Energy storage safety early warning method and system based on deep learning

Abstract

The present application discloses a deep learning-based energy storage safety early warning method and system, which relates to the field of electrical safety monitoring technology, including: inputting historical operating status data, current operating status data, historical environmental characteristic data and current environmental characteristic data into a data prediction network, performing operating status data prediction and environmental characteristic data prediction based on the data prediction network, and generating operating status data prediction quantity and environmental characteristic data prediction quantity; determining the safety state deviation of the energy storage device at the current moment according to the operating status data prediction quantity and the environmental characteristic data prediction quantity, and determining the pyrolysis particle target index at the current moment according to the current environmental characteristic data and the safety state deviation; determining the early warning method of the energy storage device at the current moment according to the pyrolysis particle target index, so as to solve the technical problem of low accuracy of early warning monitoring for energy storage devices in the prior art.

Description

Technical Field

[0001] The present application relates to the field of electrical safety monitoring technology, and in particular to a storage energy safety early warning method and system based on deep learning. Background Art

[0002] Traditional fire monitoring technologies rely primarily on temperature sensors, smoke detectors, or infrared monitoring, and suffer from issues such as delayed response, susceptibility to interference, and an inability to provide early warnings. With the increasing application of new technologies such as energy storage devices, the safety monitoring of energy storage devices is receiving increasing attention. During the operation of energy storage devices, traditional safety monitoring methods rely primarily on fixed thresholds and single-dimensional data analysis, making it difficult to comprehensively and accurately assess the safety status of the equipment. This leads to delayed warnings or high false alarm rates, and the inability to promptly identify potential safety hazards. Especially in complex and changing actual operating environments, the safety risks of energy storage devices become even more difficult to predict and control. Therefore, there is currently a lack of a method that can comprehensively process multi-dimensional data and provide accurate early warnings for energy storage devices.

[0003] The above content is only used to assist in understanding the technical solution of this application and does not constitute an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide a deep learning-based energy storage safety early warning method and system, aiming to solve the technical problem of low accuracy of early warning monitoring for energy storage equipment in the existing technology.

[0005] To achieve the above objectives, this application proposes a deep learning-based energy storage safety early warning method, which includes:

[0006] Obtaining historical operating status data and historical environmental characteristic data of the energy storage device at historical moments, and obtaining current operating status data and current environmental characteristic data of the energy storage device at the current moment;

[0007] Inputting the historical operating state data, the historical environmental characteristic data, the current operating state data, and the current environmental characteristic data into a data prediction network to obtain a predicted amount of operating state data and a predicted amount of environmental characteristic data of the energy storage device at a future time generated by the data prediction network; the data prediction network is obtained by training a long short-term memory network (LSTM) based on a training sample set;

[0008] Determining a safety state deviation of the energy storage device at the current moment based on the predicted amount of operating state data and the predicted amount of environmental characteristic data;

[0009] determining a pyrolysis particle target index at the current moment according to the current environmental characteristic data and the safety state deviation;

[0010] An early warning mode of the energy storage device at the current moment is determined according to the pyrolysis particle target index.

[0011] In some embodiments, determining the safety state deviation of the energy storage device at the current moment based on the predicted amount of operating state data and the predicted amount of environmental characteristic data includes:

[0012] Obtaining a preset operating status data reference range and a preset environmental characteristic data reference range corresponding to the energy storage device;

[0013] Calculating the difference between the predicted running state data and the preset running state data reference range to obtain the running state data deviation;

[0014] Calculating the difference between the predicted amount of environmental characteristic data and the preset reference range of environmental characteristic data to obtain an environmental characteristic data deviation;

[0015] The safety state deviation is determined according to the operating state data deviation and the environmental characteristic data deviation.

[0016] In some embodiments, determining the pyrolysis particle target index at the current moment based on the current environmental characteristic data and the safety state deviation includes:

[0017] Determining the initial index of pyrolysis particles at the current moment according to the current environmental characteristic data;

[0018] The pyrolysis particle initial index is corrected based on the safety state deviation to obtain the pyrolysis particle target index at the current moment.

[0019] In some embodiments, determining the initial index of pyrolysis particles at the current moment according to the current environmental characteristic data includes:

[0020] Determine the current time-sharing stage category at the current moment according to the current ambient temperature change rate;

[0021] Determine a target correspondence table corresponding to the current time-sharing stage category according to a preset first correspondence between the time-sharing stage category and the correspondence table, wherein the target correspondence table includes correspondences between various parameters and weighting coefficients;

[0022] Determining the weighting coefficients of various parameters in the current environmental feature data according to the target correspondence table;

[0023] The pyrolysis particle initial index is determined based on the weighted coefficients of various parameters in the current environmental characteristic data and the current environmental characteristic data, wherein the current environmental data includes the current environmental temperature change rate.

[0024] In some embodiments, the correcting the initial pyrolysis particle index based on the safety state deviation to obtain the target pyrolysis particle index at the current moment includes:

[0025] If the safety state deviation is less than a first deviation threshold, determining the pyrolysis particle initial index as a pyrolysis particle target index;

[0026] If the safety state deviation is greater than or equal to the first deviation threshold and less than a second deviation threshold, performing a weighted summation on the safety state deviation and the pyrolysis particle initial index to obtain a first summation result, and determining the first summation result as the pyrolysis particle target index, and the second deviation threshold is greater than the first deviation threshold;

[0027] If the safety state deviation is greater than or equal to the second deviation threshold, the safety state deviation and the pyrolysis particle initial index are weightedly summed to obtain a second summation result, and the second summation result is compared with the value of the preset pyrolysis particle index, and the larger value is determined as the pyrolysis particle target index.

[0028] In some embodiments, determining the warning mode of the energy storage device at the current moment according to the pyrolysis particle target index includes:

[0029] Obtain a first warning threshold and a second warning threshold, wherein the second warning threshold is greater than the first warning threshold;

[0030] Acquire an energy storage device image, wherein the energy storage device image includes the energy storage device;

[0031] When the pyrolysis particle target index is greater than or equal to a first warning threshold and less than a second warning threshold, determining that the warning mode is a first warning mode, and controlling the display of a first warning message on the user terminal, the first warning message including the pyrolysis particle target index and the energy storage device image;

[0032] When the pyrolysis particle target index is greater than or equal to the second warning threshold, the warning mode is determined to be the second warning mode, the second warning information is controlled to be displayed on the user terminal, the user terminal is controlled to emit a warning sound for prompting the second warning mode, and the energy storage device is controlled to emit a warning sound for prompting the second warning mode, and the second warning information includes the pyrolysis particle target index and the energy storage device image.

[0033] In some embodiments, after determining that the warning mode is the first warning mode, the method further includes:

[0034] generating a first control instruction, and controlling the ventilation device connected to the energy storage device to turn on based on the first control instruction; and

[0035] A second control instruction is generated, and based on the second control instruction, a load of the energy storage device is controlled to migrate from the energy storage device to a backup line.

[0036] In some embodiments, the operating state data includes voltage and current, the environmental characteristic data includes ambient temperature, target volatile gas concentration, and target smoke particle concentration, and inputting the historical operating state data, the historical environmental characteristic data, the current operating state data, and the current environmental characteristic data into a data prediction network includes:

[0037] Using a Kalman filter algorithm to smooth the historical operating state data, the historical environmental characteristic data, the current operating state data, and the current environmental characteristic data to obtain a filtered data sequence;

[0038] Normalizing various parameters in the filtered data sequence to obtain a normalized data sequence, wherein the normalized data sequence includes a voltage data normalized sequence, a current data normalized sequence, an ambient temperature data normalized sequence, a target volatile gas concentration data normalized sequence, and a target smoke particle concentration data normalized sequence;

[0039] The normalized data sequence is determined as an input sequence, and the input sequence is input into the data prediction network.

[0040] In some embodiments, the historical environmental characteristic data and the current environmental characteristic data are obtained by the following steps:

[0041] Acquiring material parameters of the energy storage device, and determining a target volatile gas concentration and a target smoke particle size according to the material parameters of the energy storage device;

[0042] Acquire historical ambient temperatures before the current moment and the current ambient temperature at the current moment, collected by a temperature sensor, and determine a current ambient temperature change rate based on the historical ambient temperatures and the current ambient temperature;

[0043] Obtaining a historical target volatile gas concentration before the current moment and a current target volatile gas concentration at the current moment, collected by an electrochemical sensor, and determining a current target volatile gas concentration change rate based on the historical target volatile gas concentration and the current target volatile gas concentration;

[0044] Obtaining a historical target smoke particle concentration before the current moment and a current target smoke particle concentration at the current moment, collected by a laser particle sensor, and determining a current target smoke particle concentration change rate based on the historical target smoke particle concentration and the current target smoke particle concentration;

[0045] Determining the historical ambient temperature, the historical target volatile gas concentration, and the historical target smoke particle concentration as the historical environmental characteristic data;

[0046] The current ambient temperature, the current ambient temperature change rate, the current target volatile gas concentration, the current target volatile gas concentration change rate, the current target smoke particle concentration change rate, and the current target smoke particle concentration are determined as the current environmental characteristic data.

[0047] In some embodiments, obtaining the first warning threshold and the second warning threshold includes:

[0048] Obtaining a first preset warning initial value;

[0049] determining a current first correction amount corresponding to the current ambient temperature according to a second mapping relationship, wherein the second mapping relationship is used to represent a mapping relationship between the ambient temperature and the first correction amount, and the ambient temperature and the first correction amount are positively correlated;

[0050] Correcting the first preset warning initial value according to the current first correction amount to obtain the first warning threshold;

[0051] Obtaining a second preset warning initial value;

[0052] determining a current second correction amount corresponding to the current ambient temperature according to the current ambient temperature and a third mapping relationship, wherein the third mapping relationship is used to represent a mapping relationship between the ambient temperature and the second correction amount, and the ambient temperature and the second correction amount are positively correlated;

[0053] The second preset warning initial value is corrected according to the current second correction amount to obtain the second warning threshold.

[0054] In addition, to achieve the above objectives, this application also proposes a deep learning-based energy storage safety early warning system, which includes:

[0055] A multi-source data acquisition module is used to obtain historical operating status data and historical environmental characteristic data of the energy storage device at historical moments, as well as current operating status data and current environmental characteristic data of the energy storage device at the current moment;

[0056] a prediction module, configured to input the historical operating status data, the historical environmental characteristic data, the current operating status data, and the current environmental characteristic data into a data prediction network, and obtain a predicted amount of operating status data and a predicted amount of environmental characteristic data of the energy storage device at a future time generated by the data prediction network; the data prediction network is obtained by training a long short-term memory network (LSTM) based on a training sample set;

[0057] a deviation determination module, configured to determine a safety state deviation of the energy storage device at the current moment based on the predicted amount of the operating state data and the predicted amount of the environmental characteristic data;

[0058] An index determination module, configured to determine a target index of pyrolysis particles at the current moment based on current environmental characteristic data and the safety state deviation;

[0059] The early warning determination module is used to determine the early warning mode of the energy storage device at the current moment according to the pyrolysis particle target index.

[0060] In addition, to achieve the above-mentioned purpose, the present application also proposes a deep learning-based energy storage safety warning system, which includes: a memory, a processor, and a deep learning-based energy storage safety warning program stored in the memory and executable on the processor, wherein the deep learning-based energy storage safety warning program is configured to implement the steps of the deep learning-based energy storage safety warning method described above.

[0061] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the deep learning-based energy storage safety warning method as described above are implemented.

[0062] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the energy storage safety early warning method based on deep learning as described above.

[0063] One or more technical solutions proposed in this application have at least the following technical effects: by obtaining historical operating status data of an energy storage device at historical moments and current operating status data at the current moment, as well as obtaining historical environmental characteristic data at historical moments and current environmental characteristic data at the current moment; inputting the historical operating status data, current operating status data, historical environmental characteristic data, and current environmental characteristic data into a data prediction network, and performing operating status data prediction and environmental characteristic data prediction based on the data prediction network to generate predicted operating status data and environmental characteristic data for the energy storage device at future moments. The data prediction network is a Long Short-Term Memory Network (LSTM) trained based on a training sample set, which can capture long-term dependencies in time series and predict future data. The safety state deviation of the energy storage device at the current moment is determined based on the predicted operating status data and predicted environmental characteristic data, and the degree of deviation of the current device from the normal state is evaluated. The pyrolysis particle target index at the current moment is determined based on the current environmental characteristic data and the safety state deviation. The pyrolysis particle target index reflects the current safety risk level of the energy storage device. Different index ranges correspond to different warning methods. Based on the pyrolysis particle target index, the warning method for the energy storage device at the current moment is determined, and corresponding response measures are taken. The deep learning-based energy storage safety warning method provided in this application integrates multi-source heterogeneous data (operation status data and environmental feature data), uses a trained LSTM network (data prediction network) for prediction and analysis, combines historical data with current data for trend prediction, predicts state changes and environmental changes of the energy storage device in advance, and calculates the safety state deviation of the energy storage device at the current moment based on the prediction results (predicted amount of environmental feature data and predicted amount of operation status data). Combined with the current environmental feature data and the safety state deviation, the warning method of the energy storage device at the current moment is determined, achieving efficient early warning and accurate safety status assessment of the energy storage device, solving the technical problem of low accuracy of warning monitoring for energy storage devices in the existing technology. It not only improves the accuracy and timeliness of energy storage safety warnings, but also enhances adaptability to the characteristics of energy storage devices, and better copes with the complex working conditions of energy storage devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0065] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0066] Figure 1 A flowchart illustrating the first embodiment of the energy storage safety early warning method based on deep learning in this application;

[0067] Figure 2 This is a schematic diagram of the module structure of the energy storage safety early warning system based on deep learning in an embodiment of the present application;

[0068] Figure 3 This is a system structure diagram of the hardware operating environment involved in the deep learning-based energy storage safety warning method in the embodiment of the present application. DETAILED DESCRIPTION

[0069] It should be understood that the specific embodiments described herein are merely used to explain the technical solutions of the present application and are not intended to limit the present application.

[0070] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0071] The main solution of the embodiment of the present application is: obtaining historical operating status data and historical environmental characteristic data of the energy storage device at historical moments, and obtaining current operating status data and current environmental characteristic data of the energy storage device at the current moment; inputting the historical operating status data, historical environmental characteristic data, current operating status data and current environmental characteristic data into a data prediction network, and obtaining the operating status data prediction amount and environmental characteristic data prediction amount of the energy storage device at the future moment generated by the data prediction network; the data prediction network is obtained by training the long short-term memory network LSTM based on the training sample set; determining the safety state deviation of the energy storage device at the current moment based on the operating status data prediction amount and the environmental characteristic data prediction amount; determining the pyrolysis particle target index at the current moment based on the current environmental characteristic data and the safety state deviation; and determining the early warning method of the energy storage device at the current moment based on the pyrolysis particle target index.

[0072] In this embodiment, for ease of description, the following description is based on the identification of a deep learning-based energy storage safety warning system as the execution subject.

[0073] Traditional electrical fire monitoring technologies mainly rely on temperature sensors, smoke detectors, or infrared monitoring. Although these methods can provide fire alarms to a certain extent, they generally have problems such as delayed response and susceptibility to environmental interference. In recent years, the application of pyrolytic particle detectors has gradually increased. Although pyrolytic particle detectors can detect micron-sized particles and gases produced by the pyrolysis of insulating materials and can provide more precise monitoring methods compared to temperature sensors, they lack the ability to intelligently integrate and analyze multi-source heterogeneous data, making it difficult to adapt to complex and changing practical application scenarios. In addition, existing technologies lack targeted solutions for safety warnings in new application scenarios such as energy storage equipment. Energy storage equipment has unique operating characteristics, including high energy density and rapid charging and discharging, which makes its safety monitoring requirements higher. Traditional safety monitoring methods can no longer meet the needs of energy storage equipment under complex working conditions. Effective and accurate safety monitoring cannot be carried out through existing single sensors or simple combinations.

[0074] In response to the above problems, this application proposes a deep learning-based energy storage safety early warning method. The deep learning-based energy storage safety early warning method integrates multi-source heterogeneous data (operating status data and environmental feature data), uses a trained LSTM network (data prediction network) for prediction and analysis, combines historical data and current data for trend prediction, predicts the state changes and environmental changes of the energy storage equipment in advance, and calculates the safety state deviation of the energy storage equipment at the current moment based on the prediction results (predicted amount of environmental feature data and predicted amount of operating status data). Combined with the current environmental feature data and the safety state deviation, the early warning method of the energy storage equipment at the current moment is determined, achieving efficient early warning and accurate safety status assessment of the energy storage equipment, solving the technical problem of low accuracy of early warning monitoring of energy storage equipment in the existing technology, not only improving the accuracy and timeliness of energy storage safety early warning, but also enhancing the adaptability to the characteristics of energy storage equipment, and better coping with the complex working conditions of energy storage equipment.

[0075] It should be noted that the execution entity of this embodiment can be a computing service device with data processing, network communication, and program execution capabilities, or an energy storage device capable of implementing the above functions. This embodiment and the following embodiments are described below using a deep learning-based energy storage safety warning system as an example.

[0076] Based on this, the embodiment of the present application provides a storage energy safety early warning method based on deep learning, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the energy storage safety warning method based on deep learning in this application.

[0077] In this embodiment, the energy storage safety early warning method based on deep learning includes steps 101 to 105:

[0078] Step 101 : Acquire historical operating status data and historical environmental characteristic data of the energy storage device at historical moments, and acquire current operating status data and current environmental characteristic data of the energy storage device at the current moment.

[0079] Specifically, the first preset number of historical operating status data for the energy storage device before the current moment and the current operating status data at the current moment are obtained, as well as the first preset number of historical environmental characteristic data before the current moment and the current environmental characteristic data at the current moment. An energy storage device is a device used to store electrical energy, such as an energy storage battery cabinet or energy storage battery cabinet. The operating status and safety of the energy storage device directly impact the reliability and stability of the entire system. The current moment is the time point for real-time monitoring of the system and serves as the reference time for data collection and analysis. The first preset number is a fixed number, representing the first preset number of sampling points within a pre-set time window (e.g., 10 minutes), for example, 1000 sampling points collected over the past 10 minutes. This application does not impose a specific numerical limit on the first preset number; in this application, the value of the first preset number is greater than or equal to 2. The historical operating status data represents the operating parameters of the energy storage device over a period of time, such as current and voltage. The first preset number of historical operating status data can reflect the historical operating trends of the energy storage device. The current operating status data is the real-time operating parameters of the energy storage device at the current moment, such as the current temperature, current, and voltage. The current operating status data can reflect the immediate status of the device. The environmental characteristic data is the physical quantity parameters that indicate fire hazards in the environment where the energy storage device is located, such as ambient temperature, target volatile gas concentration, and target smoke particle concentration. The first preset historical environmental characteristic data is the physical quantity parameters that indicate fire hazards in the environment where the energy storage device has been located over a period of time in the past. It can reflect the changing trend of the environment where the energy storage device has been located over the past period of time. The current environmental characteristic data is the physical quantity parameters that indicate fire hazards in the environment where the energy storage device is located at the current moment, which can reflect the immediate status of the current environment.

[0080] In some embodiments, multi-source sensors (such as voltage sensors and current sensors) can be used to collect operating status data of the energy storage device, sampling at a fixed frequency (e.g., 100 times per second) and storing it in a database. Simultaneously, environmental sensors (such as electrochemical sensors, temperature sensors, and laser particle sensors) can be used to collect environmental characteristic data of the energy storage device's environment, sampling at a fixed frequency and storing it synchronously with the operating status data. A first preset amount of historical operating status data (e.g., operating status data for the past 10 minutes) can be extracted from the database. This first preset amount of historical operating status data can be represented as a time series, and the current operating status data can be collected as real-time input. A first preset amount of historical environmental characteristic data (e.g., temperature, gas concentration, particle concentration, etc., data for the past 10 minutes) can be extracted from the database, and the current environmental characteristic data can be collected to supplement the real-time information. The above historical and current operating status data and historical and current environmental characteristic data are integrated to generate a multi-dimensional time series dataset, which serves as input for a subsequent deep learning model (e.g., a data prediction network). By combining historical and current operating status data, the system comprehensively reflects the changing trends of the device's status. Historical operating status data provides contextual information, helping to better understand whether the current status reflects long-term trends. By integrating multiple sensors (such as electrochemical sensors, temperature sensors, and laser particle sensors), the system monitors multiple environmental parameters (environmental characteristic data) of the energy storage device's environment in real time. This environmental characteristic data can include ambient temperature, target volatile gas concentration, and target smoke particle concentration, comprehensively covering physical quantities that may indicate fire hazards. Combining environmental characteristic data with operating status data allows for a more comprehensive assessment of the device's safety status, avoiding misjudgments caused by single data points and improving the accuracy and relevance of early warnings.

[0081] By acquiring the historical and current operating status data and environmental characteristic data of energy storage equipment, a comprehensive, multi-dimensional data foundation can be constructed. The fusion analysis of multi-dimensional data can help to more accurately identify potential safety hazards, solve the problems of delayed warnings, high false alarm rates, and inability to integrate multi-dimensional data in traditional technologies, and enhance the early hazard prediction capabilities and intelligence level of energy storage safety monitoring systems.

[0082] Step 102: Input historical operating status data, historical environmental characteristic data, current operating status data, and current environmental characteristic data into a data prediction network to obtain predicted operating status data and predicted environmental characteristic data of the energy storage device at a future time, generated by the data prediction network. The data prediction network is obtained by training a long short-term memory network (LSTM) based on a training sample set.

[0083] Specifically, first preset historical operating status data, current operating status data, first preset historical environmental characteristic data, and current environmental characteristic data are input into a data prediction network. Based on the data prediction network, operating status data prediction and environmental characteristic data prediction are performed to generate a second preset amount of predicted operating status data for the energy storage device after the current moment and a second preset amount of predicted environmental characteristic data after the current moment. The data prediction network is obtained by training a long short-term memory (LSTM) network based on a training sample set. The data prediction network is obtained by training a deep learning model (such as an LSTM) based on a training sample set. The data prediction network analyzes and predicts the first preset amount of historical operating status data, current operating status data, first preset historical environmental characteristic data, and current environmental characteristic data to generate a predicted amount of operating status data and a predicted amount of environmental characteristic data at a certain point in the future. The second preset amount of predicted operating status data is the energy storage device operating status data for a future period (e.g., the next 30 minutes or the next 1000 sampling points) predicted via a data prediction network based on historical and current operating status data. The second preset amount of predicted environmental characteristic data is the environmental characteristic data for a future period (e.g., the next 30 minutes or the next 1000 sampling points) predicted via a data prediction network based on historical and current environmental characteristic data. This application does not impose a specific numerical limit on the second preset amount. In this application, the value of the first preset amount is greater than or equal to two.

[0084] In some embodiments, the data prediction network training process can be as follows: multiple data sets are obtained, each including operating status data and environmental characteristic data. The data sets are randomly divided into training and testing sample sets using ten-fold cross-validation. The training sample set includes multiple training samples, each of which includes multiple historical operating status data of the energy storage device at a historical moment, multiple historical environmental characteristic data at a historical moment, multiple predicted operating status data quantities of the energy storage device after the historical moment, and multiple predicted environmental characteristic data quantities. An LSTM is trained using the training sample sets. The LSTM inputs are the multiple historical operating status data and multiple historical environmental characteristic data, and the output is multiple predicted operating status data quantities and multiple predicted environmental characteristic data quantities for the relevant inputs. The optimizer selected for LSTM training can be Adam. The model performance is evaluated using root mean square error (RMS), optimizing the deviation between the predicted value and the true value. The trained LSTM (i.e., the data prediction network) can use the multiple historical operating status data and multiple historical environmental characteristic data to predict the predicted operating status data quantities of the energy storage device at future moments and the predicted environmental characteristic data quantities of the energy storage device's environment. Through its internal memory cells and gating mechanisms (input gate, forget gate, and output gate), LSTMs can effectively retain and forget information, thereby capturing long-term dependencies in the input sequence. Even with long time series, they can retain important pieces of information. Consequently, data prediction networks trained on LSTMs based on training sample sets also exhibit this property. By learning long-term patterns within the training sample set, data prediction networks can process long-term dependencies in time series data, enabling more accurate predictions of future operating status and environmental characteristics. Furthermore, the data prediction network boasts high computational efficiency for prediction and control, meeting real-time requirements. The data prediction network algorithm can be processed in parallel on hardware accelerators such as GPUs, improving real-time performance. Applying LSTM to multi-parameter time series correlation analysis for energy storage systems helps address the low accuracy of traditional energy storage safety warning methods.

[0085] In some embodiments, first preset historical operating status data, current operating status data, first preset historical environmental characteristic data, and current environmental characteristic data are input into a data prediction network. Based on the trained data prediction network, the first preset historical operating status data, current operating status data, first preset historical environmental characteristic data, and current environmental characteristic data are processed using learned rules to predict the future operating status data of the energy storage device and the future environmental characteristic data of the environment in which the energy storage device is located. By using multiple historical operating status data, current operating status data, multiple historical environmental characteristic data, and current environmental characteristic data, the data prediction network can capture the trends and patterns of the evolution of the operating status data and environmental characteristic data of the energy storage device over time, and obtain accurate prediction results, namely, the predicted amount of the second preset operating status data of the energy storage device after the current moment and the predicted amount of the second preset environmental characteristic data after the current moment.

[0086] By leveraging LSTM to comprehensively analyze multi-source heterogeneous data, extracting complex nonlinear relationships, and implementing intelligent fusion analysis of multi-dimensional data, we can predict future operating states and environmental characteristics. This allows for more accurate identification of abnormal trends and reduces false alarms caused by fluctuations in single data points. Furthermore, the deep learning-based energy storage safety early warning system can issue warnings at the earliest stages of potential hazards, thereby improving response speed and early warning capabilities. Furthermore, the trained data prediction network can better adapt to the complex and ever-changing application scenarios of energy storage equipment, helping to provide more accurate safety monitoring and early warnings.

[0087] Step 103: Determine the safety state deviation of the energy storage device at the current moment based on the predicted amount of the operating state data and the predicted amount of the environmental characteristic data.

[0088] Specifically, the safety state deviation of the energy storage device at the current moment is determined based on the second preset predicted amount of operating state data and the second preset predicted amount of environmental characteristic data. The safety state deviation is the degree of deviation between the predicted amount of operating state data and the predicted amount of environmental characteristic data of the energy storage device and the safe operating state data and the environmental characteristic data in the safe state. It can be calculated by comparing the predicted value with the safety state baseline value. The calculated safety state deviation of the energy storage device at the current moment can quantify the degree of deviation of the current state of the energy storage device from the safe and normal baseline, reflecting the cumulative effect of potential risks.

[0089] In some embodiments, a second predetermined predicted amount of operating state data and a second predetermined predicted amount of environmental characteristic data, obtained through a data prediction network, are compared with a safety state baseline value, and the deviation of each parameter is calculated to comprehensively determine the safety state deviation of the energy storage device at the current moment. The safety state baseline value can be established based on historical normal data, for example, a normal range for temperature, gas concentration, etc.

[0090] Step 104 : Determine the pyrolysis particle target index at the current moment based on the current environmental characteristic data and the safety state deviation.

[0091] Specifically, the Pyrolysis Particle Target Index (PTI) is a dynamic risk score that combines safety deviations with current environmental characteristics. It reflects the degree of pyrolysis of the insulating material and quantifies the current safety risk level of the energy storage device, enabling tiered early warning decisions. The higher the Pyrolysis Particle Target Index, the higher the current risk and the greater the probability of an electrical fire, corresponding to a higher level of subsequent early warning.

[0092] In some embodiments, a specific algorithm (such as a weighted average or machine learning model) can be used to determine a target pyrolysis particle index (PPI) based on current environmental characteristic data and safety state deviation. This index reflects the current safety risk level of the energy storage device and the degree of pyrolysis of the insulating material. By calculating the PPI PPI and combining it with current environmental characteristic data, a comprehensive assessment of the safety state of the energy storage device can be achieved. This helps improve the accuracy and timeliness of early warnings and enhances adaptability to complex application scenarios, thereby addressing the technical issues of existing methods such as delayed warnings, high false alarm rates, and the inability to predict early hidden dangers.

[0093] Step 105 : Determine the warning mode of the energy storage device at the current moment according to the pyrolysis particle target index.

[0094] Specifically, the Pyrolysis Particle Target Index is a dynamic risk score generated by fusing safety state deviations with real-time environmental data (current environmental characteristic data). Its value range is 0-4000, reflecting the degree of thermal decomposition of insulating materials and quantifying the current safety risk level of energy storage equipment. Higher values ​​indicate more urgent risks. Early warning methods determine corresponding warning levels and response measures based on the different value ranges of the Pyrolysis Particle Target Index. Warning methods can include level one and level two warnings, with each warning level corresponding to different response measures. The Pyrolysis Particle Target Index is used to determine the current warning method for the energy storage equipment. Alert methods of varying urgency are matched to the Pyrolysis Particle Target Index value to avoid overreaction.

[0095] The deep learning-based energy storage safety early warning method provided in this application obtains historical operating status data of the energy storage device at historical moments and current operating status data at the current moment, as well as historical environmental feature data at historical moments and current environmental feature data at the current moment; inputs the historical operating status data, current operating status data, historical environmental feature data, and current environmental feature data into a data prediction network, and performs operating status data prediction and environmental feature data prediction based on the data prediction network to generate predicted operating status data and environmental feature data for the energy storage device at future moments. The data prediction network is obtained by training an LSTM based on a training sample set, which can capture long-term dependencies in time series and predict future data. The safety state deviation of the energy storage device at the current moment is determined based on the predicted operating status data and the predicted environmental feature data, and the degree of deviation of the current device from the normal state is evaluated. The pyrolysis particle target index at the current moment is determined based on the current environmental feature data and the safety state deviation. The pyrolysis particle target index reflects the current safety risk level of the energy storage device. Different index ranges correspond to different early warning methods. Based on the pyrolysis particle target index, the early warning method of the energy storage device at the current moment is determined, and corresponding response measures are taken. The deep learning-based energy storage safety early warning method provided in this application integrates multi-source heterogeneous data (operating status data and environmental feature data), uses a trained LSTM (data prediction network) for prediction and analysis, combines historical data with current data for trend prediction, predicts state changes and environmental changes of energy storage equipment in advance, and calculates the safety state deviation of the energy storage equipment at the current moment based on the prediction results (predicted amount of environmental feature data and predicted amount of operating status data). Combined with the current environmental feature data and the safety state deviation, the early warning method of the energy storage equipment at the current moment is determined, achieving efficient early warning and accurate safety status assessment of the energy storage equipment, solving the technical problem of low accuracy of early warning monitoring for energy storage equipment in the existing technology, not only improving the accuracy and timeliness of energy storage safety early warnings, but also enhancing adaptability to the characteristics of energy storage equipment, and better coping with the complex working conditions of energy storage equipment.

[0096] In some embodiments, determining the safety state deviation of the energy storage device at the current moment based on the predicted amount of operating state data and the predicted amount of environmental characteristic data includes:

[0097] Obtaining a preset operating status data reference range and a preset environmental characteristic data reference range corresponding to the energy storage device;

[0098] Calculate the difference between the predicted amount of running status data and the preset reference range of running status data to obtain the running status data deviation;

[0099] Calculate the difference between the predicted amount of environmental characteristic data and the preset reference range of environmental characteristic data to obtain the environmental characteristic data deviation;

[0100] Determine the safety status deviation based on the operating status data deviation and the environmental characteristic data deviation.

[0101] Specifically, the number of predicted operating status data can be a second preset number. The second preset number of predicted operating status data is the predicted operating status data of the energy storage device for a future period (e.g., the next 10 minutes), generated by the data prediction network, including voltage, current, and other data. The number of predicted environmental characteristic data can be a second preset number. The second preset number of predicted environmental characteristic data is the predicted environmental characteristic data for a future period, generated by the data prediction network, including ambient temperature, target volatile gas concentration, target smoke particle concentration, and other data. The preset reference range for operating status data can be the reasonable range of various operating parameters under normal operating conditions, based on the design specifications of the energy storage device and historical data statistics. For example, the normal voltage range is [48V, 52V]. The preset reference range for environmental characteristic data can be the normal environmental parameter range based on the safety requirements of the energy storage device's environment and historical data statistics. For example, the normal ambient temperature range is [20°C, 30°C].

[0102] As an example, the difference between the predicted running status data and the preset running status data reference range can be calculated using the following formula to obtain the running status data deviation: .

[0103] in is the deviation of the i-th operating status data, is the predicted value of the i-th operating state, It is the median value of the reference range of the corresponding preset operating status data.

[0104] The difference between the predicted environmental characteristic data and the preset environmental characteristic data reference range can be calculated using the following formula to obtain the environmental characteristic data deviation: .

[0105] in, is the j-th environmental characteristic data deviation, is the predicted value of the jth environmental characteristic, is the lower limit of the corresponding preset environmental characteristic data reference range, It is the upper limit of the reference range of the corresponding preset environmental characteristic data.

[0106] The safety state deviation at a future moment can be determined by the following formula based on the j-th environmental characteristic data deviation and the i-th operating state data deviation: .

[0107] in, is the safety state deviation, is the dynamic weight of the running status data deviation, is the dynamic weight of the environmental characteristic data deviation, n is the number of items of operating status data, and m is the number of items of environmental characteristic data. According to the above method, the safety state deviation at the second preset future moment is calculated, and the weighted sum of the safety state deviations at the second preset future moment is performed to obtain the final safety state deviation of the energy storage device at the current moment. The closer the safety state deviation is to 0, the closer the energy storage device is to the normal state, and the larger the safety state deviation is, the higher the safety risk of the energy storage device is. By calculating the deviation of the predicted values ​​of the operating status data of the energy storage device and the environmental characteristic data, a comprehensive safety state deviation index, namely the safety state deviation, is generated to comprehensively evaluate the overall safety status of the energy storage device, avoid misjudgment caused by single-dimensional analysis, quantify the degree of deviation and integrate multi-dimensional data, which helps to more accurately identify potential safety hazards.

[0108] In some embodiments, determining a pyrolysis particle target index at a current moment based on current environmental characteristic data and a safety state deviation degree includes:

[0109] Determine the initial index of pyrolysis particles at the current moment according to the current environmental characteristic data;

[0110] The initial index of the pyrolysis particles is corrected based on the safety state deviation to obtain the target index of the pyrolysis particles at the current moment.

[0111] Specifically, the Pyrolytic Particle Initial Index (PPI) is a preliminary calculation based on current environmental characteristic data. It quantifies the current Pyrolytic Particle concentration within the energy storage device and reflects the current safety risk level. The Pyrolytic Particle Target Index (PTI) is a final index calculated by comprehensively considering the Pyrolytic Particle Initial Index and the Safety Status Deviation, providing a more accurate reflection of the current safety risk level of the energy storage device.

[0112] As an example, current environmental characteristic data is obtained. This data may include the current ambient temperature, the current ambient temperature change rate, the current target volatile gas concentration, the current target volatile gas concentration change rate, the current target smoke particle concentration change rate, and the current target smoke particle concentration. This data directly reflects the state of physical quantities in the energy storage device's environment that may indicate fire hazards. A predefined algorithm or formula can be used to convert this data into a preliminary pyrolysis particle index, providing a baseline reference for subsequent corrections. Based on the safety state deviation, a preset correction algorithm is used to correct the initial pyrolysis particle index to obtain the current pyrolysis particle target index. By combining multi-dimensional data (including environmental characteristic data and the overall safety state deviation of the device), a more comprehensive and accurate safety risk assessment indicator, namely the pyrolysis particle target index, is provided. The resulting pyrolysis particle target index not only reflects the potential danger level in the current environment but also takes into account the possible future development trends of the energy storage device. This helps operation and maintenance personnel make more scientific and reasonable maintenance and emergency response decisions, significantly improving the early warning capability and safety of the deep learning-based energy storage safety early warning system.

[0113] In some embodiments, determining the initial index of pyrolysis particles at the current moment based on current environmental characteristic data includes:

[0114] Determine the current time-sharing stage category at the current moment according to the current ambient temperature change rate;

[0115] Determine a target corresponding relationship table corresponding to the current time-sharing stage category according to a preset first corresponding relationship between the time-sharing stage category and the corresponding relationship table, wherein the target corresponding relationship table includes corresponding relationships between various parameters and weighting coefficients;

[0116] According to the target correspondence table, determine the weighting coefficients of various parameters in the current environmental feature data;

[0117] The initial index of the pyrolysis particles is determined based on the weighted coefficients of various parameters in the current environmental characteristic data and the current environmental characteristic data.

[0118] Specifically, the various parameters in the current environmental characteristic data include: current ambient temperature, current ambient temperature change rate, current target volatile gas concentration, current target volatile gas concentration change rate, current target smoke particle concentration change rate, and current target smoke particle concentration. Time-based phase categories dynamically divide monitoring phases based on the temperature change rate, representing different levels of reaction and thermal decomposition risk. Time-based phase categories can include steady-state, abnormal heating, and violent decomposition phases. For example, in the steady-state phase, the equipment operates normally with minimal temperature fluctuations. The weights corresponding to the various parameters can be evenly distributed. In the abnormal heating phase, the equipment exhibits an abnormal heating trend, potentially posing a safety hazard, with temperature fluctuations dominating. In this case, the temperature change rate can be assigned a higher weight (e.g., 50%). In the violent decomposition phase, the equipment undergoes a vigorous chemical reaction, resulting in the rapid release of gases and particles, posing an extremely high safety risk. Therefore, higher weights can be assigned to the target volatile gas concentration, target volatile gas concentration change rate, and target smoke particle concentration change rate. Different time-based phase categories have different corresponding relationship tables based on the characteristics of different time-based phases. The corresponding relationship is a table containing various parameters of environmental characteristic data (such as temperature, gas concentration, particle concentration, etc.) and their weighted coefficients, and the weight distribution is dynamically adjusted according to different time-sharing stage categories.

[0119] As an example, determine the current ambient temperature change rate, and map the current ambient temperature change rate to the corresponding time-sharing stage category according to the preset threshold range. For example, if the ambient temperature change rate is low (such as less than 0.1℃ / min), determine that the current time-sharing stage category at the current moment is the steady-state stage; if the temperature change rate is high (such as 0.1℃ / min to 1℃ / min), determine that the current time-sharing stage category at the current moment is the abnormal heating stage; and when the temperature change rate is extremely high (such as >1℃ / min), determine that the current time-sharing stage category at the current moment is the severe decomposition stage. Determine the time-sharing stage category at the current moment based on the current ambient temperature change rate, identify which monitoring stage the energy storage device is currently in, and provide a basis for subsequent analysis. Specifically, the following table can represent the correspondence between the time-sharing stage category and the weighted coefficients of various parameters in the current environmental characteristic data.

[0120]

[0121] Based on the current time-sharing stage category and the first correspondence between the pre-set time-sharing stage category and the correspondence table, a target correspondence table corresponding to the current time-sharing stage category is searched and determined, which is used to indicate the subsequent selection of weighted coefficients for various parameters. When the target correspondence table is determined, the weighted coefficients of the current various parameters applicable to the current time-sharing stage can be extracted from the target correspondence table. The weighted coefficients of the current various parameters can be combined with the current ambient temperature, the current ambient temperature change rate, the current target volatile gas concentration, the current target volatile gas concentration change rate, the current target smoke particle concentration change rate, and the current target smoke particle concentration to calculate the initial pyrolysis particle index through weighted summation. For example, when the current ambient temperature is 78°C, the current ambient temperature change rate is 1.5°C / min, the current target volatile gas concentration is 350 ppm, the current target volatile gas concentration change rate is 40 ppm / min, the current target smoke particle concentration change rate is 70 (µg / m³) / min, and the current target smoke particle concentration is 480µg / m³, the initial pyrolysis particle index is calculated as 3610 by weighted summation based on the weighted coefficients of various parameters of the current environmental characteristic data.

[0122] In some embodiments, the initial index of pyrolysis particles is corrected based on the safety state deviation to obtain the current target index of pyrolysis particles, including:

[0123] If the safety state deviation is less than the first deviation threshold, the pyrolysis particle initial index is determined as the pyrolysis particle target index;

[0124] If the safety state deviation is greater than or equal to the first deviation threshold and less than the second deviation threshold, performing a weighted summation of the safety state deviation and the pyrolysis particle initial index to obtain a first summation result, and determining the first summation result as the pyrolysis particle target index, and the second deviation threshold is greater than the first deviation threshold;

[0125] If the safety state deviation is greater than or equal to the second deviation threshold, the safety state deviation and the pyrolysis particle initial index are weightedly summed to obtain a second summation result, and the second summation result is compared with the value of the preset pyrolysis particle index, and the larger value is determined as the pyrolysis particle target index.

[0126] Specifically, the safety state deviation is a comprehensive indicator obtained by calculating the difference between the predicted amount of energy storage equipment operating state data and the predicted amount of environmental characteristic data and the preset reference range, which can be used to measure the current safety risk level of the equipment. The first deviation threshold is the set low-risk critical value of the safety state deviation, which can indicate that the state of the energy storage equipment is slightly abnormal and requires attention but no immediate intervention. The second deviation threshold is the set high-risk critical value of the safety state deviation, which can indicate that the energy storage equipment is approaching a dangerous state. The preset pyrolysis particle index can be a high-risk critical value of the pyrolysis particle index set empirically, such as the preset pyrolysis particle index can be 4000. When the pyrolysis particle target index exceeds the preset pyrolysis particle index, the highest level warning needs to be directly triggered.

[0127] As an example, the safety state deviation can reflect the degree of difference between the predicted values ​​of the energy storage device's operating state data and environmental characteristic data and a preset reference range. If the safety state deviation is less than a first deviation threshold (e.g., 0.2), the energy storage device can be considered to be in normal or near-normal operating condition. No additional correction is required to the initial pyrolysis particle index; the initial pyrolysis particle index is directly determined as the target pyrolysis particle index. If the safety state deviation is greater than or equal to the first deviation threshold and less than a second deviation threshold (e.g., 0.6), the energy storage device can be considered to have certain safety hazards, but not yet a serious level. In this case, appropriate correction is required to the initial pyrolysis particle index. A weighted summation of the safety state deviation and the initial pyrolysis particle index is performed to obtain a first summation result. Specifically, the first summation result = initial pyrolysis particle index * (1 + safety state deviation). The first summation result is determined as the target pyrolysis particle index. If the safety state deviation is greater than or equal to the second deviation threshold, the energy storage device can be considered to have a serious safety hazard. At this point, a weighted summation of the safety state deviation and the initial pyrolysis particle index is performed to obtain a second summation result. This second summation result is then compared with the preset pyrolysis particle index value, and the larger value is selected as the pyrolysis particle target index. If the preset pyrolysis particle index is 3000, the weighted summation of the safety state deviation and the initial pyrolysis particle index yields a second summation result of 2800. The larger of these two values ​​is used as the final pyrolysis particle target index: pyrolysis particle target index = max(second summation result, preset pyrolysis particle index) = max(2800, 3000) = 3000. This ensures that in high-risk situations, the pyrolysis particle target index fully reflects the potential severity, thereby issuing a higher-level warning signal in a timely manner. The pyrolysis particle index is dynamically adjusted based on the safety state deviation. Combining different levels of safety state deviation, the current pyrolysis particle target index is calculated, providing a more accurate and comprehensive safety risk assessment.

[0128] In some embodiments, determining the warning mode of the energy storage device at the current moment based on the pyrolysis particle target index includes:

[0129] Obtain a first warning threshold and a second warning threshold, where the second warning threshold is greater than the first warning threshold;

[0130] Acquire an energy storage device image, where the energy storage device image includes the energy storage device;

[0131] When the pyrolysis particle target index is greater than or equal to the first warning threshold and less than the second warning threshold, determining the warning mode to be the first warning mode, and controlling the display of the first warning information on the user terminal, the first warning information including the pyrolysis particle target index and the energy storage device image;

[0132] When the pyrolysis particle target index is greater than or equal to the second warning threshold, the warning mode is determined to be the second warning mode, the second warning information is controlled to be displayed on the user end, the user end is controlled to emit a warning sound for prompting the second warning mode, and the energy storage device is controlled to emit a warning sound for prompting the second warning mode. The second warning information includes the pyrolysis particle target index and the energy storage device image.

[0133] Specifically, the first warning threshold is a preset low-risk warning threshold for the pyrolysis particle index (e.g., 1000). This indicates a potential risk to the energy storage device and can alert the user, but no immediate action is required. The second warning threshold is a preset high-risk warning threshold for the pyrolysis particle index (e.g., 3000). When the pyrolysis particle index exceeds the second warning threshold, it indicates that the device is approaching or in a dangerous state, requiring immediate intervention and triggering an emergency response (e.g., audible and visual alarms, equipment shutdown). The first warning mode is a low-risk warning mode. The user terminal displays risk information and a real-time device image, but does not trigger an audible alarm on the device. This serves to remind maintenance personnel to check the device status. For example, they can perform standardized troubleshooting, including: checking battery pack connections (using infrared thermal imaging to assist in locating abnormal temperature rise points); detecting load current limits (by comparing historical operating data); and tightening loose joints (using a torque wrench for standardized operations). Once the potential risk is eliminated, the system automatically resumes monitoring, forming a closed-loop safety loop. The second warning mode is a high-risk warning mode. The user terminal displays an emergency message and emits an audible warning. Simultaneously, the device triggers an audible alarm, drawing attention or enabling immediate emergency measures such as shutdown. The energy storage device image is an image captured by a camera connected to the energy storage device. The camera is connected to the energy storage device and the energy storage device is within the camera's shooting range. The energy storage device is photographed at a preset frequency to obtain image data of the energy storage device.

[0134] As an example, when the pyrolysis particle target index is less than the first warning threshold, no warning mode is required and continuous monitoring is maintained. When the pyrolysis particle target index is greater than or equal to the first warning threshold and less than the second warning threshold, the first warning mode is determined, and a first warning message is displayed on the user terminal. The first warning message includes the pyrolysis particle target index and image data of the energy storage device. For example, the user terminal interface may display "Current pyrolysis particle target index: 1500, please pay attention to device operating status" along with a real-time image of the energy storage device. This visual prompt can attract the user's attention and encourage maintenance personnel to further inspect the device status. When the pyrolysis particle target index is greater than or equal to the second warning threshold, the second warning mode is determined, and the user terminal is controlled to display a second warning message. The second warning message includes the pyrolysis particle target index (e.g., 3400) and image data of the energy storage device. The image data of the energy storage device included in the second warning message is the most recently captured image data. Furthermore, more obvious visual cues, such as a red alert sign or a flashing icon, can be displayed to enhance the warning effect. In addition to visual prompts, specific warning sounds are played at the user end, such as a continuous beep or voice prompt "Emergency warning: The pyrolysis particle target index is too high, please check the equipment immediately." At the same time, the energy storage device is controlled to emit a second warning warning sound, such as a continuous beep or voice prompt "The equipment is abnormal, please deal with it immediately", so that even if the user is not near the user end, the warning information can be received in time. Through a hierarchical warning mechanism, combined with equipment status information and multiple reminder methods, comprehensive monitoring and timely response to the safety status of energy storage equipment can be achieved. The first warning method provides gentle but clear prompts to identify potential problems at an early stage. The second warning method uses strong visual and auditory signals to quickly attract the user's attention and take action in high-risk situations. The multi-level warning strategy not only improves the intelligence level and safety of the system, but also effectively reduces the risk of fire accidents in energy storage equipment and ensures the stable operation of energy storage equipment.

[0135] After determining that the early warning mode is the first early warning mode, the following steps are also included:

[0136] generating a first control instruction, and controlling the ventilation device connected to the energy storage device to turn on based on the first control instruction; and

[0137] A second control instruction is generated, and based on the second control instruction, the load of the energy storage device is controlled to migrate from the energy storage device to the backup line.

[0138] Specifically, after determining that the warning mode is the first warning mode, a first control instruction is generated to activate and start the ventilation device connected to the energy storage device. The first control instruction may include specific parameters for starting the ventilation device, such as wind speed and operating time. Specifically, the first control instruction can be sent to the ventilation device controller via a non-polarity two-bus communication method, instructing it to start immediately. Activating the ventilation device helps quickly reduce the concentration and temperature of volatile gases in the environment surrounding the energy storage device, thereby reducing the risk of fire or other safety accidents.

[0139] In addition, a second control instruction is generated to control the migration of the energy storage device's load from the energy storage device to the backup line, reducing the energy storage device's workload and preventing more serious problems caused by overload. Specifically, the current load status of the energy storage device is determined, including key parameters such as current and power, to determine whether load transfer is necessary. Based on the first warning method and the load assessment results, a second control instruction is generated to instruct the partial or complete transfer of the load from the energy storage device to the backup line. This generated second control instruction can include a detailed load distribution plan to ensure a smooth transition without disrupting normal power supply. The system can execute the second control instruction through the controller, gradually transferring the load from the energy storage device to the backup line. During this process, non-critical loads can be disconnected first, and the load proportion of the backup line can be gradually increased. The coordinated control of ventilation and load migration helps to break the vicious cycle of temperature and current, extend the equipment's safety window, and load balancing across multiple lines can improve power supply reliability, effectively reduce stress on the energy storage device, avoid failure or damage caused by overload, improve the working environment of the energy storage device, and reduce potential combustion or explosion risks. The above-mentioned multi-level early warning and control strategies can improve the intelligence and safety of the system, significantly enhance the ability to respond to emergencies, and ensure the long-term stable operation of energy storage equipment.

[0140] In some embodiments, the operating state data includes voltage and current, and the environmental characteristic data includes ambient temperature, target volatile gas concentration, and target smoke particle concentration. Inputting historical operating state data, historical environmental characteristic data, current operating state data, and current environmental characteristic data into a data prediction network includes:

[0141] The Kalman filter algorithm is used to smooth the historical operating state data, historical environmental characteristic data, current operating state data and current environmental characteristic data to obtain a filtered data sequence;

[0142] Normalizing various parameters in the filtered data sequence to obtain a normalized data sequence, wherein the normalized data sequence includes a voltage data normalized sequence, a current data normalized sequence, an ambient temperature data normalized sequence, a target volatile gas concentration data normalized sequence, and a target smoke particle concentration data normalized sequence;

[0143] Perform feature extraction on the normalized data sequence to generate short-term features, long-term features, and cross-category correlation features;

[0144] The normalized data sequence, short-term features, long-term features, and cross-category correlation features are determined as an input sequence, and the input sequence is input into a data prediction network.

[0145] Specifically, it includes the voltage and current of the energy storage device, which can reflect the electrical operating status of the device. Environmental feature data includes ambient temperature, target volatile gas concentration (such as gases produced by thermal decomposition), and target smoke particle concentration (such as micron-sized particles). It can reflect the state of physical quantities in the environment where the energy storage device is located that may indicate fire hazards. Short-term features are features extracted from data within a short period of time (such as a few seconds) and can reflect instantaneous state changes. Long-term features are features extracted from data over a longer period of time (such as a few minutes) and can reflect trend changes. Cross-category correlation features are correlation features between data of different categories. For example, there may be a correlation between voltage and ambient temperature.

[0146] As an example, a Kalman filter algorithm can be used to perform noise suppression and trend smoothing on the first preset historical operating status data, the current operating status data, the first preset historical environmental characteristic data, and the current environmental characteristic data. This removes noise and smoothes the time series data. By filtering the historical and current data, the impact of sensor noise and other interference factors can be reduced, resulting in a more accurate data series, namely the filtered data series. The various parameters in the filtered data series are normalized, and the mean and standard deviation of each parameter are calculated based on a sliding window (e.g., 24 hours), which can be updated every 30 minutes. A normalization formula is applied to convert data of different dimensions to the same scale (e.g., [0, 1] or [-1, 1]). This generates a normalized data series, which helps ensure that each feature has equal importance during subsequent model training. Specifically, normalization can be performed on voltage, current, ambient temperature, target volatile gas concentration, and target smoke particle concentration, ensuring that the data is compared and analyzed at the same scale, eliminating the impact of dimensional differences and improving the stability and efficiency of model training. Multi-scale feature extraction is performed on the normalized data sequence (normalized data sequence) to generate short-term features, long-term features, and cross-category correlation features. Short-term features are statistical features extracted from data over a short period of time (e.g., seconds), such as mean and variance, which can characterize instantaneous state changes. Long-term features are trend features extracted over longer periods of time (e.g., minutes), such as slope and periodic fluctuations, which are used to characterize long-term behavioral patterns. Cross-category correlation features are correlation features between data of different categories, such as the correlation between voltage and ambient temperature, which help discover underlying complex relationships. Together, short-term features, long-term features, and cross-category correlation features form a more comprehensive data description, providing richer information input for the data prediction network. The normalized data sequence, short-term features, long-term features, and cross-category correlation features are integrated into a complete input sequence, resulting in the aforementioned input sequence, which is then fed into the data prediction network. By fusion of the above-mentioned multi-dimensional data, all available information can be fully utilized, which helps to enhance the adaptability of the deep learning-based energy storage safety early warning method provided in this application to complex application scenarios, and helps the deep learning-based energy storage safety early warning system to identify potential safety hazards at an early stage and take corresponding early warning measures, thereby significantly improving the safety and reliability of energy storage equipment, solving the problems of warning lag, high false alarm rate and inability to integrate multi-dimensional data in traditional technologies, and achieving more accurate and timely safety monitoring.

[0147] In some embodiments, historical environmental characteristic data and current environmental characteristic data are obtained by the following steps:

[0148] Obtaining material parameters of the energy storage device, and determining target volatile gas concentration and target smoke particles based on the material parameters of the energy storage device;

[0149] Obtaining the historical ambient temperature before the current moment and the current ambient temperature at the current moment collected by the temperature sensor, and determining the current ambient temperature change rate based on the historical ambient temperature and the current ambient temperature;

[0150] Obtaining a historical target volatile gas concentration before the current moment and a current target volatile gas concentration at the current moment collected by the electrochemical sensor, and determining a current target volatile gas concentration change rate based on the historical target volatile gas concentration and the current target volatile gas concentration;

[0151] Obtaining a historical target smoke particle concentration before the current moment and a current target smoke particle concentration at the current moment collected by the laser particle sensor, and determining a current target smoke particle concentration change rate based on the historical target smoke particle concentration and the current target smoke particle concentration;

[0152] Determine the historical environmental temperature, historical target volatile gas concentration and historical target smoke particle concentration as historical environmental characteristic data;

[0153] The current ambient temperature, the current ambient temperature change rate, the current target volatile gas concentration, the current target volatile gas concentration change rate, the current target smoke particle concentration change rate and the current target smoke particle concentration are determined as current environmental characteristic data.

[0154] Specifically, the material parameters of an energy storage device are the physical and chemical properties of the materials used, such as the insulation material and electrolyte composition. These parameters directly determine the target volatile gases and smoke particles that may be released during thermal decomposition. The target volatile gas concentration is the concentration of specific gases (such as hydrogen fluoride and carbon monoxide) released by the thermal decomposition of the energy storage device materials and can be measured using electrochemical sensors. The target smoke particle concentration is the concentration of micron- or nanometer-sized smoke particles produced by the thermal decomposition of the energy storage device materials and can be measured using a laser particle sensor.

[0155] As an example, parameters of the energy storage device's main components (such as cathode material, anode material, and electrolyte composition) can be obtained from the device's design documentation or material specifications. The types of gases and smoke particle characteristics that may be released during thermal decomposition can be analyzed to determine the specific types of target volatile gases and smoke particles. For example, the electrolyte in a lithium battery may decompose to produce volatile gases such as hydrogen and carbon monoxide, along with micron-sized smoke particles. Ambient temperature is a key indicator for assessing the operating status of an energy storage device. The first preset historical ambient temperature recorded by a temperature sensor before the current moment and the current ambient temperature at the current moment can be obtained. Specifically, this includes extracting ambient temperature data from a database for a period of time (e.g., the past 10 minutes). Using a temperature sensor to collect the current ambient temperature value in real time, the temperature change rate per unit time is calculated based on the historical and current ambient temperatures to obtain the current ambient temperature change rate. The formula for calculating the current ambient temperature change rate is as follows: Current ambient temperature change rate = (current ambient temperature - average temperature of the previous period) / time interval. The ambient temperature change rate can reflect the trend of rising or falling temperature and help determine whether there is abnormal temperature rise. At the same time, the target volatile gas concentration directly reflects the degree of thermal decomposition of the material within the energy storage device. The first preset historical target volatile gas concentrations collected by the electrochemical sensor before the current moment and the current target volatile gas concentration at the current moment are obtained. Specifically, the method includes: extracting target volatile gas concentration data from a database over a period of time, collecting the current target volatile gas concentration in real time through the electrochemical sensor, and calculating the concentration change rate per unit time based on the historical target volatile gas concentrations and the current target volatile gas concentration to obtain the current ambient temperature change rate. The target smoke particle concentration reflects the thermal decomposition of the material in the energy storage device. The first preset historical target smoke particle concentrations collected by the laser particle sensor before the current moment and the current target smoke particle concentration at the current moment are obtained. Specifically, the method includes: extracting target smoke particle concentration data from a database over a period of time, collecting the current target smoke particle concentration value in real time through the laser particle sensor, and calculating the concentration change rate per unit time based on the historical target smoke particle concentrations and the current target smoke particle concentration to obtain the current target smoke particle concentration change rate. The first preset historical ambient temperature, the first preset historical target volatile gas concentration and the first preset historical target smoke particle concentration are determined as the first preset historical environmental characteristic data, and the current ambient temperature, the current ambient temperature change rate, the current target volatile gas concentration, the current target volatile gas concentration change rate, the current target smoke particle concentration change rate and the current target smoke particle concentration are determined as the current environmental characteristic data.

[0156] By collecting historical and current ambient temperatures, target volatile gas concentrations, and target smoke particle concentrations, and calculating their rates of change, the dynamic changing trends of environmental conditions can be fully captured. The target volatile gas concentrations and target smoke particle concentrations directly reflect the degree of thermal decomposition of energy storage equipment materials. Combined with the ambient temperature change rate, the equipment's safety status can be more accurately assessed. By collecting and analyzing multi-dimensional data, a comprehensive environmental feature dataset is constructed, significantly enhancing the early hazard prediction capabilities and intelligence level of the deep learning-based energy storage safety warning system. This helps address the problems of delayed warnings, high false alarm rates, and the inability to integrate multi-dimensional data that exist in traditional technologies.

[0157] In some embodiments, obtaining the first warning threshold and the second warning threshold includes:

[0158] Obtaining a first preset warning initial value;

[0159] Determine a current first correction value corresponding to the current ambient temperature according to the current ambient temperature and the second mapping relationship, wherein the second mapping relationship is used to represent a mapping relationship between the ambient temperature and the first correction value, and the ambient temperature and the first correction value are positively correlated;

[0160] Correcting the first preset warning initial value according to the current first correction amount to obtain a first warning threshold;

[0161] Obtaining a second preset warning initial value;

[0162] Determining a current second correction value corresponding to the current ambient temperature according to the current ambient temperature and a third mapping relationship, wherein the third mapping relationship is used to represent a mapping relationship between the ambient temperature and the second correction value, and the ambient temperature and the second correction value are positively correlated;

[0163] According to the current second correction amount, the second preset warning initial value is corrected to obtain the second warning threshold.

[0164] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the energy storage safety warning method based on deep learning in the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0165] This application also provides a storage energy safety early warning system based on deep learning, please refer to Figure 2 , the energy storage safety early warning system based on deep learning includes:

[0166] The multi-source data acquisition module 201 is used to acquire historical operating status data and historical environmental characteristic data of the energy storage device at historical moments, and to acquire current operating status data and current environmental characteristic data of the energy storage device at the current moment;

[0167] Prediction module 202 is used to input historical operating status data, historical environmental characteristic data, current operating status data, and current environmental characteristic data into a data prediction network to obtain predicted operating status data and environmental characteristic data of the energy storage device at a future time generated by the data prediction network; the data prediction network is obtained by training a long short-term memory network (LSTM) based on a training sample set;

[0168] The deviation determination module 203 is used to determine the safety state deviation of the energy storage device at the current moment based on the predicted amount of operating state data and the predicted amount of environmental characteristic data;

[0169] An index determination module 204 is used to determine a target index of pyrolysis particles at a current moment based on current environmental characteristic data and a safety state deviation degree;

[0170] The early warning determination module 205 is used to determine the early warning mode of the energy storage device at the current moment according to the pyrolysis particle target index.

[0171] The deep learning-based energy storage safety warning system provided in this application, which utilizes the deep learning-based energy storage safety warning method described in the aforementioned embodiments, can address the technical issue of low accuracy in warning monitoring for energy storage equipment in the prior art. Compared to the prior art, the beneficial effects of the deep learning-based energy storage safety warning system provided in this application are the same as those of the deep learning-based energy storage safety warning method described in the aforementioned embodiments. Other technical features of the deep learning-based energy storage safety warning system are the same as those disclosed in the aforementioned embodiments and are not further elaborated here.

[0172] The present application provides a deep learning-based energy storage safety warning system, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the deep learning-based energy storage safety warning method in the above-mentioned embodiment 1.

[0173] Reference below Figure 3 , which shows a structural diagram of a deep learning-based energy storage safety warning system suitable for implementing an embodiment of the present application. Figure 3 The deep learning-based energy storage safety warning system shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0174] like Figure 3As shown, the deep learning-based energy storage safety warning system may include a processing device 1001 (e.g., a central processing unit, graphics processing unit, etc.), which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 1002 or programs loaded from storage device 1003 into random access memory (RAM) 1004. RAM 1004 also stores various programs and data required for the operation of the deep learning-based energy storage safety warning system. Processing device 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: an input device 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the deep learning-based energy storage safety warning system to communicate with other devices wirelessly or wired to exchange data. Although the figure shows a deep learning-based energy storage safety warning system with various systems, it should be understood that implementation or presence of all the illustrated systems is not required. More or fewer systems may alternatively be implemented or present.

[0175] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device 1003, or installed from a read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are performed.

[0176] The deep learning-based energy storage safety warning system provided in this application, which utilizes the deep learning-based energy storage safety warning method described in the above-mentioned embodiments, can address the technical issue of low accuracy in warning monitoring for energy storage equipment in the prior art. Compared to the prior art, the beneficial effects of the deep learning-based energy storage safety warning system provided in this application are the same as those of the deep learning-based energy storage safety warning method described in the above-mentioned embodiments. Other technical features of this deep learning-based energy storage safety warning system are the same as those disclosed in the above-mentioned embodiments and are not further elaborated here.

[0177] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0178] The above are only specific embodiments of the present application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0179] The present application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, and the computer-readable program instructions are used to execute the deep learning-based energy storage safety early warning method in the above-mentioned embodiment.

[0180] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0181] The above-mentioned computer-readable storage medium can be included in the energy storage safety early warning system based on deep learning; or it can exist independently without being assembled into the energy storage safety early warning system based on deep learning.

[0182] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the energy storage safety warning system based on deep learning, the energy storage safety warning system based on deep learning: obtains historical operating status data and historical environmental characteristic data of the energy storage device at historical moments, and obtains current operating status data and current environmental characteristic data of the energy storage device at the current moment; inputs the historical operating status data, historical environmental characteristic data, current operating status data and current environmental characteristic data into the data prediction network, and obtains the operating status data prediction amount and environmental characteristic data prediction amount of the energy storage device at the future moment generated by the data prediction network; the data prediction network is obtained by training the long short-term memory network LSTM based on the training sample set; determines the safety state deviation of the energy storage device at the current moment according to the operating status data prediction amount and the environmental characteristic data prediction amount; determines the pyrolysis particle target index at the current moment according to the current environmental characteristic data and the safety state deviation; and determines the warning mode of the energy storage device at the current moment according to the pyrolysis particle target index.

[0183] Computer program code for performing the operations of the present application may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0184] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0185] The modules described in the embodiments of the present application may be implemented in software or hardware, wherein the name of a module does not necessarily limit the unit itself.

[0186] The computer-readable storage medium provided in this application stores computer-readable program instructions (i.e., a computer program) for executing the aforementioned deep learning-based energy storage safety early warning method. This computer-readable storage medium can address the low accuracy of early warning monitoring for energy storage devices in the prior art. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided in this application are similar to those of the deep learning-based energy storage safety early warning method provided in the aforementioned embodiments, and are not further elaborated here.

[0187] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned deep learning-based energy storage safety early warning method.

[0188] The computer program product provided in this application can address the low accuracy of early warning monitoring for energy storage devices in the prior art. Compared to the prior art, the beneficial effects of the computer program product provided in this application are similar to those of the deep learning-based energy storage safety early warning method provided in the aforementioned embodiment, and are not further elaborated here.

[0189] The above are only some embodiments of the present application and are not intended to limit the patent scope of the present application. All equivalent structural transformations made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect application in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A deep learning-based energy storage safety early warning method, characterized in that: The energy storage safety warning based on deep learning includes: Obtaining historical operating status data and historical environmental characteristic data of the energy storage device at historical moments, and obtaining current operating status data and current environmental characteristic data of the energy storage device at the current moment; Inputting the historical operating state data, the historical environmental characteristic data, the current operating state data, and the current environmental characteristic data into a data prediction network to obtain a predicted amount of operating state data and a predicted amount of environmental characteristic data of the energy storage device at a future time generated by the data prediction network; the data prediction network is obtained by training a long short-term memory network (LSTM) based on a training sample set; Determining a safety state deviation of the energy storage device at the current moment based on the predicted amount of operating state data and the predicted amount of environmental characteristic data; Determine the current time-sharing stage category at the current moment according to the current ambient temperature change rate; Determine a target correspondence table corresponding to the current time-sharing stage category according to a preset first correspondence between the time-sharing stage category and the correspondence table, wherein the target correspondence table includes correspondences between various parameters and weighting coefficients; Determining the weighting coefficients of various parameters in the current environmental feature data according to the target correspondence table; determining an initial index of pyrolysis particles based on weighted coefficients of various parameters of the current environmental characteristic data and the current environmental characteristic data, wherein the current environmental characteristic data includes a rate of change of the current environmental temperature; If the safety state deviation is less than a first deviation threshold, determining the pyrolysis particle initial index as a pyrolysis particle target index; If the safety state deviation is greater than or equal to the first deviation threshold and less than a second deviation threshold, performing a weighted summation on the safety state deviation and the pyrolysis particle initial index to obtain a first summation result, and determining the first summation result as the pyrolysis particle target index, and the second deviation threshold is greater than the first deviation threshold; If the safety state deviation is greater than or equal to the second deviation threshold, performing a weighted summation on the safety state deviation and the pyrolysis particle initial index to obtain a second summation result, and comparing the second summation result with a preset pyrolysis particle index value, and determining the larger value as the pyrolysis particle target index; An early warning mode of the energy storage device at the current moment is determined according to the pyrolysis particle target index.

2. The energy storage safety early warning method based on deep learning according to claim 1, characterized in that: The step of determining the safety state deviation of the energy storage device at the current moment based on the predicted amount of the operating state data and the predicted amount of the environmental characteristic data includes: Obtaining a preset operating status data reference range and a preset environmental characteristic data reference range corresponding to the energy storage device; Calculating the difference between the predicted running state data and the preset running state data reference range to obtain the running state data deviation; Calculating the difference between the predicted amount of environmental characteristic data and the preset reference range of environmental characteristic data to obtain an environmental characteristic data deviation; The safety state deviation is determined according to the operating state data deviation and the environmental characteristic data deviation.

3. The energy storage safety early warning method based on deep learning according to claim 1, characterized in that: The step of determining the early warning mode of the energy storage device at the current moment according to the pyrolysis particle target index includes: Obtain a first warning threshold and a second warning threshold, wherein the second warning threshold is greater than the first warning threshold; Acquire an energy storage device image, wherein the energy storage device image includes the energy storage device; When the pyrolysis particle target index is greater than or equal to a first warning threshold and less than a second warning threshold, determining that the warning mode is a first warning mode, and controlling the display of a first warning message on the user terminal, the first warning message including the pyrolysis particle target index and the energy storage device image; When the pyrolysis particle target index is greater than or equal to the second warning threshold, the warning mode is determined to be the second warning mode, the second warning information is controlled to be displayed on the user terminal, the user terminal is controlled to emit a warning sound for prompting the second warning mode, and the energy storage device is controlled to emit a warning sound for prompting the second warning mode, and the second warning information includes the pyrolysis particle target index and the energy storage device image.

4. The energy storage safety early warning method based on deep learning according to claim 3 is characterized in that: After determining that the early warning mode is the first early warning mode, the method further includes: generating a first control instruction, and controlling the ventilation device connected to the energy storage device to turn on based on the first control instruction; and A second control instruction is generated, and based on the second control instruction, a load of the energy storage device is controlled to migrate from the energy storage device to a backup line.

5. The energy storage safety early warning method based on deep learning according to claim 1, characterized in that: The operating state data includes voltage and current, the environmental characteristic data includes ambient temperature, target volatile gas concentration, and target smoke particle concentration, and inputting the historical operating state data, the historical environmental characteristic data, the current operating state data, and the current environmental characteristic data into a data prediction network includes: Using a Kalman filter algorithm to smooth the historical operating state data, the historical environmental characteristic data, the current operating state data, and the current environmental characteristic data to obtain a filtered data sequence; Normalizing various parameters in the filtered data sequence to obtain a normalized data sequence, wherein the normalized data sequence includes a voltage data normalized sequence, a current data normalized sequence, an ambient temperature data normalized sequence, a target volatile gas concentration data normalized sequence, and a target smoke particle concentration data normalized sequence; Performing feature extraction on the normalized data sequence to generate short-term features, long-term features, and cross-category correlation features; The normalized data sequence, the short-term features, the long-term features, and the cross-category correlation features are determined as an input sequence, and the input sequence is input into the data prediction network.

6. The energy storage safety early warning method based on deep learning according to claim 1, characterized in that: The historical environmental characteristic data and the current environmental characteristic data are obtained through the following steps: Acquiring material parameters of the energy storage device, and determining a target volatile gas concentration and a target smoke particle size according to the material parameters of the energy storage device; Acquire historical ambient temperatures before the current moment and the current ambient temperature at the current moment, collected by a temperature sensor, and determine a current ambient temperature change rate based on the historical ambient temperatures and the current ambient temperature; Obtaining a historical target volatile gas concentration before the current moment and a current target volatile gas concentration at the current moment, collected by an electrochemical sensor, and determining a current target volatile gas concentration change rate based on the historical target volatile gas concentration and the current target volatile gas concentration; Obtaining a historical target smoke particle concentration before the current moment and a current target smoke particle concentration at the current moment, collected by a laser particle sensor, and determining a current target smoke particle concentration change rate based on the historical target smoke particle concentration and the current target smoke particle concentration; Determining the historical ambient temperature, the historical target volatile gas concentration, and the historical target smoke particle concentration as the historical environmental characteristic data; The current ambient temperature, the current ambient temperature change rate, the current target volatile gas concentration, the current target volatile gas concentration change rate, the current target smoke particle concentration change rate, and the current target smoke particle concentration are determined as the current environmental characteristic data.

7. A deep learning-based energy storage safety early warning system, characterized in that: The deep learning-based energy storage safety warning system includes: a memory, a processor, and a deep learning-based energy storage safety warning program stored in the memory and executable on the processor. The deep learning-based energy storage safety warning program is configured to implement the steps of the deep learning-based energy storage safety warning method according to any one of claims 1 to 6.

Images