CN117633522A - Early warning method and system for battery large model cloud platform - Google Patents
Early warning method and system for battery large model cloud platform Download PDFInfo
- Publication number
- CN117633522A CN117633522A CN202311618031.9A CN202311618031A CN117633522A CN 117633522 A CN117633522 A CN 117633522A CN 202311618031 A CN202311618031 A CN 202311618031A CN 117633522 A CN117633522 A CN 117633522A
- Authority
- CN
- China
- Prior art keywords
- data
- battery
- attention
- voltage
- convolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000000605 extraction Methods 0.000 claims abstract description 49
- 239000000779 smoke Substances 0.000 claims abstract description 39
- 230000007246 mechanism Effects 0.000 claims abstract description 16
- 230000009466 transformation Effects 0.000 claims abstract description 14
- 238000013136 deep learning model Methods 0.000 claims abstract description 12
- 230000001052 transient effect Effects 0.000 claims abstract description 7
- 230000008878 coupling Effects 0.000 claims abstract description 6
- 238000010168 coupling process Methods 0.000 claims abstract description 6
- 238000005859 coupling reaction Methods 0.000 claims abstract description 6
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims abstract description 4
- 238000001931 thermography Methods 0.000 claims description 20
- 230000003750 conditioning effect Effects 0.000 claims description 12
- 238000004146 energy storage Methods 0.000 claims description 12
- 239000002245 particle Substances 0.000 claims description 11
- 238000012423 maintenance Methods 0.000 claims description 9
- 230000004927 fusion Effects 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000004044 response Effects 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 238000013459 approach Methods 0.000 claims description 3
- 239000000498 cooling water Substances 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000007621 cluster analysis Methods 0.000 claims 1
- 238000012795 verification Methods 0.000 claims 1
- 238000003745 diagnosis Methods 0.000 abstract description 4
- 238000000354 decomposition reaction Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000001816 cooling Methods 0.000 description 2
- 239000007788 liquid Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002485 combustion reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000003331 infrared imaging Methods 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000005507 spraying Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Landscapes
- Radiation Pyrometers (AREA)
Abstract
The invention discloses a battery large model cloud platform early warning method and a system, wherein the method comprises the following steps of 1, collecting real-time data of a battery pack; step 2, dividing a group of data points into different clusters with similar properties; step 3, decomposing the signal into transient and local features based on wavelet transformation, inputting the feature values into a deep learning model of an attention mechanism, extracting shallow and deep features, and reconstructing data, wherein the extracted shallow features are input into an attention residual block for deep coupling feature extraction, and finally, reconstructing a prediction TFR of battery voltage and temperature through convolution, reconstructing a time domain model according to wavelet inverse transformation, so as to obtain predicted voltage and temperature data; and 4, performing thermal runaway alarm based on the predicted data. The invention carries out feature decomposition and extraction based on the deep learning model of the packet transformation and the attention mechanism, and simultaneously provides the combined alarm of voltage-temperature-thermal imager-smoke sensor and the like, thereby improving the diagnosis precision and the robustness of thermal runaway.
Description
Technical Field
The invention belongs to the technical field of batteries, and particularly relates to a battery large model cloud platform early warning method and system.
Background
The energy storage power station is established based on a large number of batteries, a plurality of battery modules are densely arranged in a series or parallel connection mode, and in the operation process, the energy storage batteries possibly have safety problems in terms of electricity, heat and the like, once the modules have problems, heat generated by the battery modules can quickly spread fire to the periphery in a heat transfer, heat radiation, combustion object spraying and the like mode, a certain fire scale is formed, phenomena such as explosion and the like can also occur, and harmful gas can also be generated when the batteries burn. Therefore, the fire disaster of the energy storage power station has the characteristics of quick temperature rise, quick spread, large harm and the like, and if the fire disaster can not be stopped before the occurrence of the fire disaster or can be quickly restrained in the early stage of the fire disaster, the situation can not be recovered. How to perform various risk early warning before the problem of the battery module is caused is important to the development of the energy storage industry.
Disclosure of Invention
Aiming at the defects, the invention provides a battery large model cloud platform early warning method and system, which are used for decomposing and extracting features based on a deep learning model of small packet transformation and an attention mechanism, and simultaneously providing combined alarm such as a voltage-temperature-thermal imager-smoke sensor and the like, so that the diagnosis precision and robustness of thermal runaway are improved.
The technical proposal is as follows:
the battery large model cloud platform early warning method comprises the following steps,
step 1, arranging an acquisition point on a battery pack, and arranging a thermal imaging camera on an energy storage cabinet to acquire real-time data of the battery pack;
step 2, dividing a group of data points into different groups or clusters with similar properties;
step 3, decomposing the signal into different scales based on wavelet transformation, including transient characteristics and local detail characteristics, inputting characteristic values into a deep learning model of an attention mechanism, carrying out shallow characteristic extraction, deep characteristic extraction and data reconstruction, wherein the extracted shallow characteristics are input into a plurality of attention residual blocks for deep coupling characteristic extraction, global jump connection is introduced in consideration of loss of shallow information, finally, reconstructing a predicted TFR of battery voltage and temperature through convolution calculation, and reconstructing a time domain model according to wavelet inverse transformation to obtain predicted voltage and temperature data;
and step 4, comparing whether the safety value is exceeded or not based on the predicted voltage and temperature data, and if the safety value is exceeded, giving a thermal runaway alarm.
Preferably, in step 1, voltage data, current data and temperature data of the battery are collected based on the collection points, thermal imaging data are collected based on the thermal imaging camera, and meanwhile, a smoke sensor and a gas sensor are arranged in the energy storage cabinet to collect smoke particle data and gas concentration data respectively.
Preferably, the real-time data of the voltage, the current and the temperature of the battery are classified according to the trend of the response curve, and the batteries with similar trends are divided into the same group; classifying the images based on thermal imaging according to the imaged pixels and the shapes, and dividing the images with similar pixels and similar shapes into the same group; based on the smoke sensor and the gas sensor, classification is performed based on the detected smoke particles and the gas concentration, the smoke particles in a certain interval are classified into the same group, and the gas concentrations between certain growth rate intervals are classified into the same group.
Preferably, a deep learning model of an attention mechanism is input into a current I, a voltage V, a temperature T and a time-frequency representation of an image, multiple convolution is carried out to extract shallow layer characteristics, namely, a result of a previous layer convolution is used as an input of a next layer convolution, and the shallow layer characteristics are extracted step by step; and taking the output of the final stage shallow layer feature extraction as the input of deep layer feature extraction, carrying out multi-attention deep layer feature extraction step by step based on an attention residual error module, and predicting the output voltage V and the time T after the output of the final stage deep layer feature extraction and the output of the shallow layer extraction are fused.
Preferably, the multi-noted deep feature extraction is: will input F in Respectively inputting a self attention collection module, a space attention module and a pixel attention module; the product of input and output of the self-acquisition attention module is fused into H CA ，H CA Input pixel attention module, and product of input and output of the pixel attention module is fused into H CPA The method comprises the steps of carrying out a first treatment on the surface of the The spatial attention module comprises a first layer of parallel three convolution channels, the convolution result of the first channel is connected with the convolution result of the second channel to be used as the input of the second layer of convolution, and the convolution result of the third channel is connected with the output of the second layer of convolution to be subjected to convolution again to output H SA ；H CPA And H SA The product is fused into H MA ，H MA And input F in Output the result F of multi-attention extraction after addition and fusion out 1, a step of; the pixel attention module performs attention extraction output F on the image out 2，F out 2 and F out 1 input TRANSFORM model, output F after being fused in mode out 。
Preferably, the global jump connection is to jump the first-stage convolution result of the shallow feature extraction to the last-stage convolution result of the deep feature extraction, and perform addition fusion.
Preferably, in the step 4, a voltage-temperature-thermal imaging-smoke sensor combined alarm is constructed, specifically, in the step 4.1, key parameters are judged, if the key parameters are judged to approach a threshold value, a prompt fault hidden danger is indicated, at the moment, a system alarms, electric-thermal adjustment is carried out, and in the step 4.2, otherwise, the step 1 is returned; step 4.2, judging the apparent parameters, if the apparent parameters reach the threshold value, indicating that a slight transient fault exists, alarming and stopping at the moment, performing electric-thermal regulation again, and entering step 4.3, otherwise, returning to step 1; and 4.3, judging the apparent parameters again, if the apparent parameters still reach the threshold value, carrying out system alarm, isolating and hosting, contacting maintenance personnel for maintenance, simultaneously judging whether the smoke signal is effective or not by combining the data of the smoke sensor, and starting the fire protection system if the smoke signal is effective.
Preferably, the smoke signal is effective to comprehensively judge the smoke growth rate and the duration, and the smoke growth rate exceeds a preset threshold value within a certain duration is judged to be effective.
Preferably, the electro-thermal conditioning includes electrical conditioning and thermal conditioning, the electrical conditioning: performing fast charge-slow charge adjustment and current reduction rate adjustment based on electrical characteristics of the battery system; thermal conditioning: and adjusting the flow rate of cooling water of the battery pack.
The beneficial effects of the invention are as follows:
(1) The battery large model cloud platform early warning method and system disclosed by the invention collect multi-mode data, can predict future trends by means of a plurality of related factors, has more comprehensive information and more practicability, and can mutually correct signals with different dimensions;
(2) According to the battery large model cloud platform early warning method and system disclosed by the invention, the feature decomposition and extraction are carried out on the basis of the packet transformation and the deep learning model of the attention mechanism, the non-steady state signals can be effectively analyzed by the wavelet transformation, the current and the voltage are decomposed in the time domain, the feature information in the signals is extracted, and the system has the localized characteristic, so that the details of the signals can be better captured;
the attention mechanism network leads in the weight parameter which can be learned, so that the deep learning model can distribute different attention aiming at different parts of input data, provides stronger flexibility and perception capability for the model, and can better process complex tasks and improve performance;
(3) The battery large model cloud platform early warning method and system disclosed by the invention provide combined warning of voltage-temperature-thermal imager-smoke sensor and the like, and improve the diagnosis precision and robustness of thermal runaway.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that are required to be used in the embodiments will be briefly described below.
FIG. 1 is a schematic diagram of a cloud platform early warning system according to one embodiment of the present invention;
FIG. 2 is a schematic diagram of a deep learning model of the attention mechanism of one embodiment of the present invention;
fig. 3 is a schematic view showing a surface temperature distribution of a battery in an overcharged state of the battery according to an embodiment of the present invention;
FIG. 4 is a schematic view of the surface temperature of a battery in a hot-punched state of an internal separator of the battery according to an embodiment of the present invention;
FIG. 5 is an alarm schematic of an embodiment of the present invention;
FIG. 6 is a diagram of multi-modality acquisition according to one embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention discloses a battery large model cloud platform early warning method which is used for battery pack thermal runaway early warning. Thermal runaway of a battery means that the battery undergoes an accumulated enhancing effect of current and battery temperature at the time of constant voltage charge and is gradually damaged. Thermal runaway of a battery is related to the surface temperature distribution, current and voltage of the battery.
A battery large model cloud platform early warning method, which comprises the following steps,
step 1, arranging an acquisition point on a battery pack, and arranging a thermal imaging camera, a smoke sensor and a gas sensor on an energy storage cabinet to acquire real-time data of the battery pack. The collected data includes battery data: voltage data, current data, and temperature data; internet sensor data: smoke sensor data, thermal imaging camera data, gas sensor data; external environment data: external temperature, external humidity.
Step 2, a set of data points is divided into different groups or clusters with similar properties. Classifying the real-time data of the voltage, the current and the temperature of the battery according to the trend of the response curve of the battery, and dividing the battery with similar trend into the same group; referring to fig. 3-4, based on the thermally imaged image, classifying according to the imaged pixels and shapes, dividing pixels having similar pixels, similar shapes into the same group; based on the smoke sensor and the gas sensor, classification is performed based on the detected smoke particles and the gas concentration, the smoke particles in a certain interval are classified into the same group, and the gas concentrations between certain growth rate intervals are classified into the same group.
And 3, decomposing the signal into different scales based on wavelet transformation, including transient characteristics and local detail characteristics, inputting the characteristic values into a deep learning model of an attention mechanism, and carrying out shallow characteristic extraction, deep characteristic extraction and data reconstruction.
Specifically, shallow features are first extracted from TFR of battery current, voltage and temperature; then, the extracted shallow features are input into a plurality of attention residual blocks to carry out deep coupling feature extraction; finally, the predicted TFR of the cell voltage and temperature is reconstructed by convolution calculations. The utility of the generic framework in image processing is verified. The one-dimensional data is converted into a two-dimensional table representation of similar images, as well as thermally imaged images of itself, using wavelet transforms.
And step 4, comparing whether the safety value is exceeded or not based on the predicted voltage and temperature data, and if the safety value is exceeded, giving a thermal runaway alarm.
Examples:
as shown in fig. 6, a smoke sensor and a gas sensor are arranged in the energy storage cabinet, and smoke particle data and gas concentration data are acquired; arranging a collection point on the battery pack, and collecting current, voltage and temperature data of the battery pack; an infrared imaging camera is arranged in the energy storage cabinet to collect thermal imaging data, and the temperature distribution condition of the battery pack is collected more comprehensively and specifically. And collecting points are arranged on the connector and the liquid cooling module at the front end of the battery pack, current, voltage and temperature data of the connector are collected, and data such as flow rate of the liquid cooling module are collected.
Each group of collected data is divided according to its characteristics. Taking the division of the thermal imaging image as an example, the correlation of thermal runaway with battery temperature and the characteristic division based on the thermal imaging image are understood in connection with fig. 3-4.
As shown in fig. 3, the surface temperature distribution of the battery in the overcharged state of the battery is shown. When the external circuit is short-circuited, a large current is generated, a large amount of Joule heat is released, heat is firstly generated at the electrode of the battery, and the heat at the electrode of the battery is continuously not dispersed and is transmitted into the battery, so that the temperature of the whole surface of the battery is raised. The high temperature of the battery starts to spread from two poles of the battery to the whole part of the battery at the 9 th s, the high temperature has spread to the whole surface of the battery at about 17s, once the heat in the battery is continuously accumulated, a series of chemical reactions occur in the battery, the gas generated by the reactions breaks through the battery shell, and the battery is in thermal runaway at the 35 th s. The battery environment in the battery external short-circuit thermal imaging is a dark blue part, and the temperature of the part is about 20 ℃; the battery shell part is light blue and is at about 35 ℃; the appearance of a green portion of the cell surface indicates that the portion is at about 42.5 ℃; yellow part indicates that the part is at about 50 ℃; orange part indicates that the part is at about 57.5 ℃; the red portion indicates that the portion is at about 66 ℃; the white part indicates that the part is above 72.5 ℃. The bluish to white portion is defined as a case region of the battery, and the red to white portion at 60 ℃ or higher is defined as a high temperature region of the battery surface.
As shown in fig. 4, a schematic diagram of the surface temperature distribution of the battery in the state of thermal penetration of the separator inside the battery is shown. After the internal separator is thermally pierced, the temperature at the pierced position of the battery surface rises rapidly, and the high temperature of the battery spreads to substantially the whole battery case at about 18s, and thermal runaway occurs. Since the battery has undergone thermal runaway, there is no further expansion of the high temperature area after 18 s. The battery environment in the battery internal diaphragm thermal puncture thermal imaging is a blue part, and the temperature of the part is about 30 ℃; the color and temperature information of the battery housing portion and the battery surface are the same as the battery external short circuit thermal imaging.
And dividing the acquired group of thermal imaging images according to the pixel points and the similarity of the images. Similarly, the response trend to the collected current voltage temperature is divided into the same group similarly, and will not be described here again.
As shown in fig. 1 to 2, in order to better capture the dynamic and coupling characteristics of the battery current, voltage and temperature, a hybrid deep neural network based on a fusion of residual structure and attention mechanism is proposed.
The network comprises two parts of feature extraction and data reconstruction, wherein the feature extraction is divided into two parts of shallow layer extraction and deep layer extraction. In particular, the method comprises the steps of,
and 3.1, extracting frequency characteristics from the time sequence data by utilizing wavelet analysis, and revealing time-frequency coupling characteristics.
And 3.2, mapping the time-frequency representation of the historical data to the predicted data by adopting deep learning with an attention mechanism.
The input is the time-frequency representation of current I, voltage V, temperature T and images (including thermal imaging images, smoke particle images and gas concentration images), the shallow layer characteristic extraction is carried out by multiple convolution, namely, the result of the previous layer convolution is used as the input of the next layer convolution, and the shallow layer characteristic extraction is carried out step by step; and taking the output of the final stage shallow layer feature extraction as the input of deep layer feature extraction, carrying out multi-attention deep layer feature extraction step by step based on an attention residual error module, and predicting the output voltage V and the time T after the output of the final stage deep layer feature extraction and the output of the shallow layer extraction are fused. Taking the loss of shallow information into consideration, global jump connection is introduced, namely, jumping the first-stage convolution result of the shallow feature extraction to the last-stage convolution result of the deep feature extraction, and carrying out addition fusion.
The multi-noted deep feature extraction is: will input F in Respectively inputting a self attention collection module, a space attention module and a pixel attention module;
the product of input and output of the self-acquisition attention module is fused into H CA ，H CA Input pixel attention module, and product of input and output of the pixel attention module is fused into H CPA ；
The spatial attention module comprises a first layer of parallel three convolution channels, the convolution result of the first channel is connected with the convolution result of the second channel to be used as the input of the second layer of convolution, and the convolution result of the third channel is connected with the output of the second layer of convolution to be subjected to convolution again to output H SA ；
H CPA And H SA The product is fused into H MA ，H MA And input F in Output the result F of multi-attention extraction after addition and fusion out 1；
Pixel attention module output F out 2 and F out 1 input TRANSFORM model, fusion output F on mode out 。
And finally, reconstructing the predicted TFR of the battery voltage and temperature through convolution calculation, and reconstructing a time domain model according to wavelet inverse transformation to obtain predicted voltage and temperature data.
The input F is in Comprising a number ofWord signals (battery voltage, current, temperature) and image signals (thermal imaging images, smoke sensor particles, gas concentration), amplifying after a focusing mechanism, and outputting Fout1 as digital prediction signals, wherein the digital prediction signals comprise changes of battery current, voltage and temperature; the prediction signal Fout2 is an image prediction signal including a pixel change of a thermal imaging image, a change of smoke particles, and a change of gas concentration.
And 3.3, providing combined alarm of voltage-temperature-thermal imager-smoke sensor and the like to improve diagnosis precision and robustness.
As shown in fig. 5, the current I, the voltage V, and the temperature T of the battery pack are monitored, and the current predicted value I ', the voltage predicted value V ', and the temperature predicted value T ' are predicted based on the deep learning of the attention mechanism, and whether the early warning is needed or not is determined based on the parameter determination and the algorithm model, respectively. In particular, the method comprises the steps of,
judging based on an algorithm model as follows: and judging parameters based on a simulated fault training algorithm, if the parameters are judged to be abnormal, indicating that gradual faults are monitored, giving an alarm to the system at the moment, isolating and hosting, contacting maintenance personnel for maintenance, and otherwise, returning to a monitoring state, wherein the simulated fault training algorithm is as follows: and (3) carrying out deep learning training on the data with thermal runaway or the thermal runaway data in the experiment as a basis to obtain a fault algorithm.
Based on the parameters, judging as follows: judging the key parameters, if judging that the key parameters approach a threshold value, indicating that the potential fault hazards exist, alarming the system at the moment, and performing electric-thermal regulation; at this time, further judging the apparent parameters, if judging that the apparent parameters reach a threshold value, then indicating that slight instantaneous faults exist, the electric-thermal regulation in the previous step cannot realize complete regulation, at this time, alarming and stopping, carrying out electric-thermal regulation again, then judging the apparent parameters again, aiming at feedback the electric-thermal regulation effect, if still not completely regulating, combining the monitoring results of the smoke sensor and the gas sensor, and if the concentration of VOC gas (mainly caused by leakage) exceeds the threshold value, alarming the system, isolating and hosting, and contacting maintenance personnel for maintenance; if the smoke concentration exceeds a threshold or a flame is generated, the fire protection system is started. The key parameters are the temperature rise rate of temperature and the voltage and current rise rate, and the apparent parameters are parameters which can be directly measured and comprise current, voltage and temperature. Electro-thermal conditioning includes electrical conditioning: performing fast charge-slow charge adjustment, current rate reduction adjustment, and the like based on the electrical characteristics of the battery system; thermal model adjustment: and adjusting the flow rate of cooling water of the battery pack. Since thermal runaway is a gradual change type problem, the judgment of the change trend of the key parameter and the apparent parameter (i.e. whether the threshold value is exceeded) can prevent noise interference, and the judgment is more accurate.
And (3) joint alarm: the platform provides real-time monitoring and alarm functions, performs real-time data comparison and analysis by continuously monitoring battery parameters and environmental conditions, timely discovers abnormal conditions, generates alarm notification and sends the alarm notification to a user, and notifies related personnel or systems to take corresponding measures. This helps to find problems early and to take corresponding maintenance and management measures to ensure the safety and performance of the battery.
Finally, it should be noted that: the foregoing description is only a preferred embodiment of the present invention, and the present invention is not limited thereto, but it is to be understood that modifications and equivalents of some of the technical features described in the foregoing embodiments may be made by those skilled in the art, although the present invention has been described in detail with reference to the foregoing embodiments. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A battery large model cloud platform early warning method is characterized by comprising the following steps,
step 1, arranging an acquisition point on a battery pack, and arranging a thermal imaging camera on an energy storage cabinet to acquire real-time data of the battery pack;
step 2, dividing a group of data points into different groups or clusters with similar properties;
step 3, decomposing the signal into different scales based on wavelet transformation, including transient characteristics and local detail characteristics, inputting characteristic values into a deep learning model of an attention mechanism, carrying out shallow characteristic extraction, deep characteristic extraction and data reconstruction, wherein the extracted shallow characteristics are input into a plurality of attention residual blocks for deep coupling characteristic extraction, global jump connection is introduced in consideration of loss of shallow information, finally, reconstructing a predicted TFR of battery voltage and temperature through convolution calculation, and reconstructing a time domain model according to wavelet inverse transformation to obtain predicted voltage and temperature data;
and step 4, comparing whether the safety value is exceeded or not based on the predicted voltage and temperature data, and if the safety value is exceeded, giving a thermal runaway alarm.
2. The battery large model cloud platform early warning method according to claim 1, wherein in the step 1, voltage data, current data and temperature data of a battery are collected based on a collection point, thermal imaging data are collected based on a thermal imaging camera, and meanwhile, a smoke sensor and a gas sensor are arranged in an energy storage cabinet to collect smoke particle data and gas concentration data respectively.
3. The battery large model cloud platform early warning method according to claim 2, wherein in step 2, the dividing specifically comprises: classifying according to the trend of a response curve based on real-time data of the voltage, the current and the temperature of the battery; classifying according to the imaged pixels and shapes based on the thermally imaged image; based on the smoke sensor and the gas sensor, classification is made based on the detected smoke particles, gas concentration size.
4. The battery large model cloud platform early warning method according to claim 1, wherein the deep learning model of the attention mechanism is input into time-frequency representations of current I, voltage V, temperature T and images, and shallow feature extraction is carried out by multiple convolution, namely, the result of the previous layer of convolution is used as the input of the next layer of convolution to be extracted step by step; and taking the output of the final stage shallow layer feature extraction as the input of deep layer feature extraction, carrying out multi-attention deep layer feature extraction step by step based on an attention residual error module, and predicting the output voltage V and the time T after the output of the final stage deep layer feature extraction and the output of the shallow layer extraction are fused.
5. The battery large model cloud platform pre-warning method according to claim 4, wherein the multi-attention deep feature extraction is as follows: will input F in Respectively inputting a self attention collection module, a space attention module and a pixel attention module; the product of input and output of the self-acquisition attention module is fused into H CA ，H CA Input pixel attention module, and product of input and output of the pixel attention module is fused into H CPA ；
The spatial attention module comprises a first layer of parallel three convolution channels, the convolution result of the first channel is connected with the convolution result of the second channel to be used as the input of the second layer of convolution, and the convolution result of the third channel is connected with the output of the second layer of convolution to be subjected to convolution again to output H SA ；
H CPA And H SA The product is fused into H MA ，H MA And input F in Output the result F of multi-attention extraction after addition and fusion out 1；
The pixel attention module performs attention extraction output F on the image out 2，F out 2 and F out 1 input TRANSFORM model, output F after being fused in mode out 。
6. The battery large model cloud platform early warning method according to claim 1 or 4, wherein the global jump connection is to jump the first-stage convolution result of the shallow feature extraction to the last-stage convolution result of the deep feature extraction for addition fusion.
7. The battery large model cloud platform early warning method according to claim 2, wherein in step 4, voltage-temperature-thermal imaging-smoke sensor combined warning is constructed, specifically,
step 4.1, judging the key parameters, if judging that the key parameters approach a threshold value, indicating that the potential fault hazard occurs instantaneously, alarming the system at the moment, performing electric-thermal adjustment and entering the step 4.2, otherwise, returning to the step 1;
step 4.2, judging the apparent parameters, if the apparent parameters reach the threshold value, indicating that a slight transient fault exists, alarming and stopping at the moment, performing electric-thermal regulation again, and entering step 4.3, otherwise, returning to step 1;
and 4.3, judging the apparent parameters again, if the apparent parameters still reach the threshold value, carrying out system alarm, isolating and hosting, contacting maintenance personnel for maintenance, simultaneously judging whether the smoke signal is effective or not by combining the data of the smoke sensor, and starting the fire protection system if the smoke signal is effective.
8. The battery large model cloud platform early warning method according to claim 7, wherein the smoke signal is effective for comprehensively judging the smoke growth rate and duration, and the smoke growth rate exceeds a preset threshold value within a certain duration and is judged to be effective.
9. The battery large model cloud platform warning method of claim 7, wherein the electro-thermal conditioning comprises electrical conditioning and thermal conditioning, the electrical conditioning: performing fast charge-slow charge adjustment and current reduction rate adjustment based on electrical characteristics of the battery system;
thermal conditioning: and adjusting the flow rate of cooling water of the battery pack.
10. The utility model provides a battery large model cloud platform early warning system which characterized in that includes:
the data acquisition module is used for arranging acquisition points on the battery pack, arranging a thermal imaging camera on the energy storage cabinet to acquire real-time data of the battery pack to the cloud platform, and dividing the real-time data into a training set and a verification set according to the acquired data set;
the cloud platform module comprehensively considers real-time data to perform feature extraction and cluster analysis, and divides a group of data points into different groups or clusters with similar properties;
the thermal runaway module is used for analyzing transient characteristics and local details in a group or a cluster based on wavelet transformation, taking an analysis result as an input parameter of a deep learning model of an attention mechanism, outputting a future change trend of data, and reconstructing a time domain model based on wavelet inverse transformation to obtain predicted temperature and voltage;
and the early warning module is used for selecting whether to perform thermal runaway early warning or not based on the predicted temperature and voltage data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311618031.9A CN117633522A (en) | 2023-11-30 | 2023-11-30 | Early warning method and system for battery large model cloud platform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311618031.9A CN117633522A (en) | 2023-11-30 | 2023-11-30 | Early warning method and system for battery large model cloud platform |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117633522A true CN117633522A (en) | 2024-03-01 |
Family
ID=90033392
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311618031.9A Pending CN117633522A (en) | 2023-11-30 | 2023-11-30 | Early warning method and system for battery large model cloud platform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117633522A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117848515A (en) * | 2024-03-07 | 2024-04-09 | 国网吉林省电力有限公司长春供电公司 | Switch cabinet temperature monitoring method and system |
-
2023
- 2023-11-30 CN CN202311618031.9A patent/CN117633522A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117848515A (en) * | 2024-03-07 | 2024-04-09 | 国网吉林省电力有限公司长春供电公司 | Switch cabinet temperature monitoring method and system |
| CN117848515B (en) * | 2024-03-07 | 2024-05-07 | 国网吉林省电力有限公司长春供电公司 | Switch cabinet temperature monitoring method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Model-based fault diagnosis approach on external short circuit of lithium-ion battery used in electric vehicles | |
| Yao et al. | Fault detection of the connection of lithium-ion power batteries based on entropy for electric vehicles | |
| US20220123559A1 (en) | System For Detecting, Assessing, and Displaying Battery Faults | |
| Yu et al. | Challenges and outlook for lithium-ion battery fault diagnosis methods from the laboratory to real world applications | |
| CN117633522A (en) | Early warning method and system for battery large model cloud platform | |
| CN116189103B (en) | Equipment monitoring automatic control method and control system based on visual image analysis | |
| Sun et al. | Modified relative entropy-based lithium-ion battery pack online short-circuit detection for electric vehicle | |
| CN112464813A (en) | Method and device for monitoring mountain fire | |
| CN113536440B (en) | Data processing method based on BIM operation and maintenance management system | |
| CN116342999A (en) | Open space fire smoke detection and identification method for ultrahigh voltage transformer substation | |
| CN116704733A (en) | Aging early warning method and system for aluminum alloy cable | |
| CN117612345A (en) | Power equipment state monitoring and alarming system and method | |
| CN113919198A (en) | Electrical fire monitoring and early warning method based on generation of countermeasure network | |
| CN116706973B (en) | Energy storage battery control method, system and medium based on multidimensional analysis | |
| CN116565354A (en) | Fault grading diagnosis early warning method, system and equipment for electrochemical energy storage system | |
| CN117372954A (en) | Charging station safety monitoring method, device, system and computer equipment | |
| CN116956148A (en) | Power system data interaction security threat information analysis method | |
| Du et al. | Battery fault diagnosis and thermal runaway warning based on the Feature-Exponential-Function and Dynamic Time Warping method | |
| CN104849654A (en) | Method for online monitoring breaker | |
| Yang et al. | A multi-sensor characteristic parameter fusion analysis based electrical fire detection model | |
| CN113935423A (en) | Power battery fault early warning method and system for coupling fuzzy control rule with Elman neural network | |
| CN112285631A (en) | Judgment method and system for active early warning of current transformer | |
| Li et al. | Fire detecting for dense bus ducts based on data fusion | |
| Han et al. | Research on Portable Infrared Diagnosis Device of Substation Equipment | |
| KR102668287B1 (en) | Monitorable electric vehicle charging station fire prevention distribution panel and the operating method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |