Fire monitoring method and device for lithium-ion capacitor frequency modulation system

Abstract

The present application relates to the field of intelligent monitoring, and specifically discloses a fire monitoring method and device for a lithium-ion capacitor frequency modulation system, which uses artificial intelligence-based data analysis technology to perform time series analysis on the internal and external temperatures, currents, and voltages of a liquid-cooled lithium-ion capacitor module, in order to dig out the time series coordinated distribution pattern of the internal and external temperatures of the capacitor module and the time series change trend of the working state of the module, and then by querying and responding to both, the correlation response relationship between the working state of the module and the temperature is captured, and the danger level of a fire that may occur in the module is intelligently estimated. In this way, the limitations of traditional fire monitoring methods can be effectively overcome, and more accurate and timely fire monitoring and early warning can be achieved, thereby improving the safety and reliability of lithium-ion capacitor frequency modulation systems.

Description

Technical Field

[0001] The present application relates to the field of intelligent monitoring technology, and more specifically, to a fire monitoring method and device for a lithium-ion capacitor frequency modulation system. Background Art

[0002] With the rapid development of lithium-ion battery technology, lithium-ion capacitor frequency regulation system (LC-FRS) has been widely used in energy storage, electric vehicles and smart grids. This type of system not only has the advantages of high energy density, long cycle life and environmental protection, but also has the ability to charge and discharge quickly, and can respond to grid load changes in a short time to achieve frequency regulation and power quality control. However, in practical applications, the safety issues of lithium-ion capacitor modules, especially the risk of fire, have always been one of the key factors restricting their widespread application.

[0003] Lithium-ion capacitor modules generate a lot of heat during the charging and discharging process. If the heat dissipation is poor or the system fails, the temperature inside the module will rise sharply, causing electrolyte decomposition, thermal runaway, and even fire. In addition, the working state of the module (such as current and voltage) will also affect its thermal stability. Abnormal current and voltage fluctuations may cause short circuits or thermal runaway inside the module. Therefore, it is important to monitor the fire of lithium-ion capacitor modules and promptly detect and warn of potential fire risks to ensure the safe and stable operation of the system.

[0004] Traditional fire monitoring methods mainly rely on devices such as temperature sensors and smoke detectors. Although these monitoring methods can detect the occurrence of fire to a certain extent, they have obvious limitations. For example, a single temperature threshold setting often cannot accurately reflect the complex thermal state changes inside the module, and it is difficult to adapt to the actual temperature rise of lithium-ion capacitors under different working conditions, which can easily lead to false alarms or missed alarms; and smoke detectors usually need to respond only after the fire develops to a certain stage, and cannot warn of potential fire risks in advance.

[0005] Therefore, an optimized fire monitoring method and device for a lithium-ion capacitor frequency modulation system is desired. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a fire monitoring method and device for a lithium-ion capacitor frequency modulation system, which uses artificial intelligence-based data analysis technology to perform time series analysis on the internal and external temperatures, currents and voltages of the liquid-cooled lithium-ion capacitor module, so as to dig out the time series coordinated distribution pattern of the internal and external temperatures of the capacitor module and the time series change trend of the working state of the module, and then by encoding the query response of the two, the correlation response relationship between the working state of the module and the temperature is captured, and the danger level of the fire that may occur in the module is intelligently estimated. In this way, the limitations of traditional fire monitoring methods can be effectively overcome, and more accurate and timely fire monitoring and early warning can be achieved, thereby improving the safety and reliability of lithium-ion capacitor frequency modulation systems.

[0007] Accordingly, according to one aspect of the present application, a fire monitoring method for a lithium-ion capacitor frequency modulation system is provided, comprising:

[0008] The time queue of the internal temperature of the module and the time queue of the external temperature of the module are collected by temperature sensors deployed inside and outside the liquid-cooled lithium-ion capacitor module, and the time queue of the module current and the time queue of the module voltage are collected by current sensors and voltage sensors deployed inside the liquid-cooled lithium-ion capacitor module;

[0009] Transmitting the time queue of the module internal temperature, the time queue of the module external temperature, the time queue of the module current and the time queue of the module voltage to a remote fire monitoring and early warning server through a wireless communication module;

[0010] In the remote fire monitoring and early warning server, a module internal-external temperature collaborative representation matrix and a module working state timing representation vector are extracted from the module internal temperature time queue, the module external temperature time queue, the module current time queue, and the module voltage time queue;

[0011] Performing query response encoding on the module working state time series representation vector and the module internal-external temperature collaborative representation matrix to obtain a module state-temperature distribution query response encoding vector;

[0012] Based on the module state-temperature distribution query response coding vector, the danger level of the fire in the module is determined.

[0013] In the fire monitoring method of the lithium-ion capacitor frequency modulation system, a module internal-external temperature collaborative representation matrix and a module working state timing representation vector are extracted from the module internal temperature time queue, the module external temperature time queue, the module current time queue and the module voltage time queue, including: inputting the module internal temperature time queue, the module external temperature time queue, the module current time queue and the module voltage time queue into a sequence encoder based on a forward LSTM model to obtain a module internal temperature timing association feature vector, a module external temperature timing association feature vector, a module current timing association feature vector and a module voltage timing association feature vector; inputting the module internal temperature timing association feature vector and the module external temperature timing association feature vector into a module internal-external temperature collaborative representation network to obtain the module internal-external temperature collaborative representation matrix; inputting the module current timing association feature vector and the module voltage timing association feature vector into a module working state timing representation network to obtain the module working state timing representation vector.

[0014] In the fire monitoring method of the above-mentioned lithium-ion capacitor frequency modulation system, the module internal temperature time series correlation feature vector and the module external temperature time series correlation feature vector are input into the module internal-external temperature collaborative characterization network to obtain the module internal-external temperature collaborative representation matrix, including: calculating the Kronecker product of the transposed vector of the module internal temperature time series correlation feature vector and the module external temperature time series correlation feature vector to obtain the module internal-external temperature collaborative representation matrix.

[0015] In the fire monitoring method of the lithium-ion capacitor frequency modulation system, the module working state timing characterization network is a multi-layer perceptron model including two fully connected layers.

[0016] In the fire monitoring method of the above-mentioned lithium-ion capacitor frequency modulation system, the module working state timing representation vector and the module internal-external temperature collaborative representation matrix are queried and responded to, including: using the significant features of the module internal-external temperature collaborative representation matrix as a prompt template, and performing cross-domain query response encoding based on the prompt template guidance on the module working state timing representation vector and the module internal-external temperature collaborative representation matrix to obtain the module state-temperature distribution query response encoding vector.

[0017] In the fire monitoring method of the above-mentioned lithium-ion capacitor frequency modulation system, the significant features of the module internal-external temperature collaborative representation matrix are used as a prompt template, and the module working state timing representation vector and the module internal-external temperature collaborative representation matrix are cross-domain query response encoded based on the prompt template to obtain the module state-temperature distribution query response encoding vector, including: using a key embedding matrix to linearly embed the module internal-external temperature collaborative representation matrix to obtain a key matrix; extracting the maximum value of each row vector in the key matrix to obtain a significant feature vector of the module internal and external temperature distribution as the prompt template; based on the prompt template, the module working state timing representation vector and the key matrix are cross-domain query encoded based on the attention mechanism to obtain the module state-temperature distribution query response encoding vector.

[0018] In the fire monitoring method of the above-mentioned lithium-ion capacitor frequency modulation system, based on the prompt template, the module working state timing representation vector and the key matrix are cross-domain query encoded based on the attention mechanism to obtain the module state-temperature distribution query response encoding vector, including: using the query embedding matrix and the value embedding matrix to respectively perform linear embedding encoding on the module working state timing representation vector to obtain a value vector and a query vector; the query vector, the value vector, each row vector in the key matrix and the prompt template are input into a heterogeneous converter structure based on template prompt optimization to obtain a set of module state-temperature distribution cross-domain optimized query encoding feature vectors; the positional mean vector of the set of module state-temperature distribution cross-domain optimized query encoding feature vectors is calculated to obtain the module state-temperature distribution query response encoding vector.

[0019] In the fire monitoring method for the lithium-ion capacitor frequency modulation system, the query vector, the value vector, each row vector in the key matrix and the prompt template are input into a heterogeneous converter structure based on template prompt optimization to obtain a set of module state-temperature distribution cross-domain optimization query encoding feature vectors, including: multiplying the query vector by the transposed vector of the row vector and then dividing it by the bi-norm of the prompt template to obtain an attention score matrix; multiplying the attention score matrix by the prompt template after passing it through a softmax function to obtain a template prompt optimization attention weight vector; and calculating the Hadamard product between the template prompt optimization attention weight vector and the value vector to obtain the module state-temperature distribution cross-domain optimization query encoding feature vector.

[0020] In the fire monitoring method of the above-mentioned lithium-ion capacitor frequency modulation system, the danger level of the fire in the module is determined based on the module state-temperature distribution query response coding vector, including: inputting the module state-temperature distribution query response coding vector into a classifier-based fire warning module to obtain a warning result, and the warning result is used to represent the danger level label of the fire in the module.

[0021] According to another aspect of the present application, a fire monitoring device for a lithium-ion capacitor frequency modulation system is provided, comprising:

[0022] A capacitor module operation monitoring module, used to collect a time queue of the internal temperature of the module and a time queue of the external temperature of the module through temperature sensors deployed inside and outside the liquid-cooled lithium-ion capacitor module, and to collect a time queue of the module current and a time queue of the module voltage through current sensors and voltage sensors deployed inside the liquid-cooled lithium-ion capacitor module;

[0023] A data transmission module, used for transmitting the time queue of the module internal temperature, the time queue of the module external temperature, the time queue of the module current and the time queue of the module voltage to a remote fire monitoring and early warning server through a wireless communication module;

[0024] A module operation parameter timing analysis module is used to extract a module internal-external temperature coordination representation matrix and a module working state timing representation vector from a time queue of the module internal temperature, a time queue of the module external temperature, a time queue of the module current, and a time queue of the module voltage in the remote fire monitoring and early warning server;

[0025] A query response encoding module, used for performing query response encoding on the module working state time series representation vector and the module internal-external temperature collaborative representation matrix to obtain a module state-temperature distribution query response encoding vector;

[0026] The fire prediction module is used to determine the danger level of the fire in the module based on the module state-temperature distribution query response coding vector.

[0027] Compared with the prior art, the fire monitoring method and device of the lithium-ion capacitor frequency modulation system provided by the present application adopts artificial intelligence-based data analysis technology to perform time series analysis on the internal and external temperature, current and voltage of the liquid-cooled lithium-ion capacitor module, so as to dig out the time series coordinated distribution pattern of the internal and external temperature of the capacitor module and the time series change trend of the working state of the module, and then by querying and responding to the two, the correlation response relationship between the working state and temperature of the module is captured, and the danger level of the fire that may occur in the module is intelligently estimated. In this way, the limitations of traditional fire monitoring methods can be effectively overcome, and more accurate and timely fire monitoring and early warning can be achieved, thereby improving the safety and reliability of lithium-ion capacitor frequency modulation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0029] Figure 1 Flow chart of a fire monitoring method for a lithium-ion capacitor frequency modulation system according to an embodiment of the present application.

[0030] Figure 2 Schematic diagram of data flow of a fire monitoring method for a lithium-ion capacitor frequency modulation system according to an embodiment of the present application.

[0031] Figure 3 Flow chart of step S130 in the fire monitoring method of the lithium-ion capacitor frequency modulation system according to an embodiment of the present application.

[0032] Figure 4 This is a flowchart of step S140 in the fire monitoring method of the lithium-ion capacitor frequency modulation system according to an embodiment of the present application.

[0033] Figure 5 4 is a block diagram of a fire monitoring device for a lithium-ion capacitor frequency modulation system according to an embodiment of the present application. DETAILED DESCRIPTION

[0034] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0035] Figure 1 Flow chart of a fire monitoring method for a lithium-ion capacitor frequency modulation system according to an embodiment of the present application. Figure 2FIG. 1 is a data flow diagram of a fire monitoring method for a lithium-ion capacitor frequency modulation system according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the fire monitoring method of the lithium-ion capacitor frequency modulation system of the embodiment of the present application, the steps include: S110, collecting the time queue of the internal temperature of the module and the time queue of the external temperature of the module through the temperature sensors deployed inside and outside the liquid-cooled lithium-ion capacitor module, and collecting the time queue of the module current and the time queue of the module voltage through the current sensor and voltage sensor deployed inside the liquid-cooled lithium-ion capacitor module; S120, transmitting the time queue of the internal temperature of the module, the time queue of the external temperature of the module, the time queue of the module current and the time queue of the module voltage to the remote fire control unit through the wireless communication module. S130, in the remote fire monitoring and early warning server, extracting the module internal-external temperature collaborative representation matrix and the module working state timing representation vector from the module internal temperature time queue, the module external temperature time queue, the module current time queue and the module voltage time queue; S140, performing query response encoding on the module working state timing representation vector and the module internal-external temperature collaborative representation matrix to obtain a module state-temperature distribution query response encoding vector; S150, determining the danger level of the fire in the module based on the module state-temperature distribution query response encoding vector.

[0036] In the fire monitoring method of the above-mentioned lithium-ion capacitor frequency modulation system, the step S110 collects the time queue of the internal temperature of the module and the time queue of the external temperature of the module through the temperature sensors deployed inside and outside the liquid-cooled lithium-ion capacitor module, and collects the time queue of the module current and the time queue of the module voltage through the current sensor and voltage sensor deployed inside the liquid-cooled lithium-ion capacitor module. It should be understood that the temperature inside the module is an important indicator reflecting its thermal state. During the charging and discharging process, the module will generate heat due to chemical reactions and resistance. If the heat dissipation is poor, the temperature will rise, which may lead to dangerous situations such as electrolyte decomposition and thermal runaway. The temperature state outside the module reflects the impact of the environment on the heat dissipation of the module. By monitoring the external temperature in real time, it is helpful to understand the impact of environmental factors on the thermal behavior and safety performance of the capacitor. At the same time, current and voltage are important parameters reflecting the working state of the module, and are also directly related to its thermal stability. By monitoring the current and voltage fluctuations of the module in real time, the charging and discharging process of the capacitor, load changes and other information can be effectively revealed, which helps to further evaluate its thermal safety.

[0037] In the fire monitoring method of the above-mentioned lithium-ion capacitor frequency modulation system, the step S120 transmits the time queue of the internal temperature of the module, the time queue of the external temperature of the module, the time queue of the module current and the time queue of the module voltage to the remote fire monitoring and early warning server through the wireless communication module. That is, considering that the direct local processing of the monitoring data of the module may be limited by the limitation of computing resources, there are problems with data processing capacity, storage space and security. Therefore, the present application further transmits the collected internal and external temperature data, current data and voltage data of the module to the remote fire monitoring and early warning server in real time through the wireless communication module, so as to perform more efficient and safer data processing and analysis. Among them, the remote fire monitoring and early warning server has stronger computing power and storage resources, and can perform deep learning and big data analysis on the received data, so as to more accurately identify potential fire risks. In addition, through the remote fire monitoring and early warning server, maintenance personnel can remotely monitor and manage the operating status of the lithium-ion capacitor module, detect abnormal conditions in time and handle them, which greatly improves the maintenance efficiency and safety of the system.

[0038] In the fire monitoring method of the lithium-ion capacitor frequency modulation system, the step S130 extracts the module internal-external temperature coordination representation matrix and the module working state timing representation vector from the module internal temperature time queue, the module external temperature time queue, the module current time queue and the module voltage time queue on the remote fire monitoring and early warning server. Figure 3 FIG. 1 is a flow chart of step S130 in the fire monitoring method of the lithium-ion capacitor frequency modulation system according to an embodiment of the present application. Figure 3 As shown, the step S130 includes: S131, inputting the time queue of the module internal temperature, the time queue of the module external temperature, the time queue of the module current and the time queue of the module voltage into a sequence encoder based on a forward LSTM model to obtain a module internal temperature timing association feature vector, a module external temperature timing association feature vector, a module current timing association feature vector and a module voltage timing association feature vector; S132, inputting the module internal temperature timing association feature vector and the module external temperature timing association feature vector into a module internal-external temperature collaborative representation network to obtain the module internal-external temperature collaborative representation matrix; S133, inputting the module current timing association feature vector and the module voltage timing association feature vector into a module working state timing representation network to obtain the module working state timing representation vector.

[0039] Specifically, in step S131, the time queue of the module internal temperature, the time queue of the module external temperature, the time queue of the module current and the time queue of the module voltage are input into a sequence encoder based on a forward LSTM model to obtain a module internal temperature timing association feature vector, a module external temperature timing association feature vector, a module current timing association feature vector and a module voltage timing association feature vector. It should be understood that the internal and external temperature data, current data and voltage data of the module are time-series and dynamic. Therefore, in order to fully capture the time-series dynamic change rules of the thermal state and working state of the module, in the remote fire monitoring and early warning server, the time series analysis technology is further utilized, and the forward LSTM model is used to perform time series encoding on the time queue of the internal temperature of the module, the time queue of the external temperature of the module, the time queue of the module current and the time queue of the module voltage, so as to utilize the unique gating mechanism and long-term memory capacity of the LSTM model to fully capture the time series dependency in the time queue data, and mine the time series change trends of the internal and external temperature, current and voltage of the module, so as to obtain the time series correlation feature vector of the internal temperature of the module, the time series correlation feature vector of the external temperature of the module, the time series correlation feature vector of the module current and the time series correlation feature vector of the module voltage.

[0040] Specifically, in step S132, the module internal temperature time series correlation feature vector and the module external temperature time series correlation feature vector are input into the module internal-external temperature collaborative characterization network to obtain the module internal-external temperature collaborative representation matrix. That is, considering that the temperature inside and outside the module has a certain interaction and correlation relationship, by collaboratively analyzing the two, the thermal state of the capacitor can be more accurately judged. For example, if the internal temperature of the module continues to rise while the external temperature is relatively stable, it may mean that a local hot spot appears inside the capacitor or the cooling system fails, indicating that the module is at a greater risk of thermal runaway. Therefore, in order to further capture the collaborative correlation pattern between the internal temperature and the external temperature of the module, the present application introduces a module internal-external temperature collaborative characterization network to associate and encode the module internal temperature time series correlation feature vector and the module external temperature time series correlation feature vector to obtain the module internal-external temperature collaborative representation matrix. In a specific example of the present application, the Kronecker product of the transposed vector of the module's internal temperature time series correlation eigenvector and the module's external temperature time series correlation eigenvector is calculated to reveal the nonlinear correlation and interaction between the module's internal and external temperatures, thereby obtaining a more comprehensive and accurate representation of the module's thermal state, namely, the module's internal-external temperature collaborative representation matrix.

[0041] Specifically, in step S133, the module current timing-related feature vector and the module voltage timing-related feature vector are input into the module working state timing characterization network to obtain the module working state timing representation vector. Similarly, considering that the current and voltage of the module are also interrelated, the two together describe the working state and load characteristics of the module. Therefore, in order to further obtain comprehensive module working state timing change information, the present application uses a module working state timing characterization network to perform correlation analysis on the module current timing-related feature vector and the module voltage timing-related feature vector to obtain the timing feature representation of the module working state. In a specific example of the present application, the module working state timing characterization network is a multi-layer perceptron model including two fully connected layers, which performs feature fusion and nonlinear transformation on the module current timing-related feature vector and the module voltage timing-related feature vector to learn the timing correlation relationship between the module current and voltage, thereby generating a module working state timing representation vector containing multi-dimensional information.

[0042] In the fire monitoring method of the above-mentioned lithium-ion capacitor frequency modulation system, the step S140 performs query response encoding on the module working state time series representation vector and the module internal-external temperature collaborative representation matrix to obtain a module state-temperature distribution query response encoding vector. It should be understood that the thermal state of the capacitor module is directly affected by its working state. When the working state of the module changes, its thermal state will also change accordingly. For example, under high-load operation, the module may generate more heat, resulting in an increase in temperature. Conversely, under low-load or no-load conditions, the module generates relatively less heat and the temperature will be relatively low. Therefore, by analyzing the associated response relationship between the working state of the module and its thermal state, it is possible to identify whether the module has abnormal thermal change behavior, thereby predicting and warning potential thermal runaway risks. Based on this, in the technical solution of the present application, in order to more deeply understand the thermal behavior characteristics of the module under different working conditions, the module working state time series representation vector and the module internal-external temperature collaborative representation matrix are further cross-domain query response encoding. It is worth mentioning that in order to improve the effect of query response encoding, in a specific example of the present application, the significant features of the module internal-external temperature collaborative representation matrix are used as prompt templates, and the module working state timing representation vector and the module internal-external temperature collaborative representation matrix are cross-domain query response encoded based on the prompt template guidance to obtain the module state-temperature distribution query response encoding vector.

[0043] Figure 4 FIG. 1 is a flow chart of step S140 in the fire monitoring method of the lithium-ion capacitor frequency modulation system according to an embodiment of the present application. Figure 4As shown, the step S140 includes: S141, using a key embedding matrix to perform linear embedding encoding on the module internal-external temperature collaborative representation matrix to obtain a key matrix; S142, extracting the maximum value of each row vector in the key matrix to obtain a significant feature vector of the module internal and external temperature distribution as the prompt template; S143, based on the prompt template, performing cross-domain query encoding based on the attention mechanism on the module working state timing representation vector and the key matrix to obtain the module state-temperature distribution query response encoding vector.

[0044] More specifically, the step S143 includes: using a query embedding matrix and a value embedding matrix to perform linear embedding encoding on the module working state timing representation vector respectively to obtain a value vector and a query vector; inputting the query vector, the value vector, each row vector in the key matrix and the prompt template into a heterogeneous converter structure optimized based on template prompts to obtain a set of module state-temperature distribution cross-domain optimized query encoding feature vectors; calculating the positional mean vector of the set of module state-temperature distribution cross-domain optimized query encoding feature vectors to obtain the module state-temperature distribution query response encoding vector. Among them, the query vector, the value vector, each row vector in the key matrix and the prompt template are input into a heterogeneous converter structure based on template prompt optimization to obtain a set of module state-temperature distribution cross-domain optimization query encoding feature vectors, including: multiplying the query vector by the transposed vector of the row vector and then dividing it by the bi-norm of the prompt template to obtain an attention score matrix; multiplying the attention score matrix by the prompt template after passing it through a softmax function to obtain a template prompt optimization attention weight vector; calculating the Hadamard product between the template prompt optimization attention weight vector and the value vector to obtain the module state-temperature distribution cross-domain optimization query encoding feature vector.

[0045] Accordingly, the step S140 includes: processing the module internal-external temperature coordination representation matrix and the module working state time series representation vector using the following query encoding formula to obtain the module state-temperature distribution query response encoding vector, wherein the query encoding formula is:

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052]

[0053] in, represents the module internal-external temperature synergy representation matrix, A timing representation vector representing the working state of the module, , and denote the key embedding matrix, query embedding matrix and value embedding matrix respectively, , and Represent different bias terms, represents the key matrix, , , and are the first, second, and and row vectors, The value of is the number of row vectors, represents the maximum value function, represents the prompt template, and denote the query vector and value vector respectively, represents the transpose of a vector, represents the two-norm of a vector, is the normalized exponential function, represents the matrix multiplication operation, represents the Hadamard product, The first one in the set of the module state-temperature distribution cross-domain optimization query encoding feature vectors Module status-temperature distribution cross-domain optimization query encoding feature vector, Represents the module state-temperature distribution query response encoding vector.

[0054] That is, first, the module internal-external temperature collaborative representation matrix is ​​linearly transformed to map it to a new potential feature space and converted into a key matrix to make it suitable for the subsequent query encoding process. Then, the maximum value of each row vector in the key matrix is ​​selected to combine into a significant feature vector of the module internal and external temperature distribution to capture important thermal state feature information and use it as a prompt template to guide the subsequent query encoding process to focus more on key thermal state features to reduce noise interference. Then, the query embedding matrix and the value embedding matrix are used to act on the module working state time series representation vector to generate corresponding query vectors and value vectors, and the query vector, the value vector, each row vector in the key matrix and the prompt template are interactively encoded through a specially designed heterogeneous converter structure. By introducing additional prompt templates in the traditional attention mechanism to automatically adjust the weight allocation strategy, not only the correlation response relationship between the local module thermal state characteristics and the module working state is considered, but also the overall distribution characteristics of the module thermal state are considered, which helps to generate more accurate and relevant query response outputs. Finally, the module working state time series representation vector and the cross-domain query results of each local module thermal state feature are fused by position mean to integrate the global module thermal state query information and generate a module state-temperature distribution query response encoding vector, thereby achieving an in-depth understanding of the module thermal state response characteristics.

[0055] In the fire monitoring method of the above-mentioned lithium-ion capacitor frequency modulation system, the step S150 determines the danger level of the fire in the module based on the module state-temperature distribution query response coding vector. In a specific example of the present application, the implementation method of determining the danger level of the fire in the module based on the module state-temperature distribution query response coding vector is to input the module state-temperature distribution query response coding vector into a fire warning module based on a classifier to obtain a warning result, and the warning result is used to represent the danger level label of the fire in the module. That is, a classifier model is used to construct a fire warning module, so as to utilize the powerful feature learning and classification mapping capabilities of the classifier to perform feature pattern learning on the module state-temperature distribution query response coding vector, so as to identify whether the module has a potential thermal runaway risk based on the module thermal state response pattern information contained in the module state-temperature distribution query response coding vector, and map it to the corresponding danger level label to achieve risk assessment of the fire in the module.

[0056] In particular, in the technical solution of the present application, the module internal temperature timing correlation feature vector, the module external temperature timing correlation feature vector, the module current timing correlation feature vector and the module voltage timing correlation feature vector respectively represent the timing characteristics of the module internal temperature, the module external temperature, the module current and the module voltage. Inputting the module internal temperature timing correlation feature vector and the module external temperature timing correlation feature vector into the module internal-external temperature collaborative characterization network is to use the module internal-external temperature collaborative characterization network to construct the full time domain spatial collaborative distribution of the module internal temperature and the module external temperature, and inputting the module current timing correlation feature vector and the module voltage timing correlation feature vector into the module working state timing characterization network is to use the module working state timing characterization network to capture the timing superposition effect characteristics between the module current and the module voltage. In this way, when the module working state time series representation vector and the module internal-external temperature collaborative representation matrix are query encoded based on the prompt template, the different directions of time domain collaboration and time domain superposition will cause the time-space query encoding representation of the module state-temperature distribution query response encoding vector to have a complex aggregated spatial structure based on the mismatch of the time domain space action dimension, thereby expecting to improve its classification regression convergence and generalization effects under the complex aggregated spatial structure.

[0057] Preferably, in an example of the present application, the module state-temperature distribution query response coding vector is passed through a classifier-based fire warning module to obtain a warning result, including: calculating the absolute value sum of each eigenvalue of the module state-temperature distribution query response coding vector to obtain a first module state-temperature distribution query response coding space structure value, and calculating the square root of the sum of squares of each eigenvalue of the module state-temperature distribution query response coding vector to obtain a second module state-temperature distribution query response coding space structure value; multiplying each eigenvalue of the module state-temperature distribution query response coding vector with the first module state-temperature distribution query response coding space structure value and the second module state-temperature distribution query response coding space structure value to obtain a first module state-temperature distribution query response coding structure reference value and a second module state-temperature distribution query response coding structure reference value corresponding to each eigenvalue; multiplying each eigenvalue of the module state-temperature distribution query response coding vector with the length of the module state-temperature distribution query response coding vector and the square root of the length to obtain a first module state-temperature distribution query response coding structure reference value and a second module state-temperature distribution query response coding structure reference value corresponding to each eigenvalue; multiplying each eigenvalue of the module state-temperature distribution query response coding vector with the length of the module state-temperature distribution query response coding vector and the square root of the length to obtain a second module state-temperature distribution query response coding structure reference value corresponding to each eigenvalue. The first module state-temperature distribution query response coding scale transformation value and the second module state-temperature distribution query response coding scale transformation value corresponding to the first module state-temperature distribution query response; the first module state-temperature distribution query response coding structure reference value is divided by the difference between the first module state-temperature distribution query response coding space structure value and the first module state-temperature distribution query response coding scale transformation value to obtain the first module state-temperature distribution query response coding transformation adjustment value; the second module state-temperature distribution query response coding structure reference value is divided by the difference between the second module state-temperature distribution query response coding space structure value and the second module state-temperature distribution query response coding scale transformation value to obtain the second module state-temperature distribution query response coding transformation adjustment value; the weighted sum of the first module state-temperature distribution query response coding transformation adjustment value and the second module state-temperature distribution query response coding transformation adjustment value is calculated to obtain each eigenvalue of the optimized module state-temperature distribution query response coding vector; the optimized module state-temperature distribution query response coding vector is passed through a fire warning module based on a classifier to obtain a warning result.

[0058] Here, the optimization process of the module state-temperature distribution query response encoding vector can be expressed by the following optimization formula:

[0059]

[0060]

[0061]

[0062]

[0063]

[0064]

[0065]

[0066]

[0067] in, represents the module state-temperature distribution query response encoding vector, The first one representing the module state-temperature distribution query response encoding vector eigenvalues, represents the length of the module state-temperature distribution query response encoding vector, represents the set of real numbers, represents the first module state-temperature distribution query response encoding space structure value, represents the second module state-temperature distribution query response encoding space structure value, Indicates the module state-temperature distribution query response encoding vector The first module state corresponding to the characteristic value - the temperature distribution query response code conversion adjustment value, A first module state-temperature distribution query response code conversion adjustment vector composed of the first module state-temperature distribution query response code conversion adjustment values, Indicates the module state-temperature distribution query response encoding vector The second module state corresponding to the characteristic value - temperature distribution query response code conversion adjustment value, A second module state-temperature distribution query response code conversion adjustment vector composed of each of the second module state-temperature distribution query response code conversion adjustment values, is the weight hyperparameter, Indicates adding by position point. It means point multiplication by position. Represents the optimized module state-temperature distribution query response encoding vector.

[0068] Therefore, in the preferred example, with respect to the spatial structure information of the feature set of the module state-temperature distribution query response encoding vector in the high-dimensional space, the scale-based frame transformation of each eigenvalue of the module state-temperature distribution query response encoding vector is performed by taking the class norm space structured representation of the module state-temperature distribution query response encoding vector as a reference window, and the spatial structure-based frame attention weight adjustment of each eigenvalue of the module state-temperature distribution query response encoding vector is implemented to ensure the invariance of the spatial transformation (translation, scaling and rotation) of the module state-temperature distribution query response encoding vector under the interaction of the feature space, thereby improving the convergence and generalization effect of the classification regression of the feature set of the module state-temperature distribution query response encoding vector under the complex spatial structure representation, and improving the accuracy of the warning result obtained by passing it through the classifier-based fire warning module.

[0069] In summary, the fire monitoring method of the lithium-ion capacitor frequency modulation system according to the embodiment of the present application is explained, which uses artificial intelligence-based data analysis technology to perform time series analysis on the internal and external temperatures, currents and voltages of the liquid-cooled lithium-ion capacitor module, so as to dig out the time series coordinated distribution pattern of the internal and external temperatures of the capacitor module and the time series change trend of the working state of the module, and then by querying and responding to the two, the correlation response relationship between the working state and temperature of the module is captured, and the danger level of the fire that may occur in the module is intelligently estimated. In this way, the limitations of traditional fire monitoring methods can be effectively overcome, and more accurate and timely fire monitoring and early warning can be achieved, thereby improving the safety and reliability of lithium-ion capacitor frequency modulation systems.

[0070] Figure 5 FIG. 1 is a block diagram of a fire monitoring device for a lithium-ion capacitor frequency modulation system according to an embodiment of the present application. Figure 5As shown, according to the fire monitoring device 100 of the lithium-ion capacitor frequency modulation system of the embodiment of the present application, it includes: a capacitor module operation monitoring module 110, which is used to collect the time queue of the internal temperature of the module and the time queue of the external temperature of the module through temperature sensors deployed inside and outside the liquid-cooled lithium-ion capacitor module, and collect the time queue of the module current and the time queue of the module voltage through current sensors and voltage sensors deployed inside the liquid-cooled lithium-ion capacitor module; a data transmission module 120, which is used to transmit the time queue of the internal temperature of the module, the time queue of the external temperature of the module, the time queue of the module current and the time queue of the module voltage to a remote fire monitoring and early warning system through a wireless communication module. server; a module operation parameter timing analysis module 130, used to extract the module internal-external temperature collaborative representation matrix and the module working state timing representation vector from the module internal temperature time queue, the module external temperature time queue, the module current time queue and the module voltage time queue in the remote fire monitoring and early warning server; a query response encoding module 140, used to perform query response encoding on the module working state timing representation vector and the module internal-external temperature collaborative representation matrix to obtain a module state-temperature distribution query response encoding vector; a fire prediction module 150, used to determine the danger level of the fire in the module based on the module state-temperature distribution query response encoding vector.

[0071] Here, those skilled in the art can understand that the specific operations of each module in the fire monitoring device of the lithium-ion capacitor frequency modulation system have been described in the above reference. Figures 1 to 4 The fire monitoring method of the lithium-ion capacitor frequency modulation system has been described in detail, and therefore, its repeated description will be omitted.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A fire monitoring method for a lithium-ion capacitor frequency modulation system, characterized in that: include: The time queue of the internal temperature of the module and the time queue of the external temperature of the module are collected by temperature sensors deployed inside and outside the liquid-cooled lithium-ion capacitor module, and the time queue of the module current and the time queue of the module voltage are collected by current sensors and voltage sensors deployed inside the liquid-cooled lithium-ion capacitor module; Transmitting the time queue of the module internal temperature, the time queue of the module external temperature, the time queue of the module current and the time queue of the module voltage to a remote fire monitoring and early warning server through a wireless communication module; In the remote fire monitoring and early warning server, a module internal-external temperature collaborative representation matrix and a module working state timing representation vector are extracted from the module internal temperature time queue, the module external temperature time queue, the module current time queue, and the module voltage time queue; Performing query response encoding on the module working state time series representation vector and the module internal-external temperature collaborative representation matrix to obtain a module state-temperature distribution query response encoding vector; Determining the danger level of fire in the module based on the module state-temperature distribution query response encoding vector; Among them, query response encoding is performed on the module working state timing representation vector and the module internal-external temperature collaborative representation matrix, including: using the significant features of the module internal-external temperature collaborative representation matrix as a prompt template, and performing cross-domain query response encoding on the module working state timing representation vector and the module internal-external temperature collaborative representation matrix based on the prompt template to obtain the module state-temperature distribution query response encoding vector.

2. The fire monitoring method of the lithium-ion capacitor frequency modulation system according to claim 1 is characterized in that: Extracting a module internal-external temperature coordination representation matrix and a module working state timing representation vector from the module internal temperature time queue, the module external temperature time queue, the module current time queue, and the module voltage time queue, including: Inputting the time queue of the module internal temperature, the time queue of the module external temperature, the time queue of the module current and the time queue of the module voltage into a sequence encoder based on a forward LSTM model to obtain a module internal temperature timing correlation feature vector, a module external temperature timing correlation feature vector, a module current timing correlation feature vector and a module voltage timing correlation feature vector; Inputting the module internal temperature time series correlation feature vector and the module external temperature time series correlation feature vector into the module internal-external temperature collaborative representation network to obtain the module internal-external temperature collaborative representation matrix; The module current timing-related characteristic vector and the module voltage timing-related characteristic vector are input into a module working state timing characterization network to obtain the module working state timing representation vector.

3. The fire monitoring method of the lithium-ion capacitor frequency modulation system according to claim 2 is characterized in that: Inputting the module internal temperature time series correlation feature vector and the module external temperature time series correlation feature vector into the module internal-external temperature collaborative representation network to obtain the module internal-external temperature collaborative representation matrix, including: The Kronecker product of the transposed vector of the module internal temperature time series associated feature vector and the module external temperature time series associated feature vector is calculated to obtain the module internal-external temperature cooperative representation matrix.

4. The fire monitoring method of the lithium-ion capacitor frequency modulation system according to claim 3 is characterized in that: The module working state timing characterization network is a multi-layer perceptron model including two fully connected layers.

5. The fire monitoring method of the lithium-ion capacitor frequency modulation system according to claim 4 is characterized in that: Taking the significant features of the module internal-external temperature collaborative representation matrix as a prompt template, performing cross-domain query response encoding based on the prompt template guidance on the module working state time series representation vector and the module internal-external temperature collaborative representation matrix to obtain the module state-temperature distribution query response encoding vector, including: Using a key embedding matrix to linearly embed the module internal-external temperature synergy representation matrix to obtain a key matrix; Extracting the maximum value of each row vector in the key matrix to obtain a significant characteristic vector of the temperature distribution inside and outside the module as the prompt template; Based on the prompt template, the module working state time series representation vector and the key matrix are subjected to cross-domain query encoding based on the attention mechanism to obtain the module state-temperature distribution query response encoding vector.

6. The fire monitoring method of the lithium-ion capacitor frequency modulation system according to claim 5 is characterized in that: Based on the prompt template, the module working state time series representation vector and the key matrix are cross-domain query encoded based on the attention mechanism to obtain the module state-temperature distribution query response encoding vector, including: Using a query embedding matrix and a value embedding matrix, respectively, linearly embedding the module working state time series representation vector to obtain a value vector and a query vector; Inputting the query vector, the value vector, each row vector in the key matrix and the prompt template into a heterogeneous converter structure optimized based on template prompts to obtain a set of module state-temperature distribution cross-domain optimized query encoding feature vectors; The positional mean vector of the set of the module state-temperature distribution cross-domain optimization query encoding feature vectors is calculated to obtain the module state-temperature distribution query response encoding vector.

7. The fire monitoring method of the lithium-ion capacitor frequency modulation system according to claim 6 is characterized in that: The query vector, the value vector, each row vector in the key matrix and the prompt template are input into a heterogeneous converter structure based on template prompt optimization to obtain a set of module state-temperature distribution cross-domain optimization query encoding feature vectors, including: Multiplying the query vector by the transposed vector of the row vector and dividing by the bi-norm of the hint template to obtain an attention score matrix; The attention score matrix is ​​multiplied by the prompt template after passing it through a softmax function to obtain a template prompt optimized attention weight vector; The Hadamard product between the template prompt optimization attention weight vector and the value vector is calculated to obtain the module state-temperature distribution cross-domain optimization query encoding feature vector.

8. The fire monitoring method for the lithium-ion capacitor frequency modulation system according to claim 7, characterized in that: Determining the danger level of a fire in the module based on the module state-temperature distribution query response encoding vector includes: The module state-temperature distribution query response encoding vector is input into a classifier-based fire warning module to obtain a warning result, and the warning result is used to represent a danger level label of a fire occurring in the module.

9. A fire monitoring device for a lithium-ion capacitor frequency modulation system, characterized in that: include: A capacitor module operation monitoring module, used to collect a time queue of the internal temperature of the module and a time queue of the external temperature of the module through temperature sensors deployed inside and outside the liquid-cooled lithium-ion capacitor module, and to collect a time queue of the module current and a time queue of the module voltage through current sensors and voltage sensors deployed inside the liquid-cooled lithium-ion capacitor module; A data transmission module, used for transmitting the time queue of the module internal temperature, the time queue of the module external temperature, the time queue of the module current and the time queue of the module voltage to a remote fire monitoring and early warning server through a wireless communication module; A module operation parameter timing analysis module is used to extract a module internal-external temperature coordination representation matrix and a module working state timing representation vector from a time queue of the module internal temperature, a time queue of the module external temperature, a time queue of the module current, and a time queue of the module voltage in the remote fire monitoring and early warning server; A query response encoding module, used for performing query response encoding on the module working state time series representation vector and the module internal-external temperature collaborative representation matrix to obtain a module state-temperature distribution query response encoding vector; The fire prediction module is used to determine the danger level of the fire in the module based on the module state-temperature distribution query response coding vector.

Images