thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon an application of data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

[0277] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, which should be covered by the claims of the present application.

Claims

1. A battery intelligent fault diagnosis method suitable for a power grid energy storage system, characterized in that: The method comprises: obtaining multi-source monitoring data of a target energy storage system, and performing data fusion on the multi-source monitoring data to form first monitoring data; The multi-source monitoring data comprises: Battery electrical parameter data, the electrical parameter data comprising voltage, current and temperature parameters; Power grid characteristic data, the power grid characteristic data comprising load power and scheduling plan indicators; Environmental condition data, the environmental condition data comprising environmental temperature and humidity; Historical operation record data, the historical operation record data comprising cumulative cycle times and charge-discharge curves; The data fusion to form the first monitoring data comprises: Performing uniform time step alignment on the multi-source monitoring data; Extracting statistical features of the multi-source monitoring data, the statistical features comprising mean, variance, skewness and kurtosis; Extracting time sequence features of the multi-source monitoring data, the time sequence features comprising trend features and autocorrelation coefficients; Extracting frequency domain features of the multi-source monitoring data, the frequency domain features comprising power spectral density; Pre-setting a first fault diagnosis model, and inputting the first monitoring data into the first fault diagnosis model; The first fault diagnosis model comprises: A random forest model for diagnosing main fault modes of overcharging, overdischarging, overheating, internal resistance increase and capacity attenuation; A convolutional neural network model for diagnosing complex fault modes of short circuit, damage, temperature out of control, unbalanced charging and group failure; The loss function of the first fault diagnosis model comprises positioning loss, confidence loss and classification loss; Performing intelligent battery fault diagnosis according to the output of the first fault diagnosis model.

2. The battery intelligent fault diagnostic method for power grid energy storage system of claim 1, wherein: The intelligent battery fault diagnosis according to the output of the first fault diagnosis model comprises: Calculating posterior probabilities of each fault type; When the posterior probability of a certain fault type is greater than a preset threshold, determining that the fault type is the fault; According to the fault determination result, issuing a warning signal to a battery management system.

3. The battery intelligent fault diagnostic method for power grid energy storage system of claim 2, wherein: The method further comprises a dynamic updating mechanism: According to newly obtained monitoring data and model prediction errors, calculating an incremental update value of model parameters; Performing online updating on parameters of the first fault diagnosis model; Performing fault diagnosis again according to the updated model.

4. The battery intelligent fault diagnostic method for power grid energy storage system of claim 3, wherein: The method further comprises a model evaluation mechanism: Calculating precision, recall and F1 score of the model on a validation set; Generating a confusion matrix to analyze diagnosis accuracy of each fault type; When the model evaluation index is lower than a preset standard, triggering a model retraining process.

5. A battery intelligent fault diagnosis system for grid energy storage system based on the battery intelligent fault diagnosis method for grid energy storage system according to any one of claims 1-4, characterized in that: The method further comprises a data acquisition module for acquiring multi-source monitoring data of a target energy storage system, and performing data fusion on the multi-source monitoring data to form first monitoring data; A preset module for pre-setting a first fault diagnosis model, and inputting the first monitoring data into the first fault diagnosis model; An output module for performing intelligent battery fault diagnosis according to the output of the first fault diagnosis model. 6.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to implement the steps of the battery intelligent fault diagnosis method for a power grid energy storage system according to any one of claims 1-4.

7. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the battery intelligent fault diagnosis method for a power grid energy storage system according to any one of claims 1-4.

Images