Battery state detection method and device based on multi-modal data fusion and medium

Abstract

The invention discloses a battery state detection method and device based on multi-modal data fusion and a medium, and the method comprises the steps: collecting sensor data and Raman spectrum data of an energy storage battery, and carrying out the normalization processing of the sensor data and the Raman spectrum data; performing principal component analysis on the normalized Raman spectrum data, selecting principal components based on a preset cumulative contribution rate threshold, generating dimension-reduced Raman spectrum features, and generating a fusion feature matrix based on the dimension-reduced Raman spectrum features and the normalized sensor data; inputting the fusion feature matrix into a preset prediction model, optimizing model hyper-parameters through cross validation and grid search, and completing training of the prediction model; and inputting the fusion features acquired in real time into the trained prediction model, outputting the charge state, health state and risk coefficient of the energy storage battery, if the risk coefficient exceeds a preset threshold, generating an early warning signal, and triggering an equalization strategy or a cooling strategy of battery management.

Description

Technical Field

[0001] This application relates to the technical field of battery management, and particularly to a battery state detection method, device, and medium based on multimodal data fusion. Background Art

[0002] With the rapid development of clean energy technology, energy storage batteries, as the core components of new energy systems, the accuracy and real-time nature of their state detection directly affect system safety and life management. Current mainstream detection methods mainly rely on sensor data such as voltage and current to construct evaluation models. However, such data can only reflect the external electrical behavior characteristics of the battery and cannot capture the correlation mechanism between the internal chemical state changes of the battery and its macroscopic performance, resulting in a significant increase in the prediction deviation of key parameters such as state of health and state of charge. Especially, the failure risk is prominent under battery aging or extreme working conditions.

[0003] To make up for the deficiency of internal state monitoring, some solutions introduce Raman spectroscopy technology to obtain the microstructure information of battery internal substances. However, Raman spectroscopy data has high dimensions and low signal-to-noise ratio. A single spectral line contains thousands of wavenumber points, and it is difficult to achieve efficient processing using traditional manual analysis or linear dimensionality reduction methods. Moreover, the fusion of sensor data and spectral data often uses simple superposition or weighted average, without considering the spatio-temporal alignment and non-linear coupling relationship of multi-source data, resulting in redundant fusion features and poor interpretability. Summary of the Invention

[0004] Embodiments of this application provide a battery state detection method, device, and medium based on multimodal data fusion to solve the above technical problems.

[0005] On the one hand, embodiments of this application provide a battery state detection method based on multimodal data fusion, including: Collect sensor data and Raman spectroscopy data of an energy storage battery, and perform normalization processing on the sensor data and the Raman spectroscopy data; the sensor data includes voltage, current, temperature, and internal resistance data; Perform principal component analysis on the normalized Raman spectroscopy data to select principal components based on a preset cumulative contribution rate threshold, generate reduced-dimensional Raman spectroscopy features, and generate a fusion feature matrix based on the reduced-dimensional Raman spectroscopy features and the normalized sensor data; Input the fusion feature matrix into a preset prediction model, optimize the model hyperparameters through cross-validation and grid search, and complete the training of the prediction model; Input the fusion features collected in real time into the trained prediction model, output the state of charge, state of health, and risk coefficient of the energy storage battery. If the risk coefficient exceeds a preset threshold, generate a warning signal and trigger the balancing strategy or cooling strategy of battery management.

[0006] In one implementation of the present application, principal component analysis is performed on the normalized Raman spectroscopy data to select principal components based on a preset cumulative contribution rate threshold, and generate Raman spectroscopy features after dimensionality reduction, specifically including: Perform standardization processing on the normalized Raman spectroscopy data so that the mean of the spectral intensities in the Raman spectroscopy data is zero and the standard deviation is one; Calculate the covariance matrix of the standardized Raman spectroscopy data, perform eigenvalue decomposition on the covariance matrix, and determine the cumulative contribution rate of the eigenvalues; Dynamically select the number of principal components according to the cumulative contribution rate, and project the original Raman spectroscopy data onto the selected principal component directions to generate Raman spectroscopy features after dimensionality reduction.

[0007] In one implementation of the present application, a fused feature matrix is generated based on the Raman spectroscopy features after dimensionality reduction and the normalized sensor data, specifically including: Concatenate the Raman spectroscopy features after dimensionality reduction and the normalized sensor data in a fixed order at the head and tail to generate a fused feature vector with a unified dimension; Align and integrate the fused feature vectors with a unified dimension at multiple time points according to the timestamps to form a fused feature matrix with time series.

[0008] In one implementation of the present application, the fused feature matrix is input into a preset prediction model, and the hyperparameters of the model are optimized through cross-validation and grid search to complete the training of the prediction model, specifically including: Divide the input fused feature matrix into several subsets, and use some subsets as the test set and the remaining subsets as the training set for model training; Traverse the preset range of hyperparameter combinations, and calculate the mean value of the mean squared error of the model prediction results under each hyperparameter combination; Select the hyperparameter combination with the smallest mean value of the mean squared error among each hyperparameter combination, and use the hyperparameter combination as the final model parameters of the prediction model to complete the training of the prediction model.

[0009] In one implementation of the present application, after the real-time collected fused features are input into the trained prediction model and the state of charge, health state, and risk coefficient of the energy storage battery are output, the method further includes: Generate a battery state heat map through a visualization interface to display the distribution of the state of charge and health state of the energy storage battery in real time through the battery state heat map; Transmit a warning signal to the battery management system based on the communication protocol to trigger a global operation strategy adjustment instruction; the global operation strategy adjustment instruction includes restricting the charge and discharge power of abnormal battery clusters or switching to standby battery units.

[0010] In one implementation of the present application, triggering the balancing strategy or the cooling strategy of battery management specifically includes: Based on the warning signal, determining the high-risk battery clusters whose risk coefficients exceed the preset threshold, and restricting the charging current of the high-risk battery clusters; Starting the balancing strategy, adjusting the voltage difference between individual batteries through a preset balancing method until the voltage difference drops to the preset safe range; the preset balancing method includes resistor dissipation or energy transfer.

[0011] In one implementation of the present application, triggering the balancing strategy or the cooling strategy of battery management specifically includes: Obtaining the battery temperature of the energy storage battery in real time, and dynamically controlling the coolant flow rate of the liquid cooling system according to the real-time temperature gradient; When the battery temperature exceeds the preset safety threshold, starting the cooling pump, and adjusting the circulation intensity of the coolant in the heat dissipation pipeline according to the preset cooling strategy until the battery temperature returns to within the safety threshold.

[0012] In one implementation of the present application, after triggering the balancing strategy or the cooling strategy of battery management, the method further includes: Monitoring the battery operating parameters in real time during the execution of the balancing strategy or the cooling strategy; Verifying the strategy execution effect based on the operating parameters. If the battery operating parameters after the strategy execution do not return to the preset safe range, dynamically adjusting the balancing current limit value or the coolant flow rate control instruction; Feeding back the adjusted balancing current limit value or the coolant flow rate control instruction to the battery management system through the communication protocol, and looping through the monitoring and adjustment operations until the battery operating parameters are within the preset safe range.

[0013] On the other hand, the embodiment of the present application also provides a battery state detection device based on multi-modal data fusion, and the device includes: At least one processor; And a memory communicatively connected to the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the battery state detection method based on multi-modal data fusion as described above.

[0014] On the other hand, the embodiment of the present application also provides a non-volatile computer storage medium, storing computer-executable instructions, and when the computer-executable instructions are executed, the battery state detection method based on multi-modal data fusion as described above is implemented.

[0015] The embodiments of the present application provide a battery state detection method, device, and medium based on multimodal data fusion, including at least the following beneficial effects: By synchronously collecting sensor data such as voltage, current, temperature, internal resistance, and Raman spectroscopy data, and performing normalization processing on each of them, the blind spot in monitoring the internal chemical state caused by the traditional method relying only on external electrical parameters is overcome; after dimensionality reduction of the Raman spectroscopy data by principal component analysis, the key chemical features are retained, and then fused with the sensor data into a unified feature matrix, realizing cross-modal correlation modeling of the internal microstructure changes and external electrical behavior of the battery, and significantly improving the prediction accuracy of the state of charge, state of health, and risk coefficient; dynamically selecting the number of principal components based on a preset cumulative contribution rate threshold to ensure that the high-dimensional Raman spectroscopy data is compressed to low-dimensional features while retaining effective information, avoiding information loss; generating a fusion feature matrix through time series alignment and fixed-order stitching to solve the problems of dimensional heterogeneity and spatio-temporal asynchrony of multi-source data; through a closed-loop control mechanism of prediction, execution, monitoring, and adjustment, the response lag and failure risk of the traditional open-loop strategy are completely avoided, and potential safety hazards such as thermal runaway are suppressed at the source. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments and descriptions thereof of the present application are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings: Figure 1 It is a schematic flow chart of the battery state detection method based on multimodal data fusion provided by the embodiments of the present application; Figure 2 It is a schematic internal structure diagram of the battery state detection device based on multimodal data fusion provided by the embodiments of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions of the present application will be clearly and completely described below in conjunction with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0018] The technical solutions provided by the embodiments of the present application will be described in detail below in conjunction with the drawings.

[0019] Figure 1 It is a schematic flow chart of the battery state detection method based on multimodal data fusion provided by the embodiments of the present application.

[0020] The implementation of the analysis method involved in the embodiments of this application can be a terminal device or a server, and this application does not impose special restrictions on this. For the convenience of understanding and description, the following embodiments will be described in detail taking the server as an example.

[0021] It should be noted that the server can be a single device or a system composed of multiple devices, that is, a distributed server, and this application does not make specific limitations on this.

[0022] As Figure 1 shown, the battery state detection method based on multi-modal data fusion provided by the embodiments of this application includes: Step 101, collect the sensor data and Raman spectrum data of the energy storage battery, and perform normalization processing on the sensor data and Raman spectrum data.

[0023] It should be noted that the sensor data in the embodiments of this application includes voltage, current, temperature, and internal resistance data.

[0024] In an embodiment of this application, the voltage data is collected by a high-precision voltage sensor, the current data is captured by a Hall effect current sensor, the temperature data is obtained by a temperature sensor attached to the battery surface, and the internal resistance data is dynamically measured by the AC injection method. For example, for 240 single cells in a battery cluster in an energy storage power station, data is collected once every 10 seconds. The Raman spectrum data is collected by a fiber optic probe closely attached to the battery case. The probe uses a laser with a specific wavelength to excite the substances inside the battery and receives the scattered spectrum signal. Specifically, the Raman spectrum is collected by a fiber optic probe with a wavelength of 785 nm and a probe diameter of 200 μm attached to the battery surface. The spectral range covers 200 - 2000 cm wave numbers, and a single collection contains the intensity information of 1024 wave number points. The probe is hermetically connected to the battery case through a quartz window to ensure non-destructive detection.

[0025] It can be understood that in order to eliminate the dimensional difference of multi-source data, it is necessary to perform normalization processing on the sensor data and Raman spectrum data. Exemplarily, the sensor data is mapped to a unified numerical interval according to physical attributes. The voltage value is scaled based on the battery operating voltage range, the current value is normalized according to the charge and discharge current extreme values, and the temperature value is linearly converted according to the safe temperature interval. For example, the voltage value is mapped to [0, 1], the reference range of the voltage value is 3.0 - 4.2 V, the current value is mapped to [0, 1], the reference range of the current value is -50 A to 50 A, the temperature value is mapped to [0, 1], and the reference range of the temperature value is -20 °C to 60 °C. The Raman spectrum data is then normalized by Min-Max to convert the original spectral intensity to a fixed interval.

[0026] Step 102: Perform principal component analysis on the normalized Raman spectral data to select principal components based on a preset cumulative contribution rate threshold, generate the reduced-dimensional Raman spectral features, and generate a fused feature matrix based on the reduced-dimensional Raman spectral features and the normalized sensor data.

[0027] In an embodiment of the present application, principal components analysis (PCA) first standardizes the normalized Raman spectral data so that the mean of the spectral intensities at each wavenumber point is zero and the standard deviation is one. Exemplarily, Z-score standardization is performed on the normalized spectral data so that the mean intensity at each wavenumber point is 0 and the standard deviation is 1. For example, the intensity values at some wavenumber points after standardizing a certain spectral data are [-0.3, 1.2, -0.8].

[0028] Specifically, by calculating the covariance matrix of the standardized data and performing eigenvalue decomposition on it, eigenvectors sorted by variance contribution rate are obtained. Exemplarily, calculate the covariance matrix of the standardized data (dimension 1024×1024), and obtain eigenvalues and eigenvectors through Singular Value Decomposition (SVD).

[0029] According to the preset cumulative contribution rate threshold, dynamically select the number of principal components, project the original high-dimensional spectrum onto the selected principal component directions, and generate low-dimensional Raman features. Sort by eigenvalues from large to small and stop selecting when the cumulative contribution rate reaches 95%. For example, the cumulative contribution rate of the first 30 principal components is 95.2%, and the original spectrum is compressed from 1024 dimensions to a 30-dimensional feature vector. Project the standardized spectral data onto the selected principal component directions to generate low-dimensional features, such as a 30-dimensional vector [0.83, -0.16, 0.48].

[0030] In an embodiment of the present application, feature fusion needs to solve the spatio-temporal alignment problem. Exemplarily, the reduced-dimensional Raman features (such as 30 dimensions) are concatenated with the normalized sensor data (4 dimensions: voltage, current, temperature, internal resistance) in a fixed order at the head and tail to form a 34-dimensional fused feature vector. Exemplarily, the fused feature vector = [Raman feature 1, Raman feature 2,..., Raman feature 30, voltage, current, temperature, internal resistance].

[0031] Furthermore, align and integrate the fused vectors with continuous timestamps to construct a temporal fusion feature matrix. This design significantly reduces redundant information while retaining the correlation characteristics of multi-modal data. For example, align the fused vectors at 5 consecutive time points according to the timestamps to form a 5×34-dimensional temporal fusion matrix.

[0032] Step 103: Input the fused feature matrix into a preset prediction model, optimize the model hyperparameters through cross-validation and grid search, and complete the training of the prediction model.

[0033] In an embodiment of the present application, the prediction model adopts the support vector regression (SVR) algorithm, and its kernel function is the Radial Basis Function (RBF). It can be understood that the RBF kernel solves the complex coupling relationship of multimodal data through nonlinear mapping.

[0034] Divide the fused feature matrix into several subsets, take some subsets as the test set in turn, and the rest as the training set for iterative training, and traverse the penalty coefficient (C) and kernel parameter (γ) in a preset range, and calculate the mean value of the mean square error of the prediction results under each combination. Select the hyperparameter combination with the smallest mean square error as the final model parameter. Exemplarily, this mechanism ensures the generalization ability and robustness of the model under complex working conditions.

[0035] Step 104: Input the fused features collected in real time into the trained prediction model, output the state of charge, state of health, and risk coefficient of the energy storage battery. If the risk coefficient exceeds the preset threshold, generate a warning signal and trigger the balancing strategy or cooling strategy of the battery management.

[0036] The trained SVR model outputs three parameters. The state of charge (SOC) represents the percentage of the remaining battery charge, and the state of health (SOH) reflects the degree of battery capacity attenuation; the risk coefficient is a safety score that comprehensively considers indicators such as sudden changes in internal resistance and abnormal temperature.

[0037] In an embodiment of the present application, after the prediction model outputs the state of charge, state of health, and risk coefficient of the energy storage battery, a visualization interface is generated and the global strategy is adjusted to form a closed-loop link from state monitoring to management decision-making.

[0038] The visualization interface generates a heat map based on the three types of parameters: the state of charge SOC, state of health SOH, and risk coefficient output by the prediction model in real time. It can be understood that the heat map uses the spatial distribution of the battery cluster as the base, and uses color gradient to map the state values. For example, a gradient color scale from dark red (low battery) to dark green (high battery), a gradient color from dark yellow (severe capacity attenuation) to dark blue (good health state), and when the risk coefficient exceeds the preset threshold, the contour of the corresponding battery cluster flashes a red border.

[0039] Exemplarily, on the monitoring large screen of the energy storage power station, the operator can visually identify that a certain battery cluster shows dark red, indicating that the SOC is lower than 20%, and the border flashes, indicating that the risk coefficient > 0.8. At the same time, it shows that its SOH is yellow, indicating that the capacity has decayed to 85%. The heat map converts abstract data into spatial visualization information, providing a direct basis for operation and maintenance decisions.

[0040] Through a standardized communication protocol such as the CAN bus, the warning signal is transmitted to the Battery Monitoring and Management System (BMS). The warning signal includes the battery cluster number, risk level, and recommended strategy code. The risk level is high, medium, or low. The recommended strategy code, such as 01, represents power limit, and 02 represents switching to a standby unit.

[0041] In an embodiment of the present application, the trigger for the battery management balancing strategy is the core protection mechanism for high-risk battery clusters in response to the warning signal. Through dual-level operations of current limiting and voltage balancing, the imbalance problem between individual cells within the battery cluster is solved. It can be understood that if such an imbalance is not suppressed in time, it will cause local overcharging / overdischarging, accelerating battery aging and even leading to thermal runaway.

[0042] Specifically, after the BMS parses the signal, it sends an instruction request to the EMS through the Modbus-TCP protocol to trigger two types of global strategies. When the risk coefficient output by the prediction model exceeds the preset threshold, the battery management system BMS locates the target cluster by parsing the battery cluster number in the warning signal. Exemplarily, in a battery cluster composed of 240 cells, if its risk coefficient is triggered by an abnormal increase in the voltage of a certain cell, the BMS immediately sends a current limiting instruction to the charging circuit to which the cluster belongs. Based on a preset current reduction ratio, such as 50% of the rated current, a PWM modulation signal is generated, and the impedance of the charging circuit is dynamically adjusted through the MOSFET switch circuit to achieve a step-by-step decrease in the charging current.

[0043] The balancing strategy includes two complementary methods, resistive dissipation (passive balancing) and energy transfer (active balancing), and the two are selectively activated according to the hardware configuration and the degree of failure. Resistive dissipation balancing is applicable to scenarios where the voltage deviates slightly, and energy transfer balancing is used for scenarios with severe voltage imbalance or high-efficiency applications.

[0044] The resistor dissipation type of balancing is triggered when the voltage difference between individual cells is at the critical value of the preset safety range, such as when the difference is close to but does not exceed the threshold. A power resistor is connected in parallel to the high-voltage cell, and the excess power is dissipated through Joule heat. The BMS controls the relay to close the bypass resistor circuit of the target cell, and monitors the voltage drop rate of this cell in real time. When the voltage drops back within the cluster average voltage ± allowable deviation, the circuit is disconnected. Exemplarily, in a ternary lithium battery cluster, the voltage of cell A is 0.15V higher than the cluster average. After connecting a 5Ω bypass resistor, it drops within the allowable deviation range after 120 seconds.

[0045] The energy transfer type of balancing is based on a balancing module with a bidirectional DC-DC converter, which realizes the directional transfer of energy from high-voltage cells to low-voltage cells. The BMS identifies the highest and lowest voltage cells as the energy transfer source and target, and starts the DC-DC converter to pump the energy of the high-voltage cell to the low-voltage cell in Buck-Boost mode. The duty cycle of the converter is dynamically adjusted to control the energy transfer rate. It can be understood that this method has almost no energy loss and is suitable for the energy storage power station scenario with long-term floating charge. For example, a lithium iron phosphate battery cluster reduces the maximum voltage difference from 0.25V to 0.05V within 300 seconds through active balancing.

[0046] In an embodiment of the present application, after triggering the balancing strategy or cooling strategy of the battery management, first, the battery operating parameters during the execution of the balancing strategy or cooling strategy will be monitored in real time. The battery operating parameters here cover multiple aspects. For example, when the balancing strategy is executed, the voltage, current of the battery, and the voltage difference between individual cells are all important operating parameters; while when the cooling strategy is executed, the temperature of the battery, the temperature change rate, etc. are the key operating parameters. These operating parameters can be obtained through the sensors built in the battery management system (BMS). The sensors continuously measure various physical quantities of the battery and transmit the measurement data to the processing unit of the BMS. The purpose of obtaining these operating parameters is to comprehensively understand the actual state of the battery during the strategy execution, judge whether the strategy is effective, and whether there is a need for further adjustment. For example, during the execution of the balancing strategy, if it is found that the voltage of a certain individual cell continues to increase compared with other individual cells, it indicates that the current balancing strategy may not achieve the expected effect and needs further analysis and adjustment.

[0047] Next, the execution effect of the strategy is verified based on the monitored operating parameters. Specifically, the currently monitored battery operating parameters are compared with the preset safety range. The preset safety range is set according to the type, specifications, and actual application scenarios of the battery. For example, for a certain lithium battery used in an electric vehicle, its preset safety voltage range may be 3.0V - 4.2V, and the preset safety temperature range may be -20°C - 60°C. If the battery operating parameters after the strategy execution do not return to the preset safety range, it indicates that the current implemented equalization strategy or cooling strategy fails to effectively improve the battery state. At this time, it is necessary to dynamically adjust the equalization current limit value or the coolant flow rate control instruction. The basis for dynamic adjustment is the deviation degree between the monitored operating parameters and the preset safety range. For example, when the battery temperature is too high and the temperature still does not drop within the preset safety temperature range after the cooling strategy is executed, the BMS will appropriately increase the coolant flow rate control instruction according to the temperature deviation to accelerate the cooling speed and reduce the battery temperature.

[0048] Then, the adjusted equalization current limit value or coolant flow rate control instruction is fed back to the battery management system through the communication protocol. The communication protocol is a rule that ensures accurate and reliable data transmission between components of the BMS and between the BMS and external devices. In this embodiment, common communication protocols may include the CAN (Controller Area Network) bus protocol, etc. The processing unit of the BMS will send the adjusted data to the corresponding execution module through the communication interface according to the dynamically adjusted instruction, such as the control module of the equalization circuit or the control module of the cooling system. After receiving the instruction, the execution module will immediately adjust the equalization current or coolant flow rate according to the new instruction. For example, after the control module of the equalization circuit receives the new equalization current limit value, it will adjust the parameters of components such as resistors and capacitors in the circuit to control the magnitude of the equalization current; after the control module of the cooling system receives the new coolant flow rate control instruction, it will adjust the rotation speed of the water pump or the opening degree of the valve to change the coolant flow rate.

[0049] Finally, the monitoring and adjustment operations are executed in a loop until the battery operating parameters are within the preset safety range. This loop process is continuous. The BMS will continuously monitor the battery operating parameters. Once it is found that the parameters exceed the preset safety range, the execution effect of the strategy will be verified again, and dynamic adjustment will be made according to the verification result, and then the adjusted instruction will be fed back to the battery management system. Through this loop monitoring and adjustment method, it can be ensured that the battery always operates in a safe and stable state under various complex working conditions. For example, during the long-term high-speed driving of an electric vehicle, the battery temperature may continue to rise, and the cooling system needs to continuously adjust the coolant flow rate according to the change of the battery temperature to maintain the battery temperature within the preset safety range and ensure the normal performance and life of the battery.

[0050] The above is the method embodiment proposed in this application. Based on the same inventive concept, the embodiments of this application also provide a battery state detection device based on multi-modal data fusion, and its structure is as Figure 2 shown.

[0051] Figure 2 FIG. is the internal structure schematic diagram of the battery state detection device based on multi-modal data fusion provided by the embodiments of this application. As Figure 2 shown, the device includes: At least one processor; And a memory communicatively connected to the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Collect sensor data and Raman spectrum data of the energy storage battery, and perform normalization processing on the sensor data and Raman spectrum data; the sensor data includes voltage, current, temperature and internal resistance data; Perform principal component analysis on the normalized Raman spectrum data to select principal components based on a preset cumulative contribution rate threshold, generate the Raman spectrum features after dimensionality reduction, and generate a fusion feature matrix based on the Raman spectrum features after dimensionality reduction and the normalized sensor data; Input the fusion feature matrix into a preset prediction model, optimize the model hyperparameters through cross-validation and grid search, and complete the training of the prediction model; Input the fusion features collected in real time into the trained prediction model, output the state of charge, health state and risk coefficient of the energy storage battery, and if the risk coefficient exceeds the preset threshold, generate a warning signal and trigger the balancing strategy or cooling strategy of battery management.

[0052] The embodiments of this application also provide a non-volatile computer storage medium storing computer-executable instructions, and when the computer-executable instructions are executed, they can: Collect sensor data and Raman spectrum data of the energy storage battery, and perform normalization processing on the sensor data and Raman spectrum data; the sensor data includes voltage, current, temperature and internal resistance data; Perform principal component analysis on the normalized Raman spectrum data to select principal components based on a preset cumulative contribution rate threshold, generate the Raman spectrum features after dimensionality reduction, and generate a fusion feature matrix based on the Raman spectrum features after dimensionality reduction and the normalized sensor data; Input the fusion feature matrix into a preset prediction model, optimize the model hyperparameters through cross-validation and grid search, and complete the training of the prediction model; Input the fusion features collected in real time into the trained prediction model to output the state of charge, health state, and risk coefficient of the energy storage battery. If the risk coefficient exceeds the preset threshold, an early warning signal is generated, and the balancing strategy or cooling strategy of battery management is triggered.

[0053] Each embodiment in this application is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the key points of each embodiment are the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments.

[0054] The devices and media provided in the embodiments of this application correspond one-to-one with the methods. Therefore, the devices and media also have beneficial technical effects similar to those of their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be elaborated here.

[0055] Those skilled in the art should understand that the embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0056] This application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the specified functions in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0057] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the specified functions in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0058] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide for implementing the steps in the process Figure 1 a process or multiple processes and / or blocks Figure 1 steps of the functions specified in a block or multiple blocks.

[0059] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0060] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM) and / or non-volatile memory such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0061] Computer-readable media includes permanent and non-permanent, removable and non-removable media and can store information by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0062] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0063] The above are only the embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A battery state detection method based on multi-modal data fusion, characterized in that, The method includes: Collecting sensor data and Raman spectrum data of the energy storage battery, and performing normalization processing on the sensor data and the Raman spectrum data; the sensor data includes voltage, current, temperature, and internal resistance data; Performing principal component analysis on the normalized Raman spectrum data to select principal components based on a preset cumulative contribution rate threshold, generating reduced-dimensional Raman spectrum features, and generating a fused feature matrix based on the reduced-dimensional Raman spectrum features and the normalized sensor data; Inputting the fused feature matrix into a preset prediction model, and optimizing the model hyperparameters through cross-validation and grid search to complete the training of the prediction model; Inputting the fused features collected in real time into the trained prediction model, outputting the state of charge, health state, and risk coefficient of the energy storage battery, and if the risk coefficient exceeds the preset threshold, generating a warning signal and triggering the balancing strategy or cooling strategy of the battery management.

2. The battery state detection method based on multi-modal data fusion according to claim 1, wherein Performing principal component analysis on the normalized Raman spectrum data to select principal components based on a preset cumulative contribution rate threshold, generating reduced-dimensional Raman spectrum features, specifically including: Performing standardization processing on the normalized Raman spectrum data so that the mean of the spectral intensity in the Raman spectrum data is zero and the standard deviation is one; Calculating the covariance matrix of the standardized Raman spectrum data, performing eigenvalue decomposition on the covariance matrix, and determining the cumulative contribution rate of the eigenvalues; Dynamically selecting the number of principal components according to the cumulative contribution rate, and projecting the original Raman spectrum data onto the selected principal component directions to generate reduced-dimensional Raman spectrum features.

3. The battery state detection method based on multi-modal data fusion according to claim 1, characterized in that Generating a fused feature matrix based on the reduced-dimensional Raman spectrum features and the normalized sensor data, specifically including: Performing head-to-tail splicing on the reduced-dimensional Raman spectrum features and the normalized sensor data in a fixed order to generate a fused feature vector with a unified dimension; Aligning and integrating the fused feature vectors with a unified dimension at multiple time points according to the timestamps to form a fused feature matrix with time series.

4. The battery state detection method based on multi-modal data fusion according to claim 1, characterized in that, Inputting the fused feature matrix into a preset prediction model, and optimizing the model hyperparameters through cross-validation and grid search to complete the training of the prediction model, specifically including: Dividing the input fused feature matrix into several subsets, and using some subsets as the test set and the remaining subsets as the training set for model training; Traversing the preset range of hyperparameter combinations and calculating the mean of the mean squared errors of the model prediction results under each hyperparameter combination; Selecting the hyperparameter combination with the minimum mean of the mean squared errors among each hyperparameter combination, and using the hyperparameter combination as the final model parameters of the prediction model to complete the training of the prediction model.

5. The battery state detection method based on multi-modal data fusion according to claim 1, characterized in that, After inputting the fused features collected in real time into the trained prediction model and outputting the state of charge, health state, and risk coefficient of the energy storage battery, the method further includes: Generating a battery state heat map through a visualization interface to display the distribution of the state of charge and health state of the energy storage battery in real time through the battery state heat map. Transmit the warning signal to the battery management system based on the communication protocol to trigger a global operation strategy adjustment instruction; the global operation strategy adjustment instruction includes restricting the charging and discharging power of the abnormal battery cluster or switching to the standby battery unit.

6. The battery state detection method based on multi-modal data fusion according to claim 1, characterized in that, Trigger the balancing strategy or cooling strategy of the battery management, specifically including: Based on the warning signal, determine the high-risk battery clusters whose risk coefficients exceed the preset threshold, and limit the charging current of the high-risk battery clusters; Start the balancing strategy, and adjust the voltage difference between the single cells through a preset balancing method until the voltage difference drops to the preset safe range; the preset balancing method includes resistor dissipation or energy transfer.

7. The battery state detection method based on multi-modal data fusion according to claim 1, wherein, Trigger the balancing strategy or cooling strategy of the battery management, specifically including: Obtain the battery temperature of the energy storage battery in real time, and dynamically control the coolant flow rate of the liquid cooling system according to the real-time temperature gradient; When the battery temperature exceeds the preset safety threshold, start the cooling pump, and adjust the circulation intensity of the coolant in the heat dissipation pipeline according to the preset cooling strategy until the battery temperature returns to within the safety threshold.

8. The battery state detection method based on multi-modal data fusion according to claim 1, wherein, After triggering the balancing strategy or cooling strategy of the battery management, the method further includes: Monitor the battery operation parameters in real time during the execution of the balancing strategy or cooling strategy; Verify the strategy execution effect based on the operation parameters. If the battery operation parameters after the strategy execution do not return to the preset safe range, dynamically adjust the balanced current limit value or the coolant flow rate control instruction; Feedback the adjusted balanced current limit value or the coolant flow rate control instruction to the battery management system through the communication protocol, and loop through the monitoring and adjustment operations until the battery operation parameters are within the preset safe range.

9. A battery state detection device based on multi-modal data fusion, characterized in that, The device includes: At least one processor; And a memory communicatively connected to the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the battery state detection method based on multi-modal data fusion according to any one of claims 1-8.

10. A non-volatile computer storage medium stores computer-executable instructions, characterized in that, When the computer-executable instructions are executed, the battery state detection method based on multi-modal data fusion according to any one of claims 1-8 is implemented.

Images