Safety monitoring method for local and cloud intelligent interconnection energy storage equipment

Abstract

The invention provides a local and cloud intelligent interconnection energy storage equipment safety monitoring method. The method comprises the steps of energy storage equipment data collection, analysis and cleaning, feature extraction and normalization processing, input of local and cloud algorithm modules for safety monitoring, execution of alarm processing measures, storage of uploading processing records and realization of data interaction. According to the invention, the safety monitoring efficiency and accuracy of the energy storage equipment can be improved, and stable operation of the energy storage system and real-time interaction of data are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of safety monitoring of intelligent energy storage devices. More specifically, the present invention relates to a method for safety monitoring of a locally and cloud-intelligently interconnected energy storage device. Background Art

[0002] Existing safety monitoring technologies for energy storage devices mainly rely on local monitoring systems. These systems can usually only process a limited amount of data and lack sufficient computing power to analyze large-scale data sets. In terms of technical principles, these systems may adopt basic data collection and simple anomaly detection algorithms, but often cannot effectively identify and predict complex fault patterns. In addition, due to the complex and changeable operating environment of energy storage devices, a single monitoring system is difficult to adapt to all situations, resulting in limited monitoring efficiency and accuracy. In terms of data collection and processing, existing technologies may not support multiple communication protocols, resulting in limited compatibility and integrity of data collection. At the same time, the methods of data cleaning and feature extraction may not be efficient enough to meet the needs of real-time monitoring. In terms of safety monitoring, existing algorithms may lack deep learning capabilities and are difficult to accurately predict the health status and potential risks of energy storage devices.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: The existing methods for monitoring energy storage devices have limitations in data processing and safety monitoring, cannot achieve efficient data interaction and accurate fault prediction, and need to be improved in terms of real-time performance and accuracy. Summary of the Invention

[0004] The present invention provides a method for safety monitoring of a locally and cloud-intelligently interconnected energy storage device, including:

[0005] Collecting energy storage device data by using a transmission protocol support module that supports specific multiple underlying communication protocols;

[0006] Parsing and real-time cleaning the collected data;

[0007] Performing feature extraction and normalization processing on the cleaned data;

[0008] Inputting the processed data into a local real-time diagnosis algorithm module and a cloud prediction algorithm module respectively for safety monitoring;

[0009] Executing corresponding warning and processing measures according to the results of the local and cloud algorithms, and locally saving the processing records and uploading the processing records to the cloud;

[0010] Implementing data interaction between the local EMS and the cloud platform, including data sending, control instruction issuing, execution control, and execution result feedback.

[0011] Further, during the data acquisition process, when it is detected that the device is powered on, the communication interface is connected, and the energy storage device transmits data to the server side, the original data transmitted by the device is received through the Transport function module that supports three transmission protocols, namely TCP, RUT, and QUIC. Then, the data protocol conversion is completed through the Code module. Among them, decoding is performed on the data sent by the device to decode the original data into a format that the system can process or that is convenient for developers to process, and encoding supports the instructions and data sent to the device.

[0012] Further, the real-time cleaning processes the entire time series in a sliding window mode. For random data missing situations, the linear regression estimation method is used to fill in the missing parts; for sudden outlier data, the size of the sliding window is determined based on the data characteristics and the required smoothing degree. The data points within the window are summed up, and the sum value is divided by the window size to obtain the average value for substitution. Each time the window is moved by one data point until the entire time series is processed.

[0013] Further, the outlier data judgment uses the formula

[0014]

[0015] where x i is the sample data (i = 1, 2, …, N, N is the total number of samples), is the sample mean, the detection level α is determined, and the critical value table value b ′ p (n) of the kurtosis test is looked up. When b k > b ′ p (n), it is determined that the value farthest from the mean is an outlier. After removing the outlier, the kurtosis test method is repeated to check whether there are still outliers. If not, no outliers are found.

[0016] Further, feature extraction obtains data features from multiple perspectives, including time-domain features, frequency-domain features, and time-series features, to obtain the high-level feature set Q1, ensuring that the single-feature Gini coefficient R is not less than 0.02, obtaining the low-dimensional feature set Q2, and then obtaining the fusion feature set M.

[0017] Further, the time-domain features include:

[0018] Average value: where x i is

[0019] the sample data (i = 1, 2, …, N, N is the total number of samples);

[0020] Peak-to-peak value: t2 = max(x i ) - min(x i );

[0021] Root mean square:

[0022] Shock factor: t4 = max(x i ) / t3;

[0023] Form factor:

[0024] Skewness index:

[0025] Standard deviation:

[0026] Kurtosis index:

[0027] Amplitude factor: t9 = max(x i ) / t3;

[0028] Margin factor:

[0029] Furthermore, the frequency domain features transform the time domain waveform of the signal into a frequency spectrum signal by means of discrete Fourier transform, where x(kΔt) represents the acquisition value of the signal, N is the number of sampling points, Δt is the sampling interval, k is the sequence number of the time domain discrete value, and the frequency domain features include:

[0030] Center frequency:

[0031] Average frequency:

[0032] Frequency variance:

[0033] Root mean square frequency:

[0034] Furthermore, the extraction of the timing features includes the following steps:

[0035] Establish a data relationship through the AR(p) model:

[0036] Set x i+1 = φ1x i + φ2x i-1 + … + φ p x i-p+1 + ξ i+1 , where φ j (j = 1, 2, …, p) is the linear correlation coefficient, and ξ i+1 is the noise;

[0037] Define the partial autocorrelation coefficient of the sequence under the AR(p) model as: pcaf(p) = φ p ;

[0038] Construct the Yule-Walker equation:

[0039] Construct the equation Multiply both sides of the equation by x i-k+1 , take the expectation of the resulting expression, remove the noise term, then divide the equation by N - k, and at the same time let c l = c -l , thus obtaining the autocovariance c of the sequence k , then divide both sides of the equation by c0 to obtain the autocorrelation coefficient of the sequence, and after arrangement, obtain the matrix equation Rφ = r. Since the R matrix is a symmetric full-rank matrix and invertible, calculate to obtain Then the partial autocorrelation coefficient of the lag order k

[0040] Extract features: Extract the autocorrelation coefficient and partial autocorrelation coefficient of each attribute to form time series features.

[0041] Furthermore, the normalization uses the Sigmoid function to map the variable to [0, 1]. The derivative of the Sigmoid function

[0042] Furthermore, the local algorithm unit of the local real-time diagnosis algorithm module includes a voltage anomaly detection algorithm, a cell connection fault detection algorithm, a longitudinal energy storage temperature detection algorithm based on a neural network, and an energy storage battery abnormal temperature rise rate detection algorithm; the cloud algorithm unit of the cloud prediction algorithm module includes the estimation of the state of health (SOH) of a single cell, the prediction of the voltage of a single cell, and the prediction of internal short circuit.

[0043] According to the above embodiments of the present invention, it has at least the following beneficial effects: This local and cloud intelligent interconnected energy storage device safety monitoring method can effectively integrate multiple underlying communication protocols, collect energy storage device data through the transmission protocol support module, and achieve comprehensive data collection and compatibility. By performing real-time cleaning and feature extraction on the collected data, this method can improve the accuracy and efficiency of data processing and ensure the reliability of monitoring results. Combining the local real-time diagnosis algorithm and the cloud prediction algorithm module, this method can comprehensively evaluate the safety status of the energy storage device, and timely discover and handle potential safety problems.

[0044] In addition, by implementing the data interaction between the local EMS and the cloud platform, this method can enhance the real-time and dynamic nature of the data, and improve the speed of issuing control instructions and feedback of execution results. Such a data interaction mechanism can make the monitoring and management of energy storage devices more flexible and responsive, thereby enhancing the safety and stability of the entire energy storage system. Meanwhile, this method also has the ability to accurately identify and process outliers. By removing outliers through the kurtosis test method, the quality of the data and the accuracy of the monitoring algorithm can be ensured, further enhancing the effectiveness of the safety monitoring of energy storage devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown by way of example and not limitation, wherein:

[0046] Figure 1 is a schematic flowchart of a method for safe monitoring of local and cloud intelligent interconnected energy storage devices provided by an embodiment of the present invention;

[0047] Figure 2 is a schematic diagram of the data interaction process between the local EMS and the cloud platform provided by an embodiment of the present invention;

[0048] Figure 3 schematically shows a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] The principles and spirit of the present invention will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and then implement the present invention, and do not limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to be able to fully convey the scope of the present invention to those skilled in the art.

[0050] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention can be specifically implemented in the following forms: completely hardware, completely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0051] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0052] Next, refer to Figure 1 , Figure 1Schematic flow diagram of a local and cloud intelligent interconnected energy storage device safety monitoring method provided by an embodiment of the present invention. As Figure 1 shown, a local and cloud intelligent interconnected energy storage device safety monitoring method includes:

[0053] Collect energy storage device data using a transmission protocol support module that supports specific multiple underlying communication protocols;

[0054] Parse and perform real-time cleaning on the collected data;

[0055] Extract features and perform normalization processing on the cleaned data;

[0056] Input the processed data into a local real-time diagnosis algorithm module and a cloud prediction algorithm module respectively for safety monitoring;

[0057] Execute corresponding alarm and handling measures according to the results of the local and cloud algorithms, and perform local storage of processing records and upload processing records to the cloud;

[0058] As Figure 2 shown, a schematic diagram of the data interaction process between a local EMS and a cloud platform, including realizing the data interaction between the local EMS and the cloud platform, including data sending, control instruction issuing, execution control, and execution result feedback.

[0059] It should be noted that this method involves a local and cloud intelligent interconnected energy storage device safety monitoring method, aiming to monitor the status of the energy storage device in real time to ensure its safe operation. This method is implemented through three main steps: local data collection, data preprocessing, and cloud-edge combined intelligent safety monitoring algorithm. Among them, local data collection refers to the process of directly obtaining raw data from the energy storage device, and the cloud-edge combined intelligent safety monitoring algorithm refers to an algorithm that combines the advantages of local computing and cloud computing to analyze and process data.

[0060] Specifically, the local data collection step includes support for multiple transmission protocols, underlying control, and communication data protocol parsing. In this process, first, a local data collection pre-transmission protocol support module is used to access and support several existing transmission protocols, such as TCP, RUT, QUIC, etc. The acquisition module underlying control module is responsible for input port connection signals, judging whether it meets the startup requirements, and outputting startup program signals. The communication data protocol parsing module is responsible for receiving data input signals, converting, adapting, and parsing multiple device vendor data protocols, and finally outputting the collected data. The parameter settings of these modules will be determined according to the actual device characteristics and communication protocol requirements.

[0061] Preferably, in order to improve the efficiency and accuracy of data collection, the steps of data collection can be further refined. For example, a data verification link can be added to ensure that the collected data meets the expected format and quality.

[0062] Furthermore, more advanced data compression technologies can be considered to reduce the bandwidth consumption during data transmission. Alternative solutions include using wireless sensor networks to expand the scope of data collection, or adopting more intelligent data collection strategies, such as event-triggered collection, where data collection and transmission are only performed when an abnormal event is detected.

[0063] In some embodiments, during the data collection process, when it is detected that the device is powered on, the communication interface is connected, and the energy storage device transmits data to the server side, the original data transmitted by the device is received through the Transport function module that supports three transmission protocols, namely TCP, RUT, and QUIC. After that, the data protocol conversion is completed through the Code module. Among them, decoding is for the data sent by the device, decoding the original data into a format that the system can process or is convenient for developers to process, and encoding supports the instructions and data sent to the device.

[0064] It should be noted that the data preprocessing steps in this method include data cleaning, feature extraction, and normalization. Data preprocessing refers to processing the collected original data to facilitate subsequent analysis and monitoring. Data cleaning refers to correcting errors and outliers in the data, while feature extraction refers to identifying key information helpful for monitoring from the data. Normalization processing refers to converting the data to a unified range, usually between 0 and 1.

[0065] Specifically, the data preprocessing steps involve the following modules: a data cleaning module, a feature extraction module, and a normalization module. The data cleaning module uses a sliding window mode, moving the window one data point each time until it gradually moves to the end of the sequence, and gradually processing the entire time series in turn. For randomly missing data, the linear regression estimation method is used to fill in the missing part; for sudden outlier data, the sliding window size is determined according to the characteristics of the data and the required smoothness degree, the sum of all data points within the window is calculated, and the average value obtained by dividing the sum by the window size is used to replace it. The feature extraction module is used to extract data characteristics, and the normalization module is responsible for normalizing all data and converting it into values between 0 and 1.

[0066] Preferably, to improve the effectiveness and efficiency of data preprocessing, the steps of data cleaning and feature extraction can be further refined. For example, in the data cleaning phase, more complex anomaly detection algorithms can be introduced, such as machine learning-based anomaly detection methods, to more accurately identify and handle outliers. In the feature extraction phase, various feature selection techniques can be employed, such as information gain-based or model-based feature selection, to select the most influential features.

[0067] Furthermore, different functions can be used for normalization processing, such as Min-Max scaling or Z-score standardization, to adapt to different data distribution characteristics. Alternative solutions include using other data preprocessing techniques, such as wavelet transform or principal component analysis (PCA), to further reduce the dimensionality of the data and extract key information.

[0068] In some embodiments, the real-time cleaning processes the entire time series in a sliding window mode. For random data missing cases, linear regression estimation methods are used to fill in the missing parts; for sudden outlier data, the size of the sliding window is determined based on the data characteristics and the required smoothing degree. The data points within the window are summed up and the sum value is divided by the window size to obtain the average value for replacement. Each time the window is moved by one data point until the entire time series is processed.

[0069] It should be noted that the steps of the cloud-edge combined intelligent security monitoring algorithm in this method include a local real-time diagnosis algorithm module and a cloud prediction algorithm. The cloud-edge combination here refers to the allocation of data processing and analysis tasks to the local end (edge computing) and the cloud to achieve resource optimization and efficiency improvement. The local real-time diagnosis algorithm module is responsible for the real-time monitoring and diagnosis of the key parameters of the energy storage device, while the cloud prediction algorithm conducts predictive analysis based on historical and real-time data.

[0070] Specifically, in the cloud-edge combined intelligent security monitoring algorithm, the local real-time diagnosis algorithm module includes a voltage anomaly detection algorithm, a cell connection fault detection algorithm, a neural network-based longitudinal energy storage temperature detection algorithm, and an energy storage battery abnormal temperature rise rate detection algorithm. The specific parameter settings of these algorithms will be determined according to the technical specifications and historical data of the energy storage device. For example, the voltage anomaly detection algorithm may set a normal operating voltage range, and once the measured voltage exceeds this range, the system will issue a warning. The cloud prediction algorithm module includes single cell SOH (State of Health) estimation, single cell voltage prediction, internal short circuit prediction, etc. These algorithms will use machine learning models, train the models based on historical data, and use the models to predict the future state of the device.

[0071] Preferably, to improve the effectiveness and response speed of the cloud-edge integrated intelligent security monitoring algorithm, the implementation details of the algorithm can be further refined. For example, in the local real-time diagnosis algorithm module, more sensor data such as current and pressure can be introduced to provide more comprehensive device status information. The cloud prediction algorithm module can adopt more advanced machine learning frameworks such as deep learning or reinforcement learning to improve the accuracy of prediction.

[0072] Furthermore, edge computing technology can be considered to transfer some computing tasks from the cloud to a location closer to the data source to reduce latency and bandwidth consumption. Alternative solutions include using other data processing and analysis technologies such as rule engines or expert systems to adapt to different application scenarios and requirements.

[0073] In some embodiments, the outlier data judgment uses the formula

[0074]

[0075] where x i is the sample data (i = 1, 2, …, N, N being the total number of samples), is the sample mean, the detection level α is determined, and the critical value table value b ′ p (n) of the kurtosis test is looked up. When b k > b ′ p (n), the value farthest from the mean is determined to be an outlier. After removing the outlier, the kurtosis test method is repeated to check whether there are still outliers. If not, no outliers are found.

[0076] It should be noted that the local real-time diagnosis algorithm module mentioned in this method is responsible for tasks such as voltage anomaly detection and cell connection fault detection. The local real-time diagnosis algorithm module refers to a series of algorithms running locally on the energy storage device, which can quickly respond to changes in the device status and perform diagnosis. The voltage anomaly detection algorithm is an algorithm used to monitor whether the voltage of the energy storage device is within the normal range, and the cell connection fault detection algorithm is used to detect whether the connection between cells is stable.

[0077] Specifically, the local real-time diagnosis algorithm module receives the cleaned data through the first input unit and then uses the local algorithm unit for analysis and diagnosis. For example, the voltage anomaly detection algorithm may set a threshold. When the monitored voltage exceeds this threshold, the system will consider it abnormal and issue an alarm. The cell connection fault detection algorithm may analyze the current and voltage differences between cells to determine whether the connection is normal. The specific parameters of these algorithms, such as the threshold and response time, will be set and adjusted according to the specific requirements of the energy storage device and historical operation data.

[0078] Preferably, to improve the accuracy and efficiency of the local real-time diagnosis algorithm module, the implementation details of these algorithms can be further refined. For example, the voltage anomaly detection algorithm can adopt an adaptive threshold technique to dynamically adjust the normal range of voltage according to historical data. The cell connection fault detection algorithm can integrate a machine learning model to improve the accuracy of fault detection.

[0079] Furthermore, more sensor data such as temperature and humidity can be considered to be introduced to provide more comprehensive device status information. Alternative solutions include using other monitoring techniques such as vibration analysis or sound monitoring to assist in diagnosing cell connection faults. Through these methods, the safety monitoring ability of the energy storage device can be further improved.

[0080] In some embodiments, feature extraction obtains data features from multiple perspectives, including time-domain features, frequency-domain features, and time-series features, to obtain a high-level feature set Q1, ensuring that the single-feature Gini coefficient R is not less than 0.02, obtaining a low-dimensional feature set Q2, and further obtaining a fused feature set M.

[0081] It should be noted that the cloud prediction algorithm module in this method includes single-cell SOH estimation, single-cell voltage prediction, internal short circuit prediction, etc. The cloud prediction algorithm module refers to a series of algorithms running in the cloud that use a large amount of data stored in the cloud for analysis to predict the performance and potential problems of the energy storage device. Single-cell SOH estimation refers to evaluating the health status of a single battery unit, and internal short circuit prediction refers to predicting possible short circuit situations inside the battery.

[0082] Specifically, the cloud prediction algorithm module receives the cleaned data and normalized data through the second input unit, and then uses the cloud algorithm unit for analysis and prediction. For example, single-cell SOH estimation may use historical operation data and battery models to evaluate the degradation degree of the battery unit. The single-cell voltage prediction algorithm may predict future voltage changes based on the battery's usage pattern and environmental conditions. The internal short circuit prediction algorithm may analyze the resistance and current changes inside the battery to predict the short circuit risk. The parameter settings of these algorithms will depend on the battery's technical specifications, historical performance data, and real-time monitoring data.

[0083] Preferably, to improve the prediction accuracy and efficiency of the cloud prediction algorithm module, the implementation details of these algorithms can be further refined. For example, single-cell SOH estimation can adopt machine learning techniques such as random forest or gradient boosting machine to improve the prediction accuracy. Single-cell voltage prediction can combine time series analysis and pattern recognition techniques to capture the trends and patterns of voltage changes. Internal short circuit prediction can adopt advanced signal processing techniques such as wavelet transform or Fourier transform to analyze the electrical signals inside the battery.

[0084] Furthermore, more data sources can be considered, such as the battery's usage history, maintenance records, and environmental monitoring data, to provide a more comprehensive basis for prediction. Alternative solutions include using other prediction models, such as neural networks or support vector machines, to adapt to different prediction requirements and data characteristics. Through these methods, the predictive maintenance ability of energy storage devices can be further improved.

[0085] In some embodiments, the time domain features include:

[0086] Average value: where x i is

[0087] sample data (i = 1, 2, …, N, where N is the total number of samples);

[0088] Peak-to-peak value: t2 = max(x i ) - min(x i );

[0089] Root mean square:

[0090] Shock factor: t4 = max(x i ) / t3;

[0091] Form factor:

[0092] Skewness index:

[0093] Standard deviation:

[0094] Kurtosis index:

[0095] Amplitude factor: t9 = max(x i ) / t3;

[0096] Margin factor:

[0097] It should be noted that the cloud prediction algorithm module mentioned in this method specifically includes single cell SOH estimation, single cell voltage prediction, internal short circuit prediction, etc. Here, SOH estimation refers to the assessment of the battery's state of health, which is an indicator measuring the current health status of the battery relative to its initial state; single cell voltage prediction refers to predicting the voltage change of a single battery cell; internal short circuit prediction refers to predicting possible short circuit phenomena inside the battery.

[0098] Specifically, the cloud prediction algorithm module receives the data that has been cleaned and normalized through the second input unit, and then the cloud algorithm unit uses this data for analysis and prediction. For example, estimating the SOH of a single cell may use parameters such as the historical charge and discharge data of the battery, the current voltage and temperature, and estimate the health state of the battery through algorithms such as Kalman filtering or neural networks. Predicting the voltage of a single cell may be based on the usage pattern and environmental conditions of the battery, and use time series analysis methods such as the ARIMA model to predict future voltage changes. Predicting internal short circuits may analyze the resistance and current changes inside the battery and use machine learning algorithms to predict the short circuit risk. The parameter settings of these algorithms will depend on the technical specifications of the battery, historical performance data, and real-time monitoring data.

[0099] Preferably, to improve the prediction accuracy and efficiency of the cloud prediction algorithm module, the implementation details of these algorithms can be further refined. For example, when estimating the SOH of a single cell, multiple parameters such as the number of battery cycles, charge and discharge efficiency, and internal impedance changes can be combined, and a deep learning framework such as a convolutional neural network (CNN) can be used to improve the prediction accuracy. In predicting the voltage of a single cell, external environmental factors such as temperature and humidity can be introduced to improve the generalization ability of the model. In predicting internal short circuits, more advanced signal processing techniques such as singular value decomposition (SVD) or independent component analysis (ICA) can be used to analyze the electrical signals inside the battery.

[0100] Furthermore, more data sources can be considered, such as the usage history of the battery, maintenance records, and environmental monitoring data, to provide a more comprehensive basis for prediction. Alternative solutions include using other prediction models, such as support vector machines (SVM) or ensemble learning methods, to adapt to different prediction requirements and data characteristics. Through these methods, the predictive maintenance ability of energy storage devices can be further improved.

[0101] In some embodiments, the frequency domain features transform the time domain waveform of the signal into a spectrum signal by means of discrete Fourier transform, where x(kΔt) represents the acquisition value of the signal, N is the number of sampling points, Δt is the sampling interval, k is the sequence number of the time domain discrete value, and the frequency domain features include:

[0102] Center frequency:

[0103] Average frequency:

[0104] Frequency variance:

[0105] Root mean square frequency:

[0106] It should be noted that the cloud prediction algorithm module mentioned in this method specifically includes functions such as single-cell SOH estimation, single-cell voltage prediction, and internal short-circuit prediction. Here, the cloud prediction algorithm module refers to a series of algorithms deployed in the cloud, which use the powerful computing power of the cloud to deeply analyze and predict the data of energy storage devices. Single-cell SOH estimation refers to the process of quantitatively evaluating the health status of a single battery unit, while internal short-circuit prediction refers to predicting possible short-circuit phenomena inside the battery.

[0107] Specifically, the cloud prediction algorithm module receives the data that has been cleaned and normalized through the second input unit, and then the cloud algorithm unit uses this data for analysis and prediction. For example, single-cell SOH estimation may use parameters such as the historical charge and discharge data, current voltage, and temperature of the battery, and estimate the health status of the battery through algorithms such as Kalman filtering or neural networks. Single-cell voltage prediction may use time series analysis methods such as the ARIMA model based on the usage pattern and environmental conditions of the battery to predict future voltage changes. Internal short-circuit prediction may analyze the resistance and current changes inside the battery and use machine learning algorithms to predict the short-circuit risk. The parameter settings of these algorithms will depend on the technical specifications of the battery, historical performance data, and real-time monitoring data.

[0108] Preferably, in order to improve the prediction accuracy and efficiency of the cloud prediction algorithm module, the implementation details of these algorithms can be further refined. For example, when performing single-cell SOH estimation, multiple parameters such as the number of battery cycles, charge and discharge efficiency, and internal impedance changes can be combined, and a deep learning framework such as convolutional neural network (CNN) can be used to improve the prediction accuracy. In single-cell voltage prediction, external environmental factors such as temperature and humidity can be introduced to improve the generalization ability of the model. In internal short-circuit prediction, more advanced signal processing techniques such as singular value decomposition (SVD) or independent component analysis (ICA) can be used to analyze the electrical signals inside the battery.

[0109] Furthermore, more data sources can be considered, such as the usage history of the battery, maintenance records, and environmental monitoring data, to provide a more comprehensive basis for prediction. Alternative solutions include using other prediction models, such as support vector machine (SVM) or ensemble learning methods, to adapt to different prediction requirements and data characteristics. Through these methods, the predictive maintenance ability of energy storage devices can be further improved.

[0110] In some embodiments, the temporal feature extraction includes the following steps:

[0111] Establish a data relationship through an AR(p) model:

[0112] Set x i+1 = φ1x i + φ2xi-1 + … + φ p x i-p+1 + ξ i+1 , where φ j (j = 1, 2, …, p) is the linear correlation coefficient, and ξ i+1 is noise;

[0113] Define the partial correlation coefficient of the sequence under the AR(p) model as: pcaf(p) = φ p ;

[0114] Construct the Yule - Walker equation:

[0115] Construct the equation Multiply both sides of the equation by x i-k+1 , take the expectation of the resulting expression, remove the noise term, then divide the equation by N - k, and at the same time let c l = c -l , thus obtaining the autocovariance c k , then divide both sides of the equation by c0 to obtain the autocorrelation coefficient of the sequence, and after rearrangement, obtain the matrix equation Rφ = r. Since the R matrix is a symmetric full - rank matrix and invertible, calculate to obtain Then the partial correlation coefficient of the lag order k

[0116] Extract features: Extract the autocorrelation coefficient and partial correlation coefficient of each attribute to form time - series features.

[0117] It should be noted that the time - series feature extraction mentioned in this method includes establishing data relationships through the AR(p) model. Time - series feature extraction refers to extracting features from time - series data that are helpful for analysis and prediction, and the AR(p) model is an autoregressive model used to describe the linear relationship between a variable and its past values. Here, p represents the maximum number of lag time points considered in the model.

[0118] Specifically, in the time - series feature extraction step, the AR(p) model expresses the relationship between time - series data points by setting a linear equation, for example:

[0119] X t+1 = φ1X t + φ2X t-1 + … + φ p X t-p + ∈ t+1

[0120] where X t represents the value of the time series at time point t, φ i is the model parameter, and ∈ t+1is the noise term. The determination of model parameters is usually achieved by minimizing the sum of the squares of the prediction errors, and this process is called parameter estimation. Specific methods of parameter estimation can include the least squares method, maximum likelihood estimation, etc.

[0121] Preferably, in order to improve the accuracy and efficiency of time series feature extraction, the implementation details of the AR(p) model can be further refined. For example, when determining the model parameters, information criteria (such as the Akaike Information Criterion AIC or the Bayesian Information Criterion BIC) can be used to select the optimal lag number p to avoid overfitting or underfitting of the model.

[0122] Furthermore, external explanatory variables can be introduced to construct an ARIMAX model to consider other factors that may affect the time series. In practical applications, seasonal ARIMA models can also be considered to process time series data with seasonal characteristics. Alternative solutions include using state space models and Kalman filters to process more complex time series data, or using machine learning methods such as Recurrent Neural Networks (RNN) or Long Short-Term Memory Networks (LSTM) to capture non-linear patterns and long-term dependencies in time series data. Through these methods, the performance of time series analysis can be further improved.

[0123] In some embodiments, the normalization uses the Sigmoid function to map the variable to [0,1], and the derivative of the Sigmoid function

[0124] It should be noted that the normalization process mentioned in this method uses the Sigmoid function to map the variable to the interval [0,1]. Normalization processing refers to scaling the data to a specific numerical range, usually between 0 and 1, so that different features can be compared and processed. The Sigmoid function is a mathematical function commonly used in machine learning and neural networks, and its shape is S-shaped, which can map any real value to the interval (0,1).

[0125] Specifically, the mathematical expression of the Sigmoid function is

[0126]

[0127] where e is the base of the natural logarithm, approximately equal to 2.71828. The derivative of this function, σ ′(x) = σ(x)(1 - σ(x)), which represents the slope of the function at any point. In the normalization process, the Sigmoid function is used to convert the eigenvalue of the original data into a value within the range of [0, 1], which can ensure the numerical stability and convergence of the data when performing machine learning or other algorithm processing. In terms of parameter settings, the Sigmoid function does not require additional parameters because it is a function with a fixed form.

[0128] Preferably, in order to improve the effect and applicability of the normalization process, the application method of the Sigmoid function can be further refined. For example, when processing features with different magnitudes, the features can be standardized before applying the Sigmoid function to eliminate the influence of the dimension.

[0129] Furthermore, other forms of the Sigmoid function can be considered, such as variants with adjustable parameters, to adapt to specific data distributions. Alternative solutions include using other normalization techniques, such as Min - Max scaling or Z - score normalization, which can provide different normalization effects according to the characteristics and requirements of the data. In practical applications, the choice of which normalization method depends on the characteristics of the data and the requirements of subsequent processing.

[0130] In some embodiments, the local algorithm unit of the local real - time diagnosis algorithm module includes a voltage anomaly detection algorithm, a cell connection fault detection algorithm, a neural - network - based longitudinal energy - storage temperature detection algorithm, and an energy - storage battery abnormal temperature - rise rate detection algorithm; the cloud algorithm unit of the cloud prediction algorithm module includes single - cell SOH estimation, single - cell voltage prediction, and internal short - circuit prediction.

[0131] It should be noted that the local real - time diagnosis algorithm module mentioned in this method includes a voltage anomaly detection algorithm, a cell connection fault detection algorithm, a neural - network - based longitudinal energy - storage temperature detection algorithm, and an energy - storage battery abnormal temperature - rise rate detection algorithm. The local real - time diagnosis algorithm module refers to a series of algorithms running locally on the energy - storage device, which can quickly respond to changes in the device state and perform diagnosis. The voltage anomaly detection algorithm is an algorithm used to monitor whether the voltage of the energy - storage device is within the normal range, while the cell connection fault detection algorithm is used to detect whether the connection between cells is stable.

[0132] Specifically, the local real-time diagnosis algorithm module receives the cleaned data through the first input unit and then analyzes and diagnoses it using the local algorithm unit. For example, the voltage anomaly detection algorithm may set a threshold. When the monitored voltage exceeds this threshold, the system will consider it abnormal and issue an alarm. The cell connection fault detection algorithm may analyze the current and voltage differences between cells to determine whether the connection is normal. The energy storage temperature longitudinal detection algorithm based on neural network may use neural network to analyze temperature data, predict and detect temperature anomalies. The energy storage battery abnormal temperature rise rate detection algorithm may monitor whether the rising rate of the battery temperature is abnormal. The specific parameters of these algorithms, such as thresholds and response times, will be set and adjusted according to the specific requirements of the energy storage device and historical operation data.

[0133] Preferably, to improve the accuracy and efficiency of the local real-time diagnosis algorithm module, the implementation details of these algorithms can be further refined. For example, in the voltage anomaly detection algorithm, an adaptive threshold technique can be adopted to dynamically adjust the normal range of voltage according to historical data. In the cell connection fault detection algorithm, a machine learning model can be integrated to improve the accuracy of fault detection. In the energy storage temperature longitudinal detection algorithm based on neural network, deep learning techniques such as convolutional neural network (CNN) or recurrent neural network (RNN) can be used to better capture patterns in time series data. In the energy storage battery abnormal temperature rise rate detection algorithm, physical models and statistical methods can be combined to improve the prediction ability of abnormal temperature rise. Alternative solutions include using other monitoring techniques such as vibration analysis or sound monitoring to assist in diagnosing cell connection faults. Through these methods, the safety monitoring ability of the energy storage device can be further enhanced.

[0134] The above-mentioned embodiments of the present invention have the following beneficial effects: The local and cloud intelligent interconnected energy storage device safety monitoring method described in the present invention can provide a comprehensive safety monitoring solution. It not only covers data collection, parsing, cleaning, feature extraction and normalization processing, but also integrates local real-time diagnosis and cloud prediction algorithm modules to achieve efficient monitoring of the safety of energy storage devices. This method can ensure that when an abnormal situation is detected, warning and handling measures can be quickly executed, and by locally saving and uploading processing records to the cloud, the traceability of data and the transparency of the system can be enhanced.

[0135] Furthermore, by adopting advanced data processing technologies and algorithms, this method can improve the accuracy and response speed of energy storage device monitoring. For example, data missing can be processed through the sliding window mode and linear regression estimation method, and outliers can be identified and removed by using the kurtosis test method, which can ensure the integrity and reliability of the data. The multi-angle method of feature extraction and normalization processing can enhance the generalization ability of the model and improve the accuracy of monitoring results. In addition, the combined use of local real-time diagnosis algorithms and cloud prediction algorithms can provide more accurate technical support for the health management and fault prediction of energy storage devices, thus effectively preventing and reducing potential safety risks.

[0136] Reference is now made to Figure 3 , which shows a schematic structural diagram of a structure 300 of an electronic device suitable for use in implementing some embodiments of the present invention. The electronic devices in some embodiments of the present invention may include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The terminal devices shown are merely examples and should not impose any limitations on the functions and usage scopes of the embodiments of the present invention.

[0137] As Figure 3 shown, the electronic device 300 may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to the programs stored in the read-only memory (ROM) 302 or the programs loaded from the storage device 308 into the random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other through a bus 304. The input / output (I / O) interface 305 is also connected to the bus 304.

[0138] Generally, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 can allow the electronic device 300 to communicate with other devices wirelessly or wiredly to exchange data. Although Figure 3 shows the electronic device 300 having various devices, it should be understood that it is not required to implement or include all the shown devices. More or fewer devices may be alternatively implemented or included. Figure 3Each box shown in the figure may represent a device, or multiple devices as required.

[0139] Furthermore, the storage medium of the embodiments of the present application stores program instructions capable of implementing all the above methods. Among them, the program instructions can be stored in the above storage medium in the form of a software product, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: various media that can store program codes such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs, or terminal devices such as computers, servers, mobile phones, and tablets.

[0140] The above description is only some preferred embodiments of the present invention and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A method for safety monitoring of local and cloud intelligent interconnected energy storage equipment, characterized in that: Collect energy storage device data using a transmission protocol support module that supports a specific variety of underlying communication protocols; Analyze and clean the collected data in real time; Perform feature extraction and normalization on the cleaned data; The processed data is input into the local real-time diagnosis algorithm module and the cloud prediction algorithm module for safety monitoring; Execute corresponding alarms and processing measures based on the results of local and cloud algorithms, and save processing records locally and upload processing records to the cloud; Realize data interaction between local EMS and cloud platform, including data sending, control instruction issuing, execution control and execution result feedback.

2. The method according to claim 1, characterized in that During the data collection process, when it is detected that the device is powered on, the communication interface is connected, and the energy storage device transmits data to the server, the original data transmitted by the device is received through the Transport function module that supports three transmission protocols: TCP, RUT, and QUIC. Then, the data protocol conversion is completed through the Code module. Among them, the data sent by the device is decoded, and the original data is decoded into a format that can be processed by the system or easy for developers to process. The encoding supports the instructions and data sent to the device.

3. The method according to claim 1, characterized in that The real-time cleaning adopts the sliding window mode to process the entire time series. For random data missing, the missing part is filled by linear regression estimation method. For sudden outlier data, the sliding window size is determined according to the data characteristics and the required smoothness. The data points in the window are summed and the sum is divided by the window size to obtain the average value instead. The window is moved one data point at a time until the entire time series is processed.

4. The method according to claim 3, characterized in that The formula for judging outliers is Among them, x i is the sample data, i=1,2,…,N, N is the total number of samples, is the sample mean, determine the detection level α, and check the critical value table value b′ of the kurtosis test p (n), when b k >b′ p (n), determine the deviation from the mean The farthest value is the outlier. After removing the outlier, the repeated kurtosis test method is used to test whether there are still outliers. If not, no outliers are found.

5. The method according to claim 1, characterized in that Feature extraction obtains data features from multiple angles, including time domain features, frequency domain features, and time series features, and obtains the high-order feature set Q1, ensuring that the single feature Gini coefficient R is not less than 0.02, and obtains the low-dimensional feature set Q2, and then obtains the fused feature set M.

6. The method according to claim 5, characterized in that The time domain features include: average value: where x i for Sample data: i = 1, 2, ..., N, N is the total number of samples; Peak-to-peak value: t2 = max(x i )-min(x i ); RMS: Impact factor: t4 = max(x i ) / t3; Form Factor: Skewness Index: Standard Deviation: Kurtosis index: Crest factor: t9 = max(x i ) / t3; Margin Factor:

7. The method according to claim 5, characterized in that The frequency domain feature converts the time domain waveform of the signal into a spectrum signal by means of discrete Fourier transform, wherein: n=1,2,…,N-1, x(kΔt) represents the acquisition value of the signal, N is the number of sampling points, Δt is the sampling interval, k is the sequence number of the time domain discrete value, and the frequency domain features include: Center of gravity frequency: Average frequency: Frequency variance: RMS frequency:

8. The method according to claim 5, characterized in that The temporal feature extraction comprises the following steps: Through the AR(p) model, establish the data relationship and set x i+1 =φ1x i +φ2x i-1 +…+φ p x i-p+1 +ξ i+1 , where φ j , j = 1, 2, ..., p is the linear correlation coefficient, ξ i+1 for noise; Define the partial correlation coefficient of the sequence under the AR(p) model as: pcaf(p) = φ p ; Constructing the Yule-Walker equation, constructing the equation Multiply both sides of the equation by x i-k+1 , find the expectation of the result, remove the noise term, and then divide the equation by Nk, while letting c l =c -l , thus obtaining the sequence autocovariance c k , then divide both sides of the equation by c0 to get the autocorrelation coefficient of the sequence, and rearrange it to get the matrix equation Rφ=r. Since the R matrix is ​​a symmetric full-rank matrix and reversible, we can calculate Then the partial correlation coefficient of lag number k is The autocorrelation coefficient and partial correlation coefficient of each attribute are extracted to form the time series features.

9. The method according to claim 1, characterized in that: The normalization adopts Sigmoid function Mapping variables to [0,1], derivative of the Sigmoid function 10. The method according to claim 1, characterized in that The local algorithm units of the local real-time diagnosis algorithm module include voltage anomaly detection algorithm, battery cell connection fault detection algorithm, neural network-based energy storage temperature longitudinal detection algorithm, and energy storage battery abnormal temperature rise rate detection algorithm; the cloud algorithm units of the cloud prediction algorithm module include single battery cell SOH estimation, single battery cell voltage prediction, and internal short circuit prediction.

Images