Battery pressure release valve opening state sound feature recognition method and system based on neural network

Abstract

The invention provides a method and a system for recognizing a sound feature of an open state of a battery pressure release valve based on a neural network, and relates to the technical field of safety monitoring of energy storage batteries, and the method comprises the steps: obtaining a sound signal of a target energy storage battery; preprocessing the sound signal by using an EMD denoising method to obtain a target sound signal; acquiring a feature data set in the target sound signal; performing data dimension reduction processing on the feature data set by using a principal component analysis method to obtain a target feature parameter set corresponding to the sound signal; and identifying the target feature parameter set by using a probabilistic neural network. The method achieves the judgment of the opening event of the pressure release valve in the working state of the battery, can achieve the early warning of the fault events, such as thermal runaway, enables safety personnel to take measures in advance, and avoids the threatening of the battery fault to the life safety and property loss of the public.

Description

Technical Field

[0001] The present application relates to the technical field of energy storage battery safety monitoring, and in particular to a method and system for identifying the sound characteristics of a battery pressure relief valve in an open state based on a neural network. Background Art

[0002] As energy storage lithium batteries are put into the energy storage industry in large quantities, fires and explosions in energy storage power stations caused by them often occur, threatening the safety of the staff and property in the station. Safety is a prerequisite for the large-scale promotion and application of energy storage batteries. In practical applications, the safety problem caused by thermal runaway of lithium-ion batteries is the most serious. In practical applications, when lithium-ion batteries are subjected to mechanical abuse, electrical abuse and thermal abuse, the battery materials are damaged and abnormal heating occurs. The continuous accumulation of heat intensifies the internal exothermic chemical reaction, forming a positive feedback, and finally causing thermal runaway. A large amount of heat and flammable gas are released during thermal runaway. When the density of flammable gas reaches a certain concentration, it will explode under high temperature and charged conditions, seriously affecting the safe and stable operation of energy storage power stations. At present, batteries are generally monitored by monitoring the internal temperature of the battery and analyzing the gas content generated by the battery. The warning time of these means occurs almost at the same time as thermal runaway, and it is difficult for staff to take safety measures in time to prevent thermal runaway. Summary of the invention

[0003] In order to solve the above problems, the present application provides a method and system for identifying the sound characteristics of a battery pressure relief valve in an open state based on a neural network.

[0004] In the first aspect, the present application provides a method for identifying the sound characteristics of a battery pressure relief valve in an open state based on a neural network, which adopts the following technical solution: A method for identifying the sound characteristics of a battery pressure relief valve in an open state based on a neural network, comprising: Acquire the sound signal of the target energy storage battery; Preprocessing the sound signal using an EMD denoising method to obtain a target sound signal; Acquiring a feature data set from the target sound signal; Performing data dimension reduction processing on the feature data set by using principal component analysis to obtain a target feature parameter set corresponding to the sound signal; The target feature parameter set is identified using a probabilistic neural network.

[0005] Optionally, the step of obtaining the sound signal of the target energy storage battery includes: When it is determined that the target energy storage battery is being charged, the sound signal of the target energy storage battery is collected by a preset sensor device, wherein the preset sensor device is located in the horizontal direction of the battery cell in the target energy storage battery and on the side of the battery pressure relief valve, and the sampling frequency is required to be greater than or equal to 4kHz.

[0006] Optionally, the step of preprocessing the sound signal using the EMD denoising method to obtain the target sound signal includes: According to the frequency characteristics of the battery pressure relief valve opening sound and the background noise, the collected sound signal is subjected to low-frequency noise processing and baseline drift interference elimination using an empirical mode decomposition method to obtain the target sound signal.

[0007] Optionally, the step of acquiring a feature data set from the target sound signal includes: The sound pressure level, impulse coefficient and wavelet packet decomposition energy of the sound signal are extracted from the target sound signal to generate a feature data set.

[0008] Optionally, the step of performing data dimension reduction processing on the feature data set by using a principal component analysis method to obtain a target feature parameter set corresponding to the sound signal includes: Determine information to be projected according to the feature data set; Projecting the information to be projected on the principal component to generate a projection result; Determining contribution rate information according to the projection result; A target feature parameter set is determined in the contribution rate information. Optionally, after the step of performing data dimension reduction processing on the feature data set by using the principal component analysis method to obtain a target feature parameter set corresponding to the sound signal, the step further includes: Generate a battery profile according to the tag information in the target feature parameter set; Obtain a historical work log, and traverse the historical work log according to the battery portrait to determine whether there is a similar target; If yes, then obtaining identification record information of the same type of target; A quick recognition result is generated according to the recognition record information combined with the sound signal.

[0009] Optionally, after the step of identifying the target feature parameter set by using a probabilistic neural network, the method further includes: When the recognition result is determined to be an abnormal state, determining the abnormal disaster level according to the recognition result, wherein the abnormal disaster level is a first-level alarm state and a second-level warning state; When it is determined that the abnormal disaster level is a first-level alarm state, a first strategy set is determined in the preset strategy set according to the alarm tag; Acquire current maintenance environment information, and determine executable strategies and consumption values ​​corresponding to each executable strategy in the current maintenance environment information; Determining a target alarm strategy according to the consumption value and the first strategy set; When it is determined that the abnormal disaster level is a crisis warning state, warning information is generated and sent to a preset warning port. In a second aspect, the present application provides a system for identifying the sound characteristics of a battery pressure relief valve in an open state based on a neural network, wherein the system for identifying the sound characteristics of a battery pressure relief valve in an open state based on a neural network includes: A signal acquisition module, used to acquire the sound signal of the target energy storage battery; A preprocessing module, used for preprocessing the sound signal using an EMD denoising method to obtain a target sound signal; A data set module, used to obtain a feature data set from the target sound signal; A target feature parameter module, used for performing data dimension reduction processing on the feature data set by using a principal component analysis method to obtain a target feature parameter set corresponding to the sound signal; The recognition module is used to recognize the target feature parameter set using a probabilistic neural network.

[0010] In summary, the present application includes the following beneficial technical effects: This application obtains the sound signal of the target energy storage battery; uses the EMD denoising method to preprocess the sound signal to obtain the target sound signal; obtains a feature data set from the target sound signal; uses the principal component analysis method to perform data dimension reduction processing on the feature data set to obtain a target feature parameter set corresponding to the sound signal; and uses a probabilistic neural network to identify the target feature parameter set. It can judge the pressure relief valve opening event under the working state of the battery, and can warn of thermal runaway and other fault events, so that safety personnel can take measures in advance to avoid threats to public life safety and property losses caused by battery failure. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a flowchart of a first embodiment of a method for identifying the sound characteristics of a battery pressure relief valve in an open state based on a neural network of the present application; Figure 2 This is a spectrum diagram of the first embodiment of the method for identifying the sound characteristics of the battery pressure relief valve opening state based on a neural network of the present application; Figure 3 This is a graph showing the accuracy of the probabilistic neural network in judging the opening state of the pressure relief valve according to the first implementation of the method for recognizing the sound characteristics of the opening state of the pressure relief valve of the battery based on the neural network of the present application; Figure 4It is a structural block diagram of the first embodiment of the system for identifying the sound characteristics of the battery pressure relief valve opening state based on neural network in the present application. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below through the accompanying drawings and embodiments. 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.

[0013] The present application embodiment provides a method for identifying the sound characteristics of a battery pressure relief valve in an open state based on a neural network, referring to Figure 1 , Figure 1 This is a flowchart diagram of the first embodiment of the method for identifying the sound characteristics of the battery pressure relief valve opening state based on a neural network in the present application.

[0014] Step S10: Acquire the sound signal of the target energy storage battery.

[0015] It should be noted that the step of obtaining the sound signal of the target energy storage battery includes: when it is determined that the target energy storage battery is being charged, collecting the sound signal of the target energy storage battery through a preset sensor device, wherein the preset sensor device is located in the horizontal direction of the battery cell in the target energy storage battery and on the side of the battery pressure relief valve, and the sampling frequency is required to be greater than or equal to 4kHz.

[0016] Step S20: pre-processing the sound signal using the EMD denoising method to obtain a target sound signal.

[0017] It can be understood that the step of preprocessing the sound signal using the EMD denoising method to obtain the target sound signal includes: according to the frequency characteristics of the battery pressure relief valve opening sound and the background noise, the empirical mode decomposition method is used to process the low-frequency noise of the collected sound signal and eliminate the baseline drift interference to obtain the target sound signal.

[0018] In a specific implementation, based on the frequency characteristics of the battery pressure relief valve opening sound and the background noise, the empirical mode decomposition method is used to process the low-frequency noise and eliminate the baseline drift interference on the collected sound signal. The collected sound signal is easily interfered by the low-frequency noise in the surrounding environment. This step can improve the detection accuracy of the battery pressure relief valve opening event and improve the anti-interference ability of this embodiment. In this embodiment, the sound signal is collected and stored in segments, and a segment of sound is collected every 4 seconds; based on the background sound being a wind noise signal, the sound signal is decomposed using the EMD empirical mode decomposition method to remove low-frequency noise and eliminate baseline drift interference. Figure 2 Shown are spectrum diagrams, (a) is the background noise spectrum, (b) is the original acquisition noise spectrum, and (c) is the processed noise spectrum.

[0019] Step S30: Acquire a feature data set from the target sound signal.

[0020] It should be noted that the step of obtaining a feature data set from the target sound signal includes: extracting the sound pressure level, impulse coefficient and wavelet packet decomposition energy of the sound signal from the target sound signal to generate a feature data set.

[0021] In the specific implementation, the characteristic parameters include: sound pressure level of sound signal, impulse coefficient, wavelet packet decomposition energy and other multi-dimensional characteristic parameters. They can be expressed by the following formula: ①Sound pressure level: Among them, P is the instantaneous sound pressure, P0 is the reference sound pressure, which is usually taken as 2×10-5Pa, which is the lowest sound pressure level that the human ear can perceive.

[0022] ②Impact coefficient: It represents the ratio of the peak value to the absolute mean value and can be used to detect whether there is a shock in the signal.

[0023] ③ Wavelet packet energy: Wavelet packet transform provides the energy distribution of the signal at different frequencies and time scales. i is the wavelet packet coefficient ω of each node i i The energy of , where k is the time index of the coefficient at that node.

[0024] Step S40: using principal component analysis to perform data dimension reduction processing on the feature data set to obtain a target feature parameter set corresponding to the sound signal.

[0025] It can be understood that the steps of using principal component analysis to perform data dimensionality reduction processing on a feature data set to obtain a target feature parameter set corresponding to a sound signal include: determining information to be projected based on the feature data set; projecting the information to be projected onto the principal component to generate a projection result; determining contribution rate information based on the projection result; and determining the target feature parameter set in the contribution rate information. In the specific implementation, the principal component analysis method is used in this embodiment to perform data dimensionality reduction processing on multidimensional features. The steps include: projecting the original data onto the selected principal component, calculating the contribution rate of each principal component, the formula is as follows, and the principal component with a cumulative contribution rate exceeding 80% is taken as the new feature vector in the principal component analysis method.

[0026] Contribution rate: ,in is the i-th eigenvalue.

[0027] Step S50: using a probabilistic neural network to identify the target feature parameter set.

[0028] In a specific implementation, the method for classifying and identifying characteristic parameters is to obtain the sound signal of the target energy storage battery in a deflated state to obtain an abnormal sound signal data set, numbered 1; to obtain the sound signal of the target energy storage battery in a normal state to obtain a normal sound signal data set, numbered 2; after preprocessing the sound signal data, as an optional implementation method, in this example, the sound signal is processed in frames, with a frame length of 1024 and a frame shift of 512; in the principal component analysis method, the principal component with a cumulative contribution rate exceeding 80% is taken as a new feature vector to obtain a standard battery pressure relief valve normal / abnormal feature data set, and the two set data are used as training sets for probabilistic neural network training.

[0029] It should be noted that, in this embodiment, after determining the target feature parameter set, the feature vector after dimension reduction is input into the probabilistic neural network. If the output is an abnormal state, an early warning is issued; if the output is a normal state, monitoring continues.

[0030] It can be understood that the probabilistic neural network of this embodiment has a strong nonlinear modeling capability. The training process of the model is completed in one go. Compared with other neural networks, it does not require multiple iterations of training, which can greatly improve the training efficiency and reduce the training time. In this embodiment, the radial basis function expansion coefficient of the neural network is selected to be 0.6. Figure 3 This is a graph showing the accuracy of the probabilistic neural network in judging the opening state of the pressure relief valve in an embodiment of the present invention. The accuracy of the judgment on the opening state of the test pressure relief valve can reach 89%, which is a good effect.

[0031] It should be noted that after the step of using the principal component analysis method to perform data dimensionality reduction processing on the feature data set to obtain the target feature parameter set corresponding to the sound signal, it also includes: generating a battery portrait based on the label information in the target feature parameter set; obtaining the historical work log, and traversing the historical work log according to the battery portrait to determine whether there are similar targets; if so, obtaining the recognition record information of the similar targets; and generating a quick recognition result based on the recognition record information combined with the sound signal.

[0032] It should be noted that in this embodiment, after determining the target characteristic parameters, data tags will be generated according to the specific data content to further generate a battery portrait. In this embodiment, the battery portrait is essentially a data set with a certain number of tags. Each tag corresponds to a specific data type and value. The record of each identification is stored in the historical work log, and the battery portrait can be used to match whether there are similar identification records. After matching similar identification records, the sound signal recognition logic in the corresponding identification record is obtained for quick identification. The logic of quick identification is to obtain the identification node in the historical identification record, and quick identification is achieved by inputting the information to be identified this time into the identification node.

[0033] It can be understood that after the step of using a probabilistic neural network to identify the target feature parameter set, it also includes: when the identification result is determined to be an abnormal state, determining the abnormal disaster level according to the identification result, wherein the abnormal disaster level is a first-level alarm state and a second-level warning state; when the abnormal disaster level is determined to be a first-level alarm state, determining a first policy set according to the alarm label in the preset policy set; obtaining current maintenance environment information, determining executable policies and consumption values ​​corresponding to each executable policy in the current maintenance environment information; determining a target alarm policy based on the consumption value and the first policy set; when it is determined that the abnormal disaster level is a crisis warning state, generating warning information and sending the warning information to a preset warning port.

[0034] It should be noted that, in this embodiment, after the sound feature is recognized, whether an abnormality occurs will be determined based on the recognition result. When determining an abnormal situation, the abnormal situation will be graded into a first-level alarm state and a second-level warning state. The first-level alarm state indicates that the abnormal situation needs to be handled in a timely manner, and the method with the minimum processing time / processing resources is selected in combination with the existing processing resources.

[0035] In the specific implementation, this embodiment detects the noise signal of the selected noise detection point of the energy storage lithium battery in the running state, and uses the EMD method to denoise the sound signal; extracts the sound pressure level, impact coefficient and wavelet packet energy characteristic parameters of the sound signal for the noise signal, and performs dimensionality reduction processing on the characteristic data, so that the characteristics can more efficiently and accurately describe the battery pressure relief valve state; the probabilistic neural network can judge the pressure relief valve state of the input characteristic signal under the training of a sufficient number of sample features, and the effect is good. Therefore, this embodiment is based on the characteristics that the internal chemical reaction rate of the battery changes before the thermal runaway occurs, the speed of gas generation is different, and the pressure relief valve will open and release gas in advance before the thermal runaway occurs. It can use the noise characteristics to monitor whether the battery pressure relief valve is open online, without contacting the live equipment, and the detection is safe. There is no need for the staff to go to the battery compartment to detect and analyze the battery, which saves time and effort, and has the advantages of safety, efficiency and convenient testing.

[0036] It can be understood that this embodiment can be conveniently combined with remote transmission technology to realize real-time monitoring and early thermal runaway warning of energy storage batteries, facilitate remote centralized monitoring of multiple energy storage battery compartments, and help improve the automation level of battery thermal runaway early warning status monitoring.

[0037] In a specific implementation, this embodiment uses sound signals as detection objects for detecting battery pressure relief valve opening events. The sound sensor is a non-contact sensor, which is relatively simple and convenient to install, and can effectively solve the problem that contact sensors are difficult to install. Moreover, since non-contact infrared laser detection sensors require a certain detection space, they have certain requirements for the installation space, and are easily affected by objects invading the detection space and affecting the detection accuracy. The present invention uses sound signals as detection objects, and sound can be transmitted through a variety of media such as gas and solid. Therefore, the detection space requirements are low, and the detection accuracy will not be affected by objects invading the detection space, which can effectively improve the accuracy of feature signal acquisition.

[0038] This embodiment obtains the sound signal of the target energy storage battery; uses the EMD denoising method to pre-process the sound signal to obtain the target sound signal; obtains a feature data set in the target sound signal; uses the principal component analysis method to perform data dimension reduction processing on the feature data set to obtain a target feature parameter set corresponding to the sound signal; and uses a probabilistic neural network to identify the target feature parameter set. It can judge the pressure relief valve opening event under the working state of the battery, and can warn of thermal runaway and other fault events, so that safety personnel can take measures in advance to avoid threats to public life safety and property losses caused by battery failure. In addition, an embodiment of the present application also proposes a computer-readable storage medium, on which is stored a program for identifying the sound characteristics of the open state of a battery pressure relief valve based on a neural network. When the program for identifying the sound characteristics of the open state of a battery pressure relief valve based on a neural network is executed by a processor, the steps of the method for identifying the sound characteristics of the open state of a battery pressure relief valve based on a neural network as described above are implemented.

[0039] Reference Figure 4 , Figure 4 This is a structural block diagram of the first embodiment of the system for identifying the sound characteristics of the battery pressure relief valve open state based on a neural network in the present application.

[0040] like Figure 4 As shown, the system for identifying the sound characteristics of the battery pressure relief valve opening state based on a neural network proposed in the embodiment of the present application includes: A signal acquisition module 10 is used to acquire a sound signal of a target energy storage battery; A preprocessing module 20, used for preprocessing the sound signal using an EMD denoising method to obtain a target sound signal; A data set module 30, used to obtain a feature data set from a target sound signal; The target feature parameter module 40 is used to perform data dimension reduction processing on the feature data set by using the principal component analysis method to obtain a target feature parameter set corresponding to the sound signal; The recognition module 50 is used to recognize the target feature parameter set using a probabilistic neural network.

[0041] This embodiment obtains the sound signal of the target energy storage battery; uses the EMD denoising method to pre-process the sound signal to obtain the target sound signal; obtains a feature data set in the target sound signal; uses the principal component analysis method to perform data dimension reduction processing on the feature data set to obtain a target feature parameter set corresponding to the sound signal; and uses a probabilistic neural network to identify the target feature parameter set. It can judge the pressure relief valve opening event under the working state of the battery, and can warn of thermal runaway and other fault events, so that safety personnel can take measures in advance to avoid threats to public life safety and property losses caused by battery failure.

[0042] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present application. In practical applications, technicians in this field can select part or all of them according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0043] In addition, for technical details not fully described in this embodiment, reference can be made to the method for identifying the sound characteristics of the open state of a battery pressure relief valve based on a neural network provided in any embodiment of the present application, which will not be repeated here.

[0044] In addition, it should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0045] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0046] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory (ROM) / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of each embodiment of the present application.

[0047] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for identifying the sound characteristics of a battery pressure relief valve in an open state based on a neural network, characterized in that: include: Acquire the sound signal of the target energy storage battery; Preprocessing the sound signal using an EMD denoising method to obtain a target sound signal; Acquiring a feature data set from the target sound signal; Performing data dimension reduction processing on the feature data set by using principal component analysis to obtain a target feature parameter set corresponding to the sound signal; The target feature parameter set is identified using a probabilistic neural network.

2. The method for identifying the sound characteristics of the battery pressure relief valve opening state based on a neural network according to claim 1, characterized in that: The step of obtaining the sound signal of the target energy storage battery includes: When it is determined that the target energy storage battery is being charged, the sound signal of the target energy storage battery is collected by a preset sensor device, wherein the preset sensor device is located in the horizontal direction of the battery cell in the target energy storage battery and on the side of the battery pressure relief valve, and the sampling frequency is required to be greater than or equal to 4kHz.

3. The method for identifying the sound characteristics of the battery pressure relief valve opening state based on a neural network according to claim 2, characterized in that: The step of preprocessing the sound signal using the EMD denoising method to obtain the target sound signal comprises: According to the frequency characteristics of the battery pressure relief valve opening sound and the background noise, the collected sound signal is subjected to low-frequency noise processing and baseline drift interference elimination using an empirical mode decomposition method to obtain the target sound signal.

4. The method for identifying the sound characteristics of the battery pressure relief valve opening state based on a neural network according to claim 1, characterized in that: The step of acquiring a feature data set from the target sound signal comprises: The sound pressure level, impulse coefficient and wavelet packet decomposition energy of the sound signal are extracted from the target sound signal to generate a feature data set.

5. The method for identifying the sound characteristics of the battery pressure relief valve opening state based on a neural network according to claim 1, characterized in that: The step of performing data dimension reduction processing on the feature data set by using the principal component analysis method to obtain a target feature parameter set corresponding to the sound signal includes: Determine information to be projected according to the feature data set; Projecting the information to be projected on the principal component to generate a projection result; Determining contribution rate information according to the projection result; A target feature parameter set is determined in the contribution rate information.

6. The method for identifying the sound characteristics of the battery pressure relief valve opening state based on a neural network according to claim 1, characterized in that: After the step of performing data dimension reduction processing on the feature data set by using the principal component analysis method to obtain the target feature parameter set corresponding to the sound signal, the method further includes: Generate a battery profile according to the tag information in the target feature parameter set; Obtain a historical work log, and traverse the historical work log according to the battery portrait to determine whether there is a similar target; If yes, then obtaining identification record information of the same type of target; A quick recognition result is generated according to the recognition record information combined with the sound signal.

7. The method for identifying the sound characteristics of the battery pressure relief valve opening state based on a neural network according to claim 1, characterized in that: After the step of identifying the target feature parameter set by using a probabilistic neural network, the method further includes: When the recognition result is determined to be an abnormal state, determining the abnormal disaster level according to the recognition result, wherein the abnormal disaster level is a first-level alarm state and a second-level warning state; When it is determined that the abnormal disaster level is a first-level alarm state, a first strategy set is determined in the preset strategy set according to the alarm tag; Acquire current maintenance environment information, and determine executable strategies and consumption values ​​corresponding to each executable strategy in the current maintenance environment information; Determining a target alarm strategy according to the consumption value and the first strategy set; When it is determined that the abnormal disaster level is a crisis warning state, warning information is generated and sent to a preset warning port.

8. A system for identifying the sound characteristics of a battery pressure relief valve in an open state based on a neural network, characterized in that: The system comprises: A signal acquisition module, used to acquire the sound signal of the target energy storage battery; A preprocessing module, used for preprocessing the sound signal using an EMD denoising method to obtain a target sound signal; A data set module, used to obtain a feature data set from the target sound signal; A target feature parameter module, used for performing data dimension reduction processing on the feature data set by using a principal component analysis method to obtain a target feature parameter set corresponding to the sound signal; The recognition module is used to recognize the target feature parameter set using a probabilistic neural network.

Images