Energy storage safety early warning method and system based on multimodal perception
Through multimodal perception method and Mendel randomization analysis, combined with the data of multiple detection devices, multi-level abnormality identification and safety warning of energy storage battery packs are achieved, and the incomplete single data source and early warning mechanism in the existing technology are solved. The problem of the warning improves the accuracy and reliability of early warnings.
Patent Information
- Application Number
- CN202410513894.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-26
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-04-26
AI Technical Summary
The prior art safety warning for energy storage battery packs usually only depends on a single data source, lacking the comprehensive processing of multiple monitoring indicators and multi-level abnormal identification, resulting in poor warning accuracy and reliability.
By using a multimodal perception method, by obtaining the historical monitoring data of the energy storage battery pack, using Mendel's randomization to analyze multiple monitoring indicators, establish an early warning correlation level, and lay out corresponding detection devices to build an abnormality recognition model to achieve the Decreasing identification and safety warning of abnormal situations of different levels.
It improves the accuracy and timeliness of safety warning of energy storage battery packs, provides more comprehensive and reliable safety guarantees, and ensures the safe operation of energy storage battery packs.
Smart Images

Figure CN118362923B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of safety early warning technology, and in particular to a storage energy safety early warning method and system based on multimodal sensing. Background Art
[0002] The safety of energy storage battery packs is crucial to the stability of the power grid and the safety of users. Therefore, an effective monitoring and early warning mechanism is needed to ensure its safe operation. However, there are still some problems with the existing technology for energy storage safety early warning. On the one hand, the existing technology for monitoring energy storage battery packs usually relies on only a single data source, such as voltage, current, temperature, etc., lacks a comprehensive basis for judgment, resulting in poor early warning accuracy; on the other hand, traditional safety early warning methods may simply issue an alarm based on whether a certain indicator exceeds a threshold, lacking multi-level division and comprehensive consideration of abnormal situations, and unable to comprehensively evaluate the safety status of energy storage battery packs, resulting in poor early warning reliability. Summary of the invention
[0003] This application provides a storage energy safety warning method based on multimodal perception, aiming to solve the technical problem that the existing technology for storage energy safety warning usually only focuses on the monitoring and analysis of single modal data, lacks comprehensive processing of multiple monitoring indicators, and lacks the classification and warning mechanism of different abnormal situations, resulting in poor accuracy of the warning.
[0004] In view of the above problems, the present application provides a storage energy safety early warning method and system based on multimodal perception.
[0005] In a first aspect disclosed in the present application, a storage safety early warning method based on multimodal perception is provided, the method being applied to an energy storage battery pack, the method comprising: obtaining a historical monitoring data set of the energy storage battery pack, the historical monitoring data set comprising a plurality of monitoring indicators; analyzing the plurality of monitoring indicators by using Mendelian randomization, establishing an early warning association level between the plurality of monitoring indicators and the energy storage battery pack, wherein the early warning association level comprises a first-level early warning association, a second-level early warning association, and a third-level early warning association; based on the first-level early warning association, a first-level detection device is arranged in the energy storage battery pack, and a first-level detection data set is obtained, wherein the first-level detection device comprises a battery voltage detection device and a charge state detection device; based on the second-level early warning association, A secondary detection device is arranged in the battery pack to obtain a secondary detection data set, wherein the secondary detection device set includes a temperature data detection device and a gas concentration detection device; based on the three-level warning association, a third-level detection device is arranged in the energy storage battery pack to obtain a third-level detection data set, wherein the three-level detection device set includes an image acquisition device; based on the abnormality recognition model, the first-level detection data set, the second-level detection data set, and the third-level detection data set are subjected to step-by-step descending abnormality recognition to generate an abnormality recognition result with an abnormality level identifier, wherein the step-by-step descending abnormality recognition is respectively third-level abnormality recognition, second-level abnormality recognition, and first-level abnormality recognition; the abnormality recognition result is sent to the management terminal, and a safety warning is issued based on the abnormality level.
[0006] In a second aspect disclosed in the present application, a storage safety early warning system based on multimodal perception is provided, the system is applied to a storage battery pack, the system is used for the above-mentioned storage safety early warning method based on multimodal perception, the system comprises: a historical monitoring data acquisition module, the historical monitoring data acquisition module is used to acquire a historical monitoring data set of the storage battery pack, the historical monitoring data set comprises a plurality of monitoring indicators; a warning association level establishment module, the warning association level establishment module is used to analyze the plurality of monitoring indicators by using Mendelian randomization, and establish a warning association level between the plurality of monitoring indicators and the storage battery pack, wherein the warning association level comprises a first-level warning association, a second-level warning association, and a third-level warning association; a first-level detection data acquisition module, the first-level detection data acquisition module is used to deploy a first-level detection device in the storage battery pack based on the first-level warning association, and acquire a first-level detection data set, wherein the first-level detection device comprises a battery voltage detection device and a charge state detection device; a second-level detection data acquisition module is used to obtain ... An acquisition module, wherein the secondary detection data acquisition module is used to deploy a secondary detection device in the energy storage battery pack based on the secondary warning association to acquire a secondary detection data set, wherein the secondary detection device set includes a temperature data detection device and a gas concentration detection device; a tertiary detection data acquisition module, wherein the tertiary detection data acquisition module is used to deploy a tertiary detection device in the energy storage battery pack based on the tertiary warning association to acquire a tertiary detection data set, wherein the tertiary detection device set includes an image acquisition device; an abnormality recognition module, wherein the abnormality recognition module is used to perform step-by-step decreasing abnormality recognition on the primary detection data set, the secondary detection data set, and the tertiary detection data set based on an abnormality recognition model, and generate an abnormality recognition result with an abnormality level identifier, wherein the step-by-step decreasing abnormality recognition is, in sequence, a tertiary abnormality recognition, a secondary abnormality recognition, and a primary abnormality recognition; a safety warning module, wherein the safety warning module is used to send the abnormality recognition result to a management terminal, and perform a safety warning based on the abnormality level.
[0007] The third aspect disclosed in the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, any step of the first aspect disclosed in the present application is implemented.
[0008] According to a fourth aspect disclosed in the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, any step of the first aspect disclosed in the present application is implemented.
[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:
[0010] By acquiring a historical monitoring data set and using Mendelian randomization to analyze multiple monitoring indicators, a warning correlation level between multiple monitoring indicators and energy storage battery groups was established, realizing a comprehensive analysis of multimodal data; by establishing primary, secondary, and tertiary warning associations, deploying corresponding detection devices, and constructing anomaly recognition models, an early warning mechanism for abnormal situations of different levels was realized, improving the accuracy and timeliness of early warnings; based on the anomaly recognition model, anomalies were identified for primary, secondary, and tertiary detection data sets, and safety warnings were issued according to the anomaly levels, realizing multi-level anomaly recognition and safety warnings, and more comprehensively guaranteeing the safe operation of energy storage battery groups. In summary, the energy storage safety early warning method based on multimodal perception effectively solves the problems of single-modal data processing, imperfect abnormal early warning mechanism, and lack of multi-level abnormal identification and safety warning in the existing technology, providing a more reliable guarantee for the safe operation of energy storage battery groups.
[0011] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] Figure 1 A schematic diagram of a process flow of a method for energy storage safety early warning based on multimodal perception provided in an embodiment of the present application;
[0013] Figure 2 A schematic diagram of the structure of an energy storage safety early warning system based on multimodal perception provided in an embodiment of the present application;
[0014] Figure 3 This is a diagram of the internal structure of a computer device provided in an embodiment of the present application.
[0015] Explanation of the accompanying drawings: historical monitoring data acquisition module 10, warning association level establishment module 20, first-level detection data acquisition module 30, second-level detection data acquisition module 40, third-level detection data acquisition module 50, anomaly identification module 60, safety warning module 70. DETAILED DESCRIPTION
[0016] The embodiments of the present application provide a storage energy safety early warning method based on multimodal perception, thereby solving the technical problem that the prior art for storage energy safety early warning generally only focuses on the monitoring and analysis of single modal data, lacks comprehensive processing of multiple monitoring indicators, and lacks a classification and early warning mechanism for different abnormal situations, resulting in poor accuracy of the early warning.
[0017] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically introduced below in conjunction with the accompanying drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0018] like Figure 1 As shown, an embodiment of the present application provides an energy storage safety early warning method based on multimodal perception, the method is applied to an energy storage battery pack, and the method includes:
[0019] Acquire a historical monitoring data set of an energy storage battery group, wherein the historical monitoring data set includes multiple monitoring indicators;
[0020] The energy storage safety warning method based on multimodal perception provided in the embodiment of the present application is applied to energy storage battery packs. First, the historical monitoring data of the energy storage battery pack is collected. These data can come from the battery management system, battery database, etc. The monitoring indicators include but are not limited to battery voltage, current, temperature, charge state, etc. In this way, the historical monitoring data set of the energy storage battery pack can be obtained to provide the necessary data basis for subsequent safety warnings.
[0021] Analyze the multiple monitoring indicators by using Mendelian randomization to establish warning association levels between the multiple monitoring indicators and the energy storage battery group, wherein the warning association levels include first-level warning association, second-level warning association, and third-level warning association;
[0022] Mendelian randomization is a statistical method used to evaluate the relationship between two or more variables. In this step, the Mendelian randomization method is used to analyze the correlation between various monitoring indicators and the energy storage battery pack, and to determine the degree of association between them. The specific analysis methods include calculating the correlation coefficient, regression analysis and other statistical methods. According to the results of Mendelian randomization, the monitoring indicators are divided into different warning association levels.
[0023] Among them, the first-level warning is associated with monitoring indicators that are closely related to the safety status of the energy storage battery pack, which can provide early warnings before safety problems occur. These indicators may show small changes or preliminary signs of abnormalities under abnormal conditions, such as slight fluctuations in battery voltage.
[0024] The second-level warning is associated with monitoring indicators that are related to the safety status of the energy storage battery pack. It can provide clearer warning signals when safety problems occur. These indicators may show continuous changes or large fluctuations under abnormal circumstances, such as a significant increase in temperature.
[0025] The third-level warning is associated with monitoring indicators that have a low correlation with the safety status of the energy storage battery pack but indicate serious safety issues. These indicators may appear as extreme values or abnormal conditions under abnormal circumstances, such as detecting abnormal signals such as smoke and fire.
[0026] Based on the first-level warning association, a first-level detection device is arranged in the energy storage battery pack to obtain a first-level detection data set, wherein the first-level detection device includes a battery voltage detection device and a charge state detection device;
[0027] The battery voltage detection device is used to monitor the voltage of each battery cell in the energy storage battery pack in real time. It consists of a group of sensors installed in different positions in the battery pack to capture the voltage information of the battery cells. The data collected by these sensors can be used to monitor the voltage changes of the battery cells in real time; the charge state detection device is used to monitor the charge state of each battery cell in the energy storage battery pack. The charge state of the battery refers to the available energy stored in the battery, which directly affects the performance and safety of the battery.
[0028] The data sets obtained by the battery voltage detection device and the charge state detection device constitute a primary detection data set, which includes the voltage and charge state information of each battery cell in the energy storage battery pack. This information is used to identify potential safety problems.
[0029] Based on the secondary warning association, a secondary detection device is arranged in the energy storage battery pack to obtain a secondary detection data set, wherein the secondary detection device set includes a temperature data detection device and a gas concentration detection device;
[0030] The temperature data detection device is used to monitor the temperature conditions inside the energy storage battery pack in real time. It is composed of a group of temperature sensors, which are installed at different positions of the battery pack to capture temperature data of different parts. The data collected by these sensors can be used to monitor the temperature distribution inside the battery pack in real time to detect whether there are temperature anomalies; the gas concentration detection device is used to monitor the concentration of gases that may be generated inside the energy storage battery pack. When an abnormal situation occurs in the energy storage battery pack, such as overheating or damage of the battery, gas may be released. The gas concentration detection device can detect whether there are abnormal situations by monitoring the concentration of specific gases in the air.
[0031] The data sets acquired by the temperature data detection device and the gas concentration detection device constitute a secondary detection data set, which includes the temperature distribution and gas concentration information inside the energy storage battery pack.
[0032] Based on the three-level warning association, a three-level detection device is arranged in the energy storage battery pack to obtain a three-level detection data set, wherein the three-level detection device set includes an image acquisition device;
[0033] The image acquisition device is used to capture image information inside or outside the energy storage battery pack in real time. The image acquisition device includes a camera for photographing specific areas inside or outside the energy storage battery pack. These images can provide visual information about the state of the energy storage battery pack, such as detecting whether there is smoke, fire or other abnormal conditions in the battery pack. The data set obtained by the image acquisition device constitutes a three-level detection data set, which includes image information inside or outside the energy storage battery pack.
[0034] Based on the anomaly recognition model, the first-level detection data set, the second-level detection data set, and the third-level detection data set are subjected to step-by-step descending anomaly recognition to generate anomaly recognition results with anomaly level identifiers, wherein the step-by-step descending anomaly recognition is, in sequence, the third-level anomaly recognition, the second-level anomaly recognition, and the first-level anomaly recognition;
[0035] The model prepared for anomaly identification can be a machine learning model, such as a neural network, which has been trained with training data in the training phase and has the ability to accurately identify different types of anomalies in the energy storage battery pack. The first-level detection data set, the second-level detection data set, and the third-level detection data set are input into the anomaly identification model. The model processes the data and identifies the anomalies therein, and assigns a corresponding anomaly level to each anomaly. Based on the output of the model, an anomaly identification result with an anomaly level identifier is generated. These results can help managers to promptly discover and resolve safety issues in the energy storage battery pack and ensure its normal operation.
[0036] The abnormality identification result is sent to the management terminal, and a safety warning is issued based on the abnormality level.
[0037] The abnormality identification results are sent to the management terminal through the network and other communication methods. The management terminal is a system for managing energy storage battery packs. After receiving the abnormality identification results, the management terminal analyzes the results and determines the urgency of the safety warning according to the abnormality level. Different warning strategies are adopted for different levels of abnormalities. For example, for level 1 abnormalities, an alarm can be sent to notify the operator; for level 2 abnormalities, measures can be taken automatically, such as adjusting the operating parameters of the battery pack to reduce risks; for level 3 abnormalities, emergency measures can be taken immediately, such as stopping the operation of the battery pack to ensure the safety of personnel.
[0038] Further, based on the anomaly recognition model, the first-level detection data set, the second-level detection data set, and the third-level detection data set are subjected to step-by-step descending anomaly recognition, and an anomaly recognition result with an anomaly level identifier is generated, wherein the step-by-step descending anomaly recognition is respectively the third-level anomaly recognition, the second-level anomaly recognition, and the first-level anomaly recognition, including:
[0039] The anomaly recognition model includes a primary anomaly recognition network, a secondary anomaly recognition network, and a tertiary anomaly recognition network;
[0040] The anomaly recognition model includes a primary anomaly recognition network, a secondary anomaly recognition network, and a tertiary anomaly recognition network;
[0041] Based on the three-level anomaly recognition network, performing three-level anomaly recognition on the three-level detection data set, and when the three-level anomaly recognition result is that an anomaly exists, marking it as a three-level anomaly;
[0042] When the third-level abnormality recognition result is a third-level abnormality, the third-level abnormality is used as the abnormality level identifier, the abnormality recognition process is stopped, and the abnormality level identifier is directly added to the abnormality recognition result for subsequent warning;
[0043] When the third-level anomaly recognition result is that there is no anomaly, based on the second-level anomaly recognition network, the second-level detection data set is subjected to second-level anomaly recognition, and when the second-level anomaly recognition result is that there is an anomaly, it is marked as a second-level anomaly;
[0044] When the secondary abnormality recognition result is a secondary abnormality, the secondary abnormality is used as the abnormality level identifier, the abnormality recognition process is stopped, and the abnormality level identifier is directly added to the abnormality recognition result for subsequent warning;
[0045] When the secondary anomaly recognition result is that there is no anomaly, based on the primary anomaly recognition network, the primary detection data set is subjected to primary anomaly recognition, and when the primary anomaly recognition result is that there is an anomaly, it is marked as a primary anomaly;
[0046] When the first-level abnormality recognition result is a first-level abnormality, the first-level abnormality is used as the abnormality level identifier, and the abnormality level identifier is added to the abnormality recognition result for subsequent warning;
[0047] When the first-level abnormality recognition result is that there is no abnormality, the abnormality recognition result is output as there is no abnormality.
[0048] The anomaly recognition model includes a first-level anomaly recognition network, a second-level anomaly recognition network, and a third-level anomaly recognition network, which are used to identify anomalies of different levels. Among them, the first-level anomaly recognition network is used to identify anomalies associated with the first-level warning. It is trained on the first-level detection data set such as battery voltage and charge state, and can identify anomalies related to it; the second-level anomaly recognition network is used to identify anomalies associated with the second-level warning. It is trained on the second-level detection data set such as temperature and gas concentration, and can identify anomalies related to it; the third-level anomaly recognition network is used to identify anomalies associated with the third-level warning. It is trained on the image acquisition data set and can identify anomalies related to it, such as smoke, fire, etc. Each network focuses on identifying a specific type of anomaly and generates a corresponding anomaly level identifier. This structure can improve the accuracy and efficiency of anomaly recognition.
[0049] The three-level detection data set, including the smoke detection image, is input into the three-level anomaly recognition network. The three-level anomaly recognition network processes the input data to identify whether there is an abnormality. If the three-level anomaly recognition network detects an anomaly, it is identified as a three-level anomaly.
[0050] When the result of the third-level anomaly recognition is that there is no abnormality, the second-level anomaly recognition is performed, and the second-level detection data set, including temperature data and gas concentration data, is input into the second-level anomaly recognition network. The second-level anomaly recognition network processes the input data to identify whether there is an abnormal situation. If the second-level anomaly recognition network detects an abnormality, it will be identified as a second-level anomaly.
[0051] When the result of the second-level abnormality recognition is that there is no abnormality, the first-level abnormality recognition is performed, and the first-level detection data set, including the battery voltage detection data and the charge state detection data, is input into the first-level abnormality recognition network. The first-level abnormality recognition network processes the input data to identify whether there is an abnormal situation. If the first-level abnormality recognition network detects an abnormality, it is identified as a first-level abnormality.
[0052] Among them, the specific first-level anomaly identification, second-level anomaly identification, and third-level anomaly identification processes are elaborated in subsequent steps.
[0053] Among all abnormal situations, level 3 abnormality is the most serious abnormality with the highest priority, followed by level 2 abnormality and finally level 1 abnormality. The highest level abnormality is used as the final abnormality level identification, that is, if there is a level 3 abnormality, it is used as the abnormality level identification; if there is no level 3 abnormality but there is a level 2 abnormality, the level 2 abnormality is used as the abnormality level identification; if there is neither a level 3 abnormality nor a level 2 abnormality but there is a level 1 abnormality, the level 1 abnormality is used as the abnormality level identification; if there is no abnormal situation, the abnormality level is identified as no abnormality.
[0054] Furthermore, constructing the primary anomaly recognition network includes:
[0055] Retrieve historical battery voltage detection records and historical charge state detection records;
[0056] The historical battery voltage detection record includes a historical battery voltage detection data set and a corresponding battery voltage abnormality marking result set as first construction data;
[0057] The historical charge state detection record includes a historical charge state detection data set and a corresponding charge state abnormality marking result set as second construction data;
[0058] Based on the neural network, the battery voltage detection sub-network and the charge state detection sub-network are constructed;
[0059] Using the first construction data to train the battery voltage detection subnetwork until convergence, and using the second construction data to train the charge state detection subnetwork until convergence;
[0060] The trained battery voltage detection subnetwork and the charge state detection subnetwork are connected in parallel to obtain the primary abnormality recognition network.
[0061] From the monitoring system or record database of the energy storage battery pack, retrieve the historical battery voltage detection records and historical charge state detection records over a period of time.
[0062] The historical battery voltage detection records are subdivided into two parts, a historical battery voltage detection data set and a corresponding battery voltage abnormality marking result set, wherein the historical battery voltage detection data set contains historical battery voltage detection data collected from the energy storage battery pack monitoring system, which includes the measured values of the battery voltage; the battery voltage abnormality marking result set contains the abnormal marking results for the historical battery voltage detection data set, and each historical battery voltage data point is associated with a label to identify whether the data point is an abnormal situation.
[0063] The historical battery voltage detection data set and the corresponding abnormal labeling result set are integrated into one data set as the first constructed data, wherein each battery voltage data point has a corresponding abnormal labeling result for supervised learning training model.
[0064] The historical charge state detection records are also divided into two parts, a historical charge state detection data set and a corresponding charge state abnormality marking result set, wherein the historical charge state detection data set includes historical charge state detection data collected from the energy storage battery pack monitoring system, which includes the measurement values of the charge state, such as charging state, discharging state, etc.; the charge state abnormality marking result set includes the abnormal marking results for the historical charge state detection data set, and each historical charge state data point is associated with a label to identify whether the data point is an abnormal situation.
[0065] The historical charge state detection data set and the corresponding abnormal marking result set are integrated into one data set as the second constructed data, wherein each charge state data point has a corresponding abnormal marking result.
[0066] A battery voltage detection subnetwork and a charge state detection subnetwork are constructed based on neural networks. These subnetworks are used to train and learn historical battery voltage detection data and historical charge state detection data, respectively, so as to accurately identify abnormal patterns of different battery voltages and charge states.
[0067] The battery voltage detection subnetwork is trained using the first constructed data, i.e., the historical battery voltage detection data set and the corresponding battery voltage abnormality marking result set. The network parameters are adjusted through optimization methods such as back propagation algorithm and gradient descent so that it can accurately identify battery voltage abnormalities. The training is stopped when the training error converges or reaches a preset number of iterations.
[0068] Similarly, the second constructed data, namely the historical charge state detection data set and the corresponding charge state abnormality marking result set, is used to train the charge state detection sub-network so that it can accurately identify charge state abnormalities.
[0069] By training the battery voltage detection sub-network and the charge state detection sub-network, they can learn the abnormal patterns of battery voltage and charge state in historical data and acquire the ability to identify abnormal situations.
[0070] The trained battery voltage detection subnetwork and charge state detection subnetwork are connected in parallel. This can be achieved by connecting the outputs of the two subnetworks to a shared output layer and ensuring that they work in parallel at the same level. The parallel-connected network is used as a first-level anomaly recognition network, which can simultaneously receive battery voltage detection data and charge state detection data as input and produce anomaly recognition results as output.
[0071] Furthermore, based on the primary anomaly identification network, performing primary anomaly identification on the primary detection data set includes:
[0072] The first-level detection data set includes battery voltage detection data and charge state detection data;
[0073] Inputting the battery voltage detection data into the battery voltage detection subnetwork and outputting a battery voltage abnormality recognition result;
[0074] Inputting the charge state detection data into the charge state detection subnetwork and outputting a charge state abnormality recognition result;
[0075] When any one of the battery voltage abnormality identification result and the charge state abnormality identification result is abnormal, a first-level abnormality identification result is output as the presence of an abnormality.
[0076] The first-level detection data set is a data set obtained by real-time data collection by a battery voltage detection device and a charge state detection device, and includes battery voltage detection data and charge state detection data.
[0077] The battery voltage detection data is taken as input and input into the battery voltage detection subnetwork, which identifies abnormal situations through the patterns and associations in the learned historical data and outputs the abnormal recognition result of the battery voltage, which is a label indicating whether the battery voltage is abnormal.
[0078] Similarly, the charge state detection data is taken as input and input into the charge state detection subnetwork, which also identifies abnormal situations through the patterns and associations in the learned historical data and outputs the abnormal charge state recognition result, which is a label indicating whether the charge state is abnormal.
[0079] A logical judgment is performed. When any of the battery voltage abnormality identification results or the charge state abnormality identification results is abnormal, the first-level abnormality identification result is marked as abnormal. This means that as long as any sub-network determines that an abnormality exists, the overall first-level abnormality identification result will be marked as abnormal.
[0080] Furthermore, based on the secondary anomaly recognition network, performing secondary anomaly recognition on the secondary detection data set includes:
[0081] The secondary anomaly recognition network has embedded temperature warning thresholds and gas concentration warning thresholds;
[0082] The secondary detection data set includes real-time temperature data and real-time gas concentration data;
[0083] When the real-time temperature data reaches the temperature warning threshold, and / or the real-time gas concentration data reaches the gas concentration warning threshold, a secondary abnormality identification result is output as the presence of an abnormality.
[0084] The secondary abnormality recognition network pre-sets a temperature warning threshold to determine whether the temperature has reached an abnormal level. When the real-time temperature data exceeds the temperature warning threshold, it will be determined as an abnormal situation. Similarly, a gas concentration warning threshold is pre-set to determine whether the gas concentration has reached an abnormal level. When the real-time gas concentration data exceeds the gas concentration warning threshold, it will be determined as an abnormal situation. The setting of these warning thresholds is determined based on the characteristics and safety requirements of the energy storage battery pack, and can be determined through historical data analysis, engineering experience or professional knowledge.
[0085] The secondary detection data set is a data set obtained by real-time data collection by a temperature data detection device and a gas concentration detection device, and includes real-time temperature data and real-time gas concentration data.
[0086] When the real-time temperature data reaches the temperature warning threshold and / or the real-time gas concentration data reaches the gas concentration warning threshold, the secondary abnormality recognition result is output as the existence of an abnormality, which means that as long as any parameter in the real-time data exceeds the set warning threshold, it is judged as an abnormal situation.
[0087] Furthermore, constructing the three-level anomaly recognition network includes:
[0088] The three-level detection data set includes smoke detection images and corresponding image acquisition states;
[0089] In the image acquisition state, a calibration detection image and a set of sample smoke detection images are acquired, wherein the calibration detection image is acquired by acquiring images of a normal energy storage battery pack, and the set of sample smoke detection images is acquired by acquiring images of energy storage battery packs with different abnormalities;
[0090] Based on the calibration detection image and the sample smoke detection image set, obtaining a sample similarity set;
[0091] Based on the twin neural network, a smoke recognition image processor is constructed, wherein the smoke recognition image processor includes a first image processing channel and a second image processing channel, and the first image processing channel and the second image processing channel have the same structure and shared parameters;
[0092] The smoke recognition image processor is trained using the calibration detection image, the sample smoke detection image set, and the sample similarity set until a convergence condition is reached to obtain the three-level anomaly recognition network.
[0093] The three-level detection data set is a data set obtained by the image acquisition device through real-time image acquisition, including smoke detection images, and the corresponding image acquisition status is recorded during image acquisition, including lighting conditions, relative position of the image acquisition device, equipment parameters, etc.
[0094] Under the same image acquisition state as the real-time image acquisition process, that is, maintaining exactly the same lighting conditions, relative position of the image acquisition device, equipment parameters, etc., image acquisition is performed on the normal energy storage battery pack to obtain a calibration detection image. The calibration detection image is used to capture image data under normal conditions so that the model can learn image features under normal conditions. Under the same state, image acquisition is performed on energy storage battery packs with different degrees of abnormality to obtain a set of sample smoke detection images. The sample smoke detection image set contains various degrees of abnormal smoke conditions, ranging from mild to severe conditions. These images are used to train the model to identify abnormal features under various smoke levels.
[0095] For each sample smoke detection image, calculate its similarity with the calibration detection image. Various image similarity measurement methods, such as structural similarity index, peak signal-to-noise ratio, etc., can be used to calculate the similarity between images. The similarity between each smoke detection image and its corresponding calibration detection image is used as part of the sample similarity set for training the smoke anomaly recognition model.
[0096] The twin neural network is a special neural network structure, which consists of two sub-networks with the same structure, processing different input data respectively. The two sub-networks share the same parameters, which means that the feature representations they learn are similar. The constructed smoke recognition image processor consists of two image processing channels, including the first image processing channel and the second image processing channel. The two channels have the same structure, including the same neural network hierarchy and structure, and they share the same parameters. Such a smoke recognition image processor can effectively use the structure of the twin neural network to process smoke detection images and realize smoke anomaly recognition.
[0097] The calibration detection image, sample smoke detection image set and sample similarity set are used as training data, where the calibration detection image is used to capture image features under normal conditions, the sample smoke detection image set is used to capture smoke features with different degrees of abnormality, and the sample similarity set is used to quantify the similarity information between images.
[0098] The calibration detection images and sample smoke detection image sets are used as training samples, and the sample similarity set is used to guide model learning. Through the back-propagation algorithm and the optimizer, the parameters of the smoke recognition image processor are continuously adjusted to adapt to different degrees of smoke conditions and improve the accuracy of smoke anomaly recognition. During the training process, the performance of the model on the validation set is monitored, and whether the training has converged is judged based on performance indicators, such as whether the model has converged based on indicators such as the accuracy on the validation set. When the training process reaches the convergence condition, the parameters of the smoke recognition image processor are fixed and used as a three-level anomaly recognition network, which can process the input smoke detection image and output the corresponding anomaly recognition result.
[0099] Furthermore, based on the three-level anomaly recognition network, performing three-level anomaly recognition on the three-level detection data set includes:
[0100] Inputting the smoke detection image and the calibration detection image into the smoke recognition image processor, outputting the similarity, and obtaining the smoke suspicion degree;
[0101] When the smoke suspicion level is greater than or equal to a preset smoke suspicion level threshold, the third-level abnormality recognition result is output as the presence of an abnormality.
[0102] The smoke detection image and the calibration detection image are input into the two image processing channels of the smoke recognition image processor respectively. The two channels extract image features respectively and calculate the similarity between them. The output similarity reflects the similarity between the smoke detection image and the calibration detection image. According to the similarity, the smoke suspicion is obtained. The higher the similarity, the smaller the smoke suspicion, indicating that the possibility of smoke abnormality is smaller; the lower the similarity, the greater the smoke suspicion, indicating that the possibility of smoke abnormality is greater.
[0103] A preset smoke suspicion threshold is set to determine whether there is a smoke anomaly. This threshold can be determined based on actual needs and safety standards, such as experience or experimental data. When the smoke suspicion is greater than or equal to the preset smoke suspicion threshold, it means that the possibility of smoke anomaly is high. At this time, the output of the third-level abnormality recognition result is that there is an abnormality, indicating that the energy storage battery pack has a serious smoke anomaly and immediate countermeasures need to be taken.
[0104] In summary, the energy storage safety early warning method based on multimodal perception provided in the embodiment of the present application has the following technical effects:
[0105] 1. By acquiring a collection of historical monitoring data and analyzing multiple monitoring indicators using Mendelian randomization, we established a warning correlation level between multiple monitoring indicators and energy storage battery groups, and achieved a comprehensive analysis of multimodal data;
[0106] 2. By establishing the association of the first, second and third level warnings, deploying the corresponding detection devices and building an abnormality recognition model, an early warning mechanism for abnormal situations of different levels is realized, which improves the accuracy and timeliness of the early warning;
[0107] 3. Based on the anomaly recognition model, anomaly recognition is performed on the first-level, second-level, and third-level detection data sets, and safety warnings are issued according to the anomaly levels, realizing multi-level anomaly recognition and safety warnings, and more comprehensively ensuring the safe operation of the energy storage battery pack.
[0108] In summary, the energy storage safety warning method based on multimodal perception effectively solves the problems existing in the prior art, such as single-modal data processing, imperfect abnormal warning mechanism, and lack of multi-level abnormality identification and safety warning, and provides a more reliable guarantee for the safe operation of energy storage battery packs.
[0109] Based on the same inventive concept as the energy storage safety early warning method based on multimodal perception in the aforementioned embodiment, Figure 2 As shown, the present application provides an energy storage safety early warning system based on multimodal perception, the system is applied to an energy storage battery pack, and the system includes:
[0110] A historical monitoring data acquisition module 10, wherein the historical monitoring data acquisition module 10 is used to acquire a historical monitoring data set of an energy storage battery group, wherein the historical monitoring data set includes multiple monitoring indicators;
[0111] A warning association level establishment module 20, the warning association level establishment module 20 is used to analyze the multiple monitoring indicators by using Mendelian randomization, and establish a warning association level between the multiple monitoring indicators and the energy storage battery group, wherein the warning association level includes a first-level warning association, a second-level warning association, and a third-level warning association;
[0112] A first-level detection data acquisition module 30, the first-level detection data acquisition module 30 is used to deploy a first-level detection device in the energy storage battery pack based on the first-level warning association, and obtain a first-level detection data set, wherein the first-level detection device includes a battery voltage detection device and a charge state detection device;
[0113] A secondary detection data acquisition module 40, the secondary detection data acquisition module 40 is used to deploy a secondary detection device in the energy storage battery pack based on the secondary warning association, and obtain a secondary detection data set, wherein the secondary detection device set includes a temperature data detection device and a gas concentration detection device;
[0114] A three-level detection data acquisition module 50, which is used to deploy a three-level detection device in the energy storage battery pack based on the three-level warning association to obtain a three-level detection data set, wherein the three-level detection device set includes an image acquisition device;
[0115] An anomaly identification module 60, the anomaly identification module 60 is used to perform step-by-step descending anomaly identification on the first-level detection data set, the second-level detection data set, and the third-level detection data set based on the anomaly identification model, and generate an anomaly identification result with an anomaly level identifier, wherein the step-by-step descending anomaly identification is, in sequence, the third-level anomaly identification, the second-level anomaly identification, and the first-level anomaly identification;
[0116] The security warning module 70 is used to send the abnormality identification result to the management terminal and issue a security warning based on the abnormality level.
[0117] Furthermore, the system further includes an anomaly identification module to perform the following operation steps:
[0118] The anomaly recognition model includes a primary anomaly recognition network, a secondary anomaly recognition network, and a tertiary anomaly recognition network;
[0119] The anomaly recognition model includes a primary anomaly recognition network, a secondary anomaly recognition network, and a tertiary anomaly recognition network;
[0120] Based on the three-level anomaly recognition network, performing three-level anomaly recognition on the three-level detection data set, and when the three-level anomaly recognition result is that an anomaly exists, marking it as a three-level anomaly;
[0121] When the third-level abnormality recognition result is a third-level abnormality, the third-level abnormality is used as the abnormality level identifier, the abnormality recognition process is stopped, and the abnormality level identifier is directly added to the abnormality recognition result for subsequent warning;
[0122] When the third-level anomaly recognition result is that there is no anomaly, based on the second-level anomaly recognition network, the second-level detection data set is subjected to second-level anomaly recognition, and when the second-level anomaly recognition result is that there is an anomaly, it is marked as a second-level anomaly;
[0123] When the secondary abnormality recognition result is a secondary abnormality, the secondary abnormality is used as the abnormality level identifier, the abnormality recognition process is stopped, and the abnormality level identifier is directly added to the abnormality recognition result for subsequent warning;
[0124] When the secondary anomaly recognition result is that there is no anomaly, based on the primary anomaly recognition network, the primary detection data set is subjected to primary anomaly recognition, and when the primary anomaly recognition result is that there is an anomaly, it is marked as a primary anomaly;
[0125] When the first-level abnormality recognition result is a first-level abnormality, the first-level abnormality is used as the abnormality level identifier, and the abnormality level identifier is added to the abnormality recognition result for subsequent warning;
[0126] When the first-level abnormality recognition result is that there is no abnormality, the abnormality recognition result is output as there is no abnormality.
[0127] Furthermore, the system also includes a primary anomaly identification network construction module to perform the following operation steps:
[0128] Retrieve historical battery voltage detection records and historical charge state detection records;
[0129] The historical battery voltage detection record includes a historical battery voltage detection data set and a corresponding battery voltage abnormality marking result set as first construction data;
[0130] The historical charge state detection record includes a historical charge state detection data set and a corresponding charge state abnormality marking result set as second construction data;
[0131] Based on the neural network, the battery voltage detection sub-network and the charge state detection sub-network are constructed;
[0132] Using the first construction data to train the battery voltage detection subnetwork until convergence, and using the second construction data to train the charge state detection subnetwork until convergence;
[0133] The trained battery voltage detection subnetwork and the charge state detection subnetwork are connected in parallel to obtain the primary abnormality recognition network.
[0134] Furthermore, the system further includes a primary anomaly identification module to perform the following operation steps:
[0135] The first-level detection data set includes battery voltage detection data and charge state detection data;
[0136] Inputting the battery voltage detection data into the battery voltage detection subnetwork and outputting a battery voltage abnormality recognition result;
[0137] Inputting the charge state detection data into the charge state detection subnetwork and outputting a charge state abnormality recognition result;
[0138] When any one of the battery voltage abnormality identification result and the charge state abnormality identification result is abnormal, a first-level abnormality identification result is output as the presence of an abnormality.
[0139] Furthermore, the system further includes a secondary anomaly identification module to perform the following operation steps:
[0140] The secondary anomaly recognition network has embedded temperature warning thresholds and gas concentration warning thresholds;
[0141] The secondary detection data set includes real-time temperature data and real-time gas concentration data;
[0142] When the real-time temperature data reaches the temperature warning threshold, and / or the real-time gas concentration data reaches the gas concentration warning threshold, a secondary abnormality identification result is output as the presence of an abnormality.
[0143] Furthermore, the system also includes a three-level anomaly identification network construction module to perform the following operation steps:
[0144] The three-level detection data set includes smoke detection images and corresponding image acquisition states;
[0145] In the image acquisition state, a calibration detection image and a set of sample smoke detection images are acquired, wherein the calibration detection image is acquired by acquiring images of a normal energy storage battery pack, and the set of sample smoke detection images is acquired by acquiring images of energy storage battery packs with different abnormalities;
[0146] Based on the calibration detection image and the sample smoke detection image set, obtaining a sample similarity set;
[0147] Based on the twin neural network, a smoke recognition image processor is constructed, wherein the smoke recognition image processor includes a first image processing channel and a second image processing channel, and the first image processing channel and the second image processing channel have the same structure and shared parameters;
[0148] The smoke recognition image processor is trained using the calibration detection image, the sample smoke detection image set, and the sample similarity set until a convergence condition is reached to obtain the three-level anomaly recognition network.
[0149] Furthermore, the system also includes a three-level anomaly identification module to perform the following operation steps:
[0150] Inputting the smoke detection image and the calibration detection image into the smoke recognition image processor, outputting the similarity, and obtaining the smoke suspicion degree;
[0151] When the smoke suspicion level is greater than or equal to a preset smoke suspicion level threshold, the third-level abnormality recognition result is output as the presence of an abnormality.
[0152] Through the above detailed description of the energy storage safety warning method based on multimodal sensing in this specification, those skilled in the art can clearly understand the energy storage safety warning system based on multimodal sensing in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.
[0153] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as shown in FIG. Figure 3 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus, wherein the processor of the computer device is used to provide computing and control capabilities; the memory of the computer device includes a non-volatile storage medium and an internal memory, the non-volatile storage medium stores an operating system, a computer program and a database, and the internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium; the network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement a storage energy safety early warning method based on multimodal perception.
[0154] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0155] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0156] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. The energy storage safety early warning method based on multimodal perception is characterized by: The method is applied to an energy storage battery pack, and the method comprises: Acquire a historical monitoring data set of an energy storage battery group, wherein the historical monitoring data set includes multiple monitoring indicators; Analyze the multiple monitoring indicators by using Mendelian randomization to establish warning association levels between the multiple monitoring indicators and the energy storage battery group, wherein the warning association levels include first-level warning association, second-level warning association, and third-level warning association; Based on the first-level warning association, a first-level detection device is arranged in the energy storage battery pack to obtain a first-level detection data set, wherein the first-level detection device includes a battery voltage detection device and a charge state detection device; Based on the secondary warning association, a secondary detection device is arranged in the energy storage battery pack to obtain a secondary detection data set, wherein the secondary detection device set includes a temperature data detection device and a gas concentration detection device; Based on the three-level warning association, a three-level detection device is arranged in the energy storage battery pack to obtain a three-level detection data set, wherein the three-level detection device set includes an image acquisition device; Based on the anomaly recognition model, the first-level detection data set, the second-level detection data set, and the third-level detection data set are subjected to step-by-step descending anomaly recognition to generate anomaly recognition results with anomaly level identifiers, wherein the step-by-step descending anomaly recognition is, in sequence, the third-level anomaly recognition, the second-level anomaly recognition, and the first-level anomaly recognition; The abnormality identification result is sent to the management terminal, and a safety warning is issued based on the abnormality level.
2. The method according to claim 1, characterized in that Based on the anomaly recognition model, the first-level detection data set, the second-level detection data set, and the third-level detection data set are subjected to step-by-step descending anomaly recognition to generate anomaly recognition results with anomaly level identifiers, wherein the step-by-step descending anomaly recognition is respectively the third-level anomaly recognition, the second-level anomaly recognition, and the first-level anomaly recognition, including: The anomaly recognition model includes a primary anomaly recognition network, a secondary anomaly recognition network, and a tertiary anomaly recognition network; Based on the three-level anomaly recognition network, performing three-level anomaly recognition on the three-level detection data set, and when the three-level anomaly recognition result is that an anomaly exists, marking it as a three-level anomaly; When the third-level abnormality recognition result is a third-level abnormality, the third-level abnormality is used as the abnormality level identifier, the abnormality recognition process is stopped, and the abnormality level identifier is directly added to the abnormality recognition result for subsequent warning; When the third-level anomaly recognition result is that there is no anomaly, based on the second-level anomaly recognition network, the second-level detection data set is subjected to second-level anomaly recognition, and when the second-level anomaly recognition result is that there is an anomaly, it is marked as a second-level anomaly; When the secondary abnormality recognition result is a secondary abnormality, the secondary abnormality is used as the abnormality level identifier, the abnormality recognition process is stopped, and the abnormality level identifier is directly added to the abnormality recognition result for subsequent warning; When the secondary anomaly recognition result is that there is no anomaly, based on the primary anomaly recognition network, the primary detection data set is subjected to primary anomaly recognition, and when the primary anomaly recognition result is that there is an anomaly, it is marked as a primary anomaly; When the first-level abnormality recognition result is a first-level abnormality, the first-level abnormality is used as the abnormality level identifier, and the abnormality level identifier is added to the abnormality recognition result for subsequent warning; When the first-level abnormality recognition result is that there is no abnormality, the abnormality recognition result is output as there is no abnormality.
3. The method according to claim 2, characterized in that Constructing the first-level anomaly recognition network includes: Retrieve historical battery voltage detection records and historical charge state detection records; The historical battery voltage detection record includes a historical battery voltage detection data set and a corresponding battery voltage abnormality marking result set as first construction data; The historical charge state detection record includes a historical charge state detection data set and a corresponding charge state abnormality marking result set as second construction data; Based on the neural network, the battery voltage detection sub-network and the charge state detection sub-network are constructed; Using the first construction data to train the battery voltage detection subnetwork until convergence, and using the second construction data to train the charge state detection subnetwork until convergence; The trained battery voltage detection subnetwork and the charge state detection subnetwork are connected in parallel to obtain the primary abnormality recognition network.
4. The method according to claim 3, characterized in that Based on the primary anomaly identification network, performing primary anomaly identification on the primary detection data set includes: The first-level detection data set includes battery voltage detection data and charge state detection data; Inputting the battery voltage detection data into the battery voltage detection subnetwork and outputting a battery voltage abnormality recognition result; Inputting the charge state detection data into the charge state detection subnetwork and outputting a charge state abnormality recognition result; When any one of the battery voltage abnormality identification result and the charge state abnormality identification result is abnormal, a first-level abnormality identification result is output as the presence of an abnormality.
5. The method according to claim 2, characterized in that Based on the secondary anomaly recognition network, performing secondary anomaly recognition on the secondary detection data set includes: The secondary anomaly recognition network has embedded temperature warning thresholds and gas concentration warning thresholds; The secondary detection data set includes real-time temperature data and real-time gas concentration data; When the real-time temperature data reaches the temperature warning threshold, and / or the real-time gas concentration data reaches the gas concentration warning threshold, a secondary abnormality identification result is output as the presence of an abnormality.
6. The method according to claim 2, characterized in that Constructing the three-level anomaly recognition network includes: The three-level detection data set includes smoke detection images and corresponding image acquisition states; In the image acquisition state, a calibration detection image and a set of sample smoke detection images are acquired, wherein the calibration detection image is acquired by acquiring images of a normal energy storage battery pack, and the set of sample smoke detection images is acquired by acquiring images of energy storage battery packs with different abnormalities; Based on the calibration detection image and the sample smoke detection image set, obtaining a sample similarity set; Based on the twin neural network, a smoke recognition image processor is constructed, wherein the smoke recognition image processor includes a first image processing channel and a second image processing channel, and the first image processing channel and the second image processing channel have the same structure and shared parameters; The smoke recognition image processor is trained using the calibration detection image, the sample smoke detection image set, and the sample similarity set until a convergence condition is reached to obtain the three-level anomaly recognition network.
7. The method according to claim 6, characterized in that Based on the three-level anomaly recognition network, performing three-level anomaly recognition on the three-level detection data set includes: Inputting the smoke detection image and the calibration detection image into the smoke recognition image processor, outputting the similarity, and obtaining the smoke suspicion degree; When the smoke suspicion level is greater than or equal to a preset smoke suspicion level threshold, the third-level abnormality recognition result is output as the presence of an abnormality.
8. The energy storage safety early warning system based on multimodal perception is characterized by: The system is applied to an energy storage battery pack and is used to implement the energy storage safety early warning method based on multimodal perception according to any one of claims 1 to 7. The system comprises: A historical monitoring data acquisition module, which is used to acquire a historical monitoring data set of the energy storage battery group, wherein the historical monitoring data set includes multiple monitoring indicators; A warning association level establishment module, the warning association level establishment module is used to analyze the multiple monitoring indicators by using Mendelian randomization, and establish a warning association level between the multiple monitoring indicators and the energy storage battery group, wherein the warning association level includes a first-level warning association, a second-level warning association, and a third-level warning association; A first-level detection data acquisition module, the first-level detection data acquisition module is used to deploy a first-level detection device in the energy storage battery pack based on the first-level warning association to obtain a first-level detection data set, wherein the first-level detection device includes a battery voltage detection device and a charge state detection device; A secondary detection data acquisition module, the secondary detection data acquisition module is used to deploy a secondary detection device in the energy storage battery pack based on the secondary warning association, and obtain a secondary detection data set, wherein the secondary detection device set includes a temperature data detection device and a gas concentration detection device; A three-level detection data acquisition module, which is used to deploy a three-level detection device in the energy storage battery pack based on the three-level warning association to acquire a three-level detection data set, wherein the three-level detection device set includes an image acquisition device; An anomaly recognition module, the anomaly recognition module is used to perform step-by-step descending anomaly recognition on the first-level detection data set, the second-level detection data set, and the third-level detection data set based on the anomaly recognition model, and generate an anomaly recognition result with an anomaly level identifier, wherein the step-by-step descending anomaly recognition is three-level anomaly recognition, second-level anomaly recognition, and first-level anomaly recognition in sequence; A security warning module is used to send the abnormality identification result to the management terminal and issue a security warning based on the abnormality level.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the energy storage safety early warning method based on multimodal perception described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the energy storage safety early warning method based on multimodal perception described in any one of claims 1 to 7 are implemented.
