Maintenance-free lead-acid battery charging and discharging self-adaptive regulation method and system based on internet of things

Abstract

The application belongs to the technical field of electric power and electricity, and relates to a maintenance-free lead-acid battery charging and discharging self-adaptive regulation and control method and system based on Internet of Things. The method collects multi-dimensional characteristic parameters such as voltage ripple, current fluctuation, temperature gradient and internal resistance change rate in real time through multi-dimensional sensors, constructs a fault characteristic vector and inputs the pre-trained comprehensive fault prediction model to obtain the prediction probability of sulfuration, plate softening and low liquid level. When the probability is higher than a threshold value, the adaptive repair pulse module, the power reduction control module and the automatic liquid level compensation module are triggered to perform self-healing repair. Meanwhile, the model is iteratively optimized based on the prediction result and the repair effect. The application realizes the closed-loop control of early prediction, automatic repair and performance recovery of the battery fault, significantly prolongs the service life of the battery, reduces the maintenance cost, and is suitable for communication base station standby power supply, photovoltaic energy storage and other scenes.

Description

Technical Field

[0001] This invention belongs to the field of power and electrical technology, specifically relating to an adaptive control method and system for charging and discharging of maintenance-free lead-acid batteries based on the Internet of Things. Background Technology

[0002] During long-term use, lead-acid batteries will gradually accumulate faults such as sulfation, plate softening, and electrolyte level drop, leading to battery performance degradation or even damage. Although existing battery charge and discharge control systems have the ability to monitor single parameters such as voltage, current, and temperature, single parameter monitoring is difficult to fully capture the early evolution characteristics of faults and cannot make reliable predictions in the fault symptom stage, resulting in delayed maintenance intervention and making it difficult to support truly maintenance-free operation.

[0003] Common lead-acid battery maintenance solutions primarily employ a strategy of periodic inspection combined with passive replacement. While some improved solutions introduce online testing or cyclic parameter recording and analysis functions, they essentially remain reactive measures, lacking predictive fault diagnosis capabilities based on multi-dimensional feature fusion. Because fault occurrence cannot be predicted in advance, single-parameter monitoring struggles to establish a mapping relationship between fault feature vectors and predictive models. This results in significant lag in the system's response to progressive faults such as gradual sulfide formation, slow plate degradation, and continuous electrolyte evaporation. Consequently, batteries are only forced to be maintained or replaced after the fault has fully developed, severely hindering the achievement of maintenance-free goals. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an Internet of Things-based adaptive regulation method and system for charging and discharging of maintenance-free lead-acid batteries, which can effectively solve the problems mentioned in the background.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present invention provides an adaptive regulation method for charging and discharging of a maintenance-free lead-acid battery based on the Internet of Things, the method comprising the following steps:

[0007] By using a multi-dimensional sensor module to collect real-time voltage ripple signals, current fluctuation signals, temperature gradient distribution data, and internal resistance change rate data of lead-acid batteries, multi-dimensional characteristic parameters characterizing the battery's operating status are obtained.

[0008] Based on the multidimensional feature parameters, a fault feature vector is constructed, and the fault feature vector is input into a pre-trained comprehensive fault prediction model to obtain the predicted probability of sulfidation fault, the predicted probability of plate softening fault, and the predicted probability of low liquid level fault.

[0009] When the probability of each of the aforementioned fault predictions exceeds the corresponding preset threshold, the adaptive repair pulse module is triggered to perform sulfidation fault repair, the power descent control module is triggered to perform discharge depth limitation, and the automatic liquid level compensation module is triggered to perform electrolyte replenishment.

[0010] Based on the fault prediction results, repair execution status, and real-time operating parameters of the battery, the prediction accuracy of the comprehensive fault prediction model is evaluated, and the model parameters of the comprehensive fault prediction model are iteratively optimized according to the evaluation results.

[0011] Furthermore, the comprehensive fault prediction model includes a feature preprocessing layer, a feature fusion layer, a feature extraction layer, and a classification output layer connected in sequence;

[0012] The feature fusion layer fuses the standardized feature vectors based on a weighted fusion algorithm. The weight coefficients in the weighted fusion algorithm are determined based on historical fault sample data during the model training phase. Different combinations of weight coefficients correspond to different fault type sensitivities.

[0013] The feature extraction layer employs a deep neural network containing multiple stacked fully connected hidden layers, which outputs intermediate feature representations after multi-level nonlinear transformation of the fused feature vectors.

[0014] The classification output layer calculates the predicted probabilities of sulfidation fault, plate softening fault, and low liquid level fault, which are all within the range of 0 to 1, based on the intermediate feature representation.

[0015] Furthermore, obtaining the multidimensional feature parameters includes:

[0016] Separate the AC ripple component from the voltage ripple signal, extract the fundamental frequency, harmonic amplitude, and energy ratio of each harmonic to the fundamental frequency, and calculate the ripple waveform distortion rate.

[0017] The amplitude, frequency, and waveform distortion rate of the AC fluctuation component are extracted from the current fluctuation signal.

[0018] Based on the multi-point temperature data collected by the temperature sensor array, the three-dimensional temperature field inside the battery is reconstructed by interpolation to obtain the temperature gradient of the plate area, the temperature gradient of the electrolyte area, and the temperature gradient of the shell surface.

[0019] For the internal resistance change rate data, by applying a low-frequency AC excitation signal to the battery circuit, the real part of the dynamic complex impedance is calculated as the ohmic internal resistance, and the absolute value of the internal resistance, the rate of change of the internal resistance, and the acceleration of the change of the internal resistance are extracted.

[0020] Furthermore, the process of the adaptive repair pulse module performing sulfurization fault repair includes:

[0021] The degree of sulfide formation is determined based on the predicted probability of sulfidation failure.

[0022] Based on the degree of sulfide formation, the corresponding pulse frequency, pulse amplitude, and pulse duty cycle are retrieved from a preset pulse parameter library. The pulse frequency is adjustable from 0.5 kHz to 5 kHz, the pulse amplitude is adjustable from 0.05 to 0.3 times the rated voltage of the battery, and the duty cycle is adjustable from 5% to 40%.

[0023] The corresponding pulse sequence signal is generated, amplified by power, and applied between the positive and negative plates of the battery through the output isolation unit;

[0024] During pulse application, the battery's terminal voltage and internal resistance are continuously monitored. When the internal resistance drops below 95% of the value before pulse charging starts and the terminal voltage recovers to above 98% of the battery's rated voltage, pulse charging is stopped.

[0025] Furthermore, the process by which the power descent control module performs the depth of discharge limitation includes:

[0026] The degree of plate softening is determined based on the predicted probability of plate softening failure.

[0027] According to the degree of softening of the electrode plate, the corresponding target discharge depth limit is retrieved from the preset discharge depth mapping table, wherein the mild softening level corresponds to the 80% discharge depth limit, the moderate softening level corresponds to the 60% discharge depth limit, and the severe softening level corresponds to the 40% discharge depth limit.

[0028] Obtain the target discharge depth parameter from the current discharge demand command. When the target discharge depth parameter is greater than the target discharge depth limit, forcibly adjust the target discharge depth parameter to the target discharge depth limit and then output the discharge control command.

[0029] Furthermore, the process of the automatic liquid level compensation module performing electrolyte replenishment includes:

[0030] The electrolyte level is monitored in real time by a liquid level sensor. When the current liquid level is lower than a preset liquid level threshold, the amount of distilled water that needs to be added is calculated. The amount of distilled water is the difference between the total capacity of the standard electrolyte of the battery and the actual capacity of the electrolyte corresponding to the current liquid level.

[0031] The solenoid valve group is opened to allow distilled water in the distilled water storage tank to be injected into the electrolyte chamber of the battery through the pipeline system, and the injection flow rate is regulated by the flow control valve.

[0032] When the electrolyte level reaches 95% of the total capacity of the standard electrolyte in the battery, close the solenoid valve assembly and stop adding water.

[0033] Furthermore, the evaluation of the prediction accuracy of the comprehensive fault prediction model and the iterative optimization of the model parameters based on the evaluation results include:

[0034] After each fault prediction and repair execution cycle, the actual fault type is compared with the fault prediction result to calculate the prediction accuracy.

[0035] The battery capacity recovery rate, internal resistance recovery rate, and terminal voltage recovery rate after the repair are compared with the preset repair effect threshold to calculate the repair effect evaluation index.

[0036] The comprehensive evaluation score is calculated based on a weighted average of the prediction accuracy and the repair effect evaluation indicators.

[0037] When the comprehensive evaluation score is lower than the preset optimization threshold, the parameters of the comprehensive fault prediction model are iteratively optimized using an incremental learning algorithm. Incremental learning is triggered when the newly collected fault sample data accumulates to a preset number threshold, and the learning rate of incremental learning is set to 10% of the learning rate in the initial training phase.

[0038] Furthermore, the training process of the comprehensive fault prediction model includes:

[0039] Collect multidimensional historical feature data and corresponding historical fault labels of historical sample batteries. The historical fault labels are three-dimensional binary vectors, which correspond to sulfation fault, plate softening fault and low liquid level fault, respectively.

[0040] The multidimensional historical feature data is input into the comprehensive fault prediction model to obtain the predicted probability of each fault type.

[0041] With the goal of minimizing the cross-entropy loss between the predicted probability and the historical fault label, the model parameters are updated using the backpropagation algorithm and the adaptive moment estimation optimization algorithm, and the training is iterated until the loss converges.

[0042] Furthermore, the multidimensional feature parameters are transmitted to the cloud server via an Internet of Things (IoT) communication network, and the comprehensive fault prediction model runs on the cloud server.

[0043] The cloud server generates a repair instruction based on the output of the comprehensive fault prediction model and sends the repair instruction to the local controller, which then controls the adaptive repair pulse module, power descent control module, and automatic liquid level compensation module to perform corresponding actions.

[0044] On the other hand, the present invention provides an Internet of Things-based maintenance-free lead-acid battery charge and discharge adaptive control system, comprising:

[0045] The multi-dimensional sensor module is used to collect voltage ripple signals, current fluctuation signals, temperature gradient distribution data, and internal resistance change rate data of lead-acid batteries in real time, so as to obtain multi-dimensional characteristic parameters that characterize the battery's operating status.

[0046] The fault prediction module has a built-in pre-trained comprehensive fault prediction model, which is used to receive the multi-dimensional feature parameters and construct the fault feature vector, and output the probability of sulfurization fault prediction, the probability of plate softening fault prediction, and the probability of low liquid level fault prediction.

[0047] An adaptive repair pulse module is used to perform low-frequency pulse charging repair when the predicted probability of the sulfidation fault exceeds a preset threshold.

[0048] The power derating control module is used to limit the depth of discharge of the battery when the predicted probability of plate softening failure exceeds a preset threshold.

[0049] An automatic liquid level compensation module is used to replenish distilled water to restore the electrolyte level when the probability of a low liquid level fault exceeds a preset threshold.

[0050] The model optimization module is used to evaluate the prediction accuracy of the comprehensive fault prediction model based on the fault prediction results, repair execution status, and real-time operating parameters of the battery, and to iteratively optimize the model parameters based on the evaluation results.

[0051] Compared with the prior art, the present invention has the following beneficial technical effects:

[0052] 1. The IoT-based adaptive charging and discharging control method for maintenance-free lead-acid batteries provided by this invention firstly acquires voltage ripple signals, current fluctuation signals, temperature gradient distribution data, and internal resistance change rate data of the lead-acid battery in real time through multi-dimensional sensors to obtain multi-dimensional feature parameters characterizing the battery's operating state. Then, a fault feature vector is constructed based on the multi-dimensional feature parameters, and the fault feature vector is input into a pre-trained comprehensive fault prediction model to obtain the predicted probabilities of sulfation faults, plate softening faults, and low liquid level faults. Further, when each of the predicted fault probabilities is detected to exceed its corresponding preset threshold, the adaptive repair pulse module, power descent control module, and automatic liquid level compensation module are triggered to execute corresponding self-healing repair actions. Finally, based on the fault prediction results, repair execution status, and real-time operating parameters of the battery, the prediction accuracy of the comprehensive fault prediction model is evaluated, and the model parameters of the comprehensive fault prediction model are iteratively optimized according to the evaluation results.

[0053] 2. This invention overcomes the limitations of existing technologies that only monitor single parameters or passively handle faults by constructing a comprehensive fault prediction model based on multi-parameter fusion. It achieves predictive diagnosis of early faults such as sulfation, plate softening, and low liquid level. By integrating and managing multiple independent functions such as fault prediction, adaptive repair pulse, power descent limiting, and automatic liquid level compensation, a closed-loop control process of detection, prediction, and repair is constructed, achieving truly maintenance-free operation. This invention enables lead-acid batteries to automatically complete early fault prediction, self-healing repair, and performance recovery without manual intervention, significantly extending battery life, reducing maintenance costs, and improving the operational reliability and availability of battery energy storage systems. Attached Figure Description

[0054] Figure 1 A schematic diagram of the overall scheme for an IoT-based adaptive control method for charging and discharging of maintenance-free lead-acid batteries;

[0055] Figure 2 This is a schematic diagram illustrating the core principle of a comprehensive fault prediction model based on multi-parameter fusion.

[0056] Figure 3 This is a schematic diagram of a closed-loop control process based on detection, prediction, and repair.

[0057] Figure 4 A schematic diagram of the logic flow for multi-dimensional operational status data acquisition and feature extraction of lead-acid batteries;

[0058] Figure 5 A schematic diagram of the adaptive repair pulse generation and sulfidation fault repair control logic flow;

[0059] Figure 6 This is a schematic diagram of the automatic liquid level compensation logic flow. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 6 Specific embodiments are provided to further illustrate the present invention in detail.

[0061] Example:

[0062] like Figure 1 As shown in the figure, this embodiment provides an IoT-based adaptive control method for the charging and discharging of maintenance-free lead-acid batteries. This method comprises multiple collaborative technical modules that interact with each other through standardized data interfaces, forming a complete closed-loop control architecture for detection, prediction, and repair. The specific structure, function, and collaborative relationships of each technical module are described in detail below.

[0063] Firstly, the multi-dimensional sensor module involved in this embodiment of the invention undertakes the function of real-time operation status data acquisition. Specifically, this module includes a voltage sensor subunit, a current sensor subunit, a temperature sensor array, and an internal resistance measurement circuit subunit, totaling four subunits.

[0064] The voltage sensor subunit is electrically coupled to the positive and negative terminals of the lead-acid battery via electrical connection lines, continuously acquiring the terminal voltage signal across the battery. The terminal voltage signal contains a DC component reflecting the battery's state of charge, and is also superimposed with an AC ripple component related to the internal electrochemical processes. The amplitude, frequency, and waveform distortion rate of the AC ripple component are all correlated with the smoothness of the internal electrochemical reaction process of the battery.

[0065] The current sensor subunit is installed in the main charging and discharging circuit of the battery to collect the current signal flowing through the battery. The current signal also contains a DC component and an AC fluctuation component caused by the charging and discharging strategy and internal state fluctuations. The amplitude, frequency, and waveform distortion rate of the AC fluctuation component in the current signal can characterize the uniformity of the internal chemical reaction of the battery during charging and discharging.

[0066] The temperature sensor array consists of multiple distributed temperature sensing elements; in this embodiment, the number of temperature sensing elements is 6 to 12. These sensing elements are attached to different heat-sensitive areas of the battery casing, including the surface of the casing sidewalls, near the positive and negative terminals, and the connection strip area between adjacent individual cells. The multi-point temperature data collected by the temperature sensor array is transmitted to the central processing unit via an internal data bus. The central processing unit executes a temperature interpolation algorithm to calculate and reconstruct the temperature gradient distribution inside the battery. In this embodiment, the temperature interpolation algorithm uses an inverse distance weighted interpolation method, where the weighting coefficient of each temperature sensing element's measurement value is inversely proportional to the spatial distance from the element to the location to be determined. The temperature gradient distribution data includes the temperature difference between the active material region of the electrode plates and the electrolyte region, the temperature difference between the upper and lower layers of the electrolyte in the vertical direction, and the horizontal temperature difference between different directional regions of the casing.

[0067] The internal resistance measurement circuit subunit integrates a signal source and a response signal acquisition unit. During measurement, the subunit applies a low-frequency AC excitation signal to the battery circuit. The frequency of the excitation signal is set to 0.01 Hz to 10 Hz. This frequency band effectively penetrates the electrochemical double-layer capacitance effect of the battery, allowing the measurement signal to primarily reflect the ohmic impedance characteristics of the battery itself. The internal resistance measurement circuit subunit acquires the battery's response current to the excitation signal. Based on the amplitude ratio and phase difference between the excitation voltage and the response current, the dynamic complex impedance of the battery is calculated. The calculation formula is: ,in, This refers to dynamic complex impedance, measured in ohms. The applied excitation voltage phasor, in volts; The acquired response current phasor, in amperes. The real part of the dynamic complex impedance. By taking dynamic complex impedance The real part is obtained as follows: ,in, This is the real part of the dynamic complex impedance, in ohms, which corresponds to the ohmic internal resistance of the battery. The operator for taking the real part of a complex number. The real part of a dynamic complex impedance. As a key indicator characterizing the internal ohmic properties of a battery.

[0068] The various raw electrical signals acquired by the multi-dimensional sensor module are transmitted to the central processing unit and then transformed into structured feature parameters that can be directly used by the fault prediction module through a feature extraction process. The feature extraction process is executed by the signal processing unit, which contains multiple parallel signal analysis channels.

[0069] In the voltage signal analysis channel, the AC ripple component is separated from the original terminal voltage signal using a bandpass filter bank. The passband range of the bandpass filter bank is set to 50 Hz to 500 Hz. The separated ripple component is then processed by a Fast Fourier Transform (FFT) to extract the fundamental frequency, harmonic amplitudes, and the energy ratio of each harmonic to the fundamental frequency. The ripple waveform distortion rate is obtained by calculating the ratio of the total energy of all harmonic components (excluding the fundamental frequency) to the fundamental frequency energy.

[0070] The current signal analysis channel adopts a similar processing architecture to the voltage signal analysis channel, extracting the amplitude, frequency, and waveform distortion rate of the AC fluctuation component from the original current signal.

[0071] The temperature signal analysis channel organizes the multi-point temperature data collected by the temperature sensor array into a two-dimensional temperature distribution matrix, and then uses an interpolation algorithm to calculate the three-dimensional temperature field inside the battery. In this embodiment, the interpolation algorithm uses the radial basis function interpolation method, constructing the interpolation function using the spatial coordinates and measured values ​​of each temperature sensing element as sample points. The calculation results of the three-dimensional temperature field are output as the temperature gradient of the electrode area, the temperature gradient of the electrolyte area, and the temperature gradient of the casing surface.

[0072] The internal resistance signal analysis channel performs frequency domain analysis on the dynamic complex impedance measurement data and extracts three key feature values: the absolute value of the dynamic internal resistance; the rate of change of the internal resistance with measurement time, which is obtained by performing a first-order difference on the time series of the absolute value of the internal resistance; and the acceleration of the change of the internal resistance, i.e., the rate of change of the rate of change, which is obtained by performing a first-order difference on the time series of the rate of change of the internal resistance.

[0073] The four signal analysis channels described above extract features from four dimensions: voltage ripple, current fluctuation, temperature gradient distribution, and internal resistance change rate. These four types of feature parameters together constitute a multi-dimensional feature parameter set, which is then encapsulated in a structured data format and transmitted to the subsequent fault prediction module.

[0074] The core functional unit of the fault prediction module in this embodiment is the comprehensive fault prediction model. The comprehensive fault prediction model is logically divided into four layers: feature preprocessing layer, feature fusion layer, feature extraction layer, and classification output layer.

[0075] The feature preprocessing layer is responsible for data validation and format standardization of the input multidimensional feature parameter set. The data validation process checks for outliers or missing values ​​in each feature parameter. When an outlier is detected, it is filled with the historical average of that feature parameter; when a missing value is detected, it is filled with a preset default value. The format standardization process maps the numerical range of each feature parameter to a standard numerical range to eliminate the impact of differences in units and numerical magnitudes between different feature parameters on subsequent processing.

[0076] The feature fusion layer performs multi-dimensional feature fusion processing on the standardized feature vectors based on a weighted fusion algorithm. The weight coefficients in the weighted fusion algorithm are determined based on historical fault sample data during the model training phase; different combinations of weight coefficients correspond to different fault type sensitivities. The dimension of the fused feature vector remains consistent with the dimension of the original feature vector.

[0077] The feature extraction layer is implemented using a deep neural network architecture. The deep neural network consists of multiple stacked fully connected hidden layers, each containing several neurons. The fused feature vector input to the feature extraction layer undergoes multiple levels of non-linear transformations within each fully connected hidden layer, gradually being abstracted into higher-level feature representations. The feature extraction layer ultimately outputs an intermediate feature representation that captures the complex interaction patterns between the multi-dimensional original features.

[0078] The classification output layer calculates the predicted probability for each preset fault type based on intermediate feature representations. The preset fault types include three categories: sulfation fault, plate softening fault, and low liquid level fault. For each fault type, the classification output layer outputs a probability value between 0 and 1, representing the confidence level that the battery's current state belongs to that fault type.

[0079] The aforementioned feature preprocessing layer, feature fusion layer, feature extraction layer, and classification output layer together constitute the complete reasoning path of the comprehensive fault prediction model. After being processed sequentially at each level, the original multidimensional feature parameters are finally transformed into the prediction probability of each fault type.

[0080] The comprehensive fault prediction model needs to be trained offline before practical application. The training dataset is constructed as follows: multi-dimensional historical feature data of historical sample batteries are collected, and each sample is labeled with a corresponding historical fault label. The format of the historical fault label is a three-dimensional binary vector. The three elements of the vector correspond to sulfation fault, plate softening fault, and low liquid level fault, respectively. A value of 1 indicates that the sample has the corresponding type of fault, and a value of 0 indicates that the corresponding type of fault does not exist.

[0081] During training, multi-dimensional historical feature data is input into the comprehensive fault prediction model, and the model outputs the predicted probability of each fault type. The training objective is to minimize the difference between the predicted probability and the historical fault label, which is quantified using the cross-entropy loss function. (Cross-entropy loss function) The calculation formula is: ,in, This is an index for the fault type, with values ​​of 1, 2, and 3, corresponding to sulfidation fault, plate softening fault, and low liquid level fault, respectively. For the first The true label for the fault, with a value of 0 or 1; The model predicts the first The probability of each type of failure ranges from 0 to 1. Based on the cross-entropy loss function. The values ​​are calculated using the backpropagation algorithm to determine the gradients of the parameters in each layer of the model, and then the adaptive moment estimation optimization algorithm is used to update the model parameters based on these gradients. This training process is iteratively executed until the cross-entropy loss function is reached. When the value converges to below the preset loss threshold, or the number of iterations reaches the preset maximum number of training rounds, the training of the comprehensive fault prediction model is completed.

[0082] The predicted probabilities of each fault type output by the comprehensive fault prediction model need to be determined by a decision threshold before being converted into the final fault prediction result. The decision threshold determination process compares the predicted probability of each fault type with its corresponding preset decision threshold one by one. When the predicted probability of a certain fault type exceeds its corresponding preset decision threshold, the decision result indicates that the fault type is in a warning state.

[0083] The decision threshold corresponding to a sulfidation fault is called the first sulfidation threshold, the decision threshold corresponding to a plate softening fault is called the first plate softening threshold, and the decision threshold corresponding to a low liquid level fault is called the first liquid level threshold. The specific values ​​of the first sulfidation threshold, the first plate softening threshold, and the first liquid level threshold are configured during the system initialization phase according to the reliability requirements and false alarm tolerance of the application scenario.

[0084] In this embodiment, the values ​​of the first sulfidation threshold, the first electrode softening threshold, and the first liquid level threshold are all set to 0.7.

[0085] When the predicted probability of a sulfidation fault exceeds the first sulfidation threshold, the fault prediction module generates a sulfidation fault warning signal and transmits it to the adaptive repair pulse module. When the predicted probability of a plate softening fault exceeds the first plate softening threshold, the fault prediction module generates a plate softening fault warning signal and transmits it to the power sag control module.

[0086] When the predicted probability of a low liquid level fault exceeds the first liquid level threshold, the fault prediction module generates a low liquid level fault warning signal and transmits the low liquid level fault warning signal to the automatic liquid level compensation module.

[0087] Through the above decision threshold judgment process, the continuous probability value output by the comprehensive fault prediction model is converted into discrete early warning signals, and each early warning signal triggers the corresponding repair module to perform self-healing repair actions.

[0088] For the adaptive repair pulse module, after receiving the sulfation fault warning signal, a low-frequency pulse charging process is executed to remove the sulfides generated on the surface of the lead-acid battery plates.

[0089] The hardware architecture of the adaptive repair pulse module consists of three parts: a signal generation unit, a power amplification unit, and an output isolation unit.

[0090] The signal generation unit is implemented by a programmable logic device or a digital signal processor. Based on received control commands, the signal generation unit generates a pulse sequence signal with adjustable frequency, adjustable amplitude, and adjustable duty cycle. The frequency adjustment range of the pulse sequence signal covers the low-frequency range; in this embodiment, the low-frequency range is specifically set to 0.5 kHz to 5 kHz. The lower limit frequency of this band ensures that the pulse can act on the decomposition process of sulfide crystals. The amplitude adjustment range of the pulse sequence signal is set to 0.05 to 0.3 times the rated voltage of the battery, and the duty cycle adjustment range is set to 5% to 40%.

[0091] The power amplifier unit amplifies the pulse sequence signal output from the signal generation unit, enabling it to reach a power level sufficient to charge lead-acid batteries. The output power of the power amplifier unit is adjustable from 10 W to 500 W, covering the charging needs of batteries with different capacities.

[0092] The output isolation unit is located between the power amplifier unit and the main charging / discharging circuit. The output isolation unit employs transformer isolation or optocoupler isolation technology to achieve electrical isolation, preventing high-voltage pulse signals from interfering with or damaging other control circuits in the system.

[0093] The aforementioned signal generation unit, power amplification unit, and output isolation unit are cascaded in sequence to jointly complete the conversion from control commands to pulse charging signals safely applied to the battery terminals.

[0094] When performing sulfidation fault repair, the adaptive repair pulse module first determines the sulfide formation level based on the predicted probability of the sulfidation fault. A preset mapping relationship exists between the predicted probability of the sulfidation fault and the sulfide formation level. In this embodiment, the preset mapping relationship is as follows: a predicted probability of sulfidation fault greater than or equal to 0.5 and less than 0.7 corresponds to a light sulfidation level; greater than or equal to 0.7 and less than 0.9 corresponds to a moderate sulfidation level; and greater than or equal to 0.9 corresponds to a severe sulfidation level.

[0095] The degree of sulfide formation is classified into three levels: mild sulfidation, moderate sulfidation, and severe sulfidation. Based on the degree of sulfide formation, the adaptive repair pulse module retrieves the corresponding pulse frequency, pulse amplitude, and pulse duty cycle parameter combinations from a preset pulse parameter library. In this embodiment, the specific contents of the pulse parameter library are as follows: Mild sulfidation corresponds to a pulse frequency of 5 kHz, a pulse amplitude of 0.05 times the battery's rated voltage, and a duty cycle of 10%; moderate sulfidation corresponds to a pulse frequency of 2 kHz, a pulse amplitude of 0.15 times the battery's rated voltage, and a duty cycle of 20%; severe sulfidation corresponds to a pulse frequency of 0.5 kHz, a pulse amplitude of 0.3 times the battery's rated voltage, and a duty cycle of 40%.

[0096] The process of selecting different pulse parameters based on the degree of sulfidation described above ensures that the repair intensity matches the severity of sulfidation.

[0097] Once the pulse parameters are determined, the signal generation unit generates a corresponding pulse sequence signal based on the pulse parameters. After being amplified by the power amplifier unit, the pulse sequence signal is applied between the positive and negative plates of the lead-acid battery through the output isolation unit.

[0098] During the application of the pulse charging signal, the multi-dimensional sensor module continuously monitors changes in the battery's terminal voltage and internal resistance. Terminal voltage monitoring tracks the battery's voltage response characteristics during pulse charging. Internal resistance monitoring assesses the degree of improvement in the battery's internal ohmic characteristics.

[0099] The adaptive repair pulse module evaluates the sulfide removal effect in real time based on terminal voltage monitoring data and internal resistance monitoring data, and compares the evaluation results with a preset repair effect threshold. In this embodiment, the preset repair effect threshold is specifically defined as the internal resistance value decreasing to below 95% of the internal resistance value before pulse charging is initiated, and the terminal voltage recovering to above 98% of the battery's rated voltage. When the evaluation results simultaneously meet both of the above conditions, it is determined that the sulfide has been removed to the preset degree.

[0100] When the evaluation results indicate that the sulfides have been removed to a preset level, the adaptive repair pulse module stops applying the pulse charging signal and generates a repair completion indicator signal, thus ending the current sulfide fault repair process.

[0101] For the power derating control module, after receiving the plate softening fault warning signal, the depth of discharge limiting control process is executed to slow down the degradation rate of the lead-acid battery plate structure.

[0102] After receiving a plate softening fault warning signal, the power shaving control module first determines the degree of plate softening based on the predicted probability of the plate softening fault. There is a preset mapping relationship between the predicted probability of plate softening fault and the degree of plate softening. In this embodiment, the preset mapping relationship is as follows: a predicted probability of plate softening fault greater than or equal to 0.5 and less than 0.7 corresponds to a mild softening level; greater than or equal to 0.7 and less than 0.9 corresponds to a moderate softening level; and greater than or equal to 0.9 corresponds to a severe softening level.

[0103] The degree of plate softening is divided into three levels: mild softening, moderate softening, and severe softening.

[0104] Based on the degree of electrode softening, the power derating control module retrieves the corresponding target discharge depth limit from a preset discharge depth mapping table. The discharge depth mapping table establishes a correspondence between the degree of electrode softening and the maximum permissible discharge depth. In this embodiment, the specific contents of the discharge depth mapping table are as follows: a target discharge depth limit of 80% for a mild softening level, 60% for a moderate softening level, and 40% for a severe softening level.

[0105] The process described above for determining the target depth of discharge limit based on the degree of electrode softening ensures that the stringency of the depth of discharge limit matches the severity of electrode softening.

[0106] The power descent control module receives the discharge demand command issued by the current charge / discharge regulation system. The discharge demand command includes the target discharge depth parameter.

[0107] The power descent control module compares the target discharge depth parameter in the discharge demand command with the retrieved target discharge depth limit.

[0108] When the target discharge depth parameter in the discharge demand command is greater than the target discharge depth limit, the power derating control module will forcibly adjust the target discharge depth parameter to the target discharge depth limit and generate an adjusted discharge control command to output to the charge / discharge actuator.

[0109] When the target discharge depth parameter in the discharge demand command is less than or equal to the target discharge depth limit, the power descent control module directly forwards the discharge demand command as a discharge control command without making any adjustments to the discharge demand command.

[0110] Through the above control logic, the power descent control module ensures that the actual depth of discharge of the battery in any discharge cycle does not exceed the upper limit of the depth of discharge that matches the current health status of the plates, thereby slowing down the further degradation of the plate structure.

[0111] The automatic liquid level compensation module in this embodiment executes an automatic distilled water replenishment process after receiving a low liquid level fault warning signal, in order to restore the electrolyte level of the lead-acid battery to the normal range.

[0112] The hardware configuration of the automatic liquid level compensation module consists of four components: a liquid level sensor, a solenoid valve assembly, a distilled water storage tank, and a pipeline control system.

[0113] The level sensor is installed inside or near the electrolyte chamber of the lead-acid battery to monitor the electrolyte level in real time. The level sensor transmits the monitored level data as an electrical signal to the control unit of the automatic level compensation module.

[0114] The control unit compares the current electrolyte level with a preset level threshold. In this embodiment, the preset level threshold is set to 30% of the total height of the electrolyte chamber. When the current level is lower than the preset level threshold, the control unit triggers a water replenishment process.

[0115] Once the water replenishment process is triggered, the control unit calculates the amount of distilled water that needs to be added. This calculation is based on the difference between the battery's standard electrolyte capacity parameter and the actual capacity parameter corresponding to the current electrolyte level. Specifically, the control unit calculates the difference between the battery's factory-calibrated standard total electrolyte capacity and the actual electrolyte capacity corresponding to the current electrolyte level as the amount of distilled water that needs to be added.

[0116] The control unit sends a water replenishment control command to the solenoid valve assembly. Upon receiving the water replenishment control command, the solenoid valve assembly opens the corresponding channel, allowing distilled water from the distilled water storage tank to be injected into the electrolyte chamber of the battery through the piping system.

[0117] The pipeline system is equipped with a flow control valve to regulate the injection rate of distilled water and prevent electrolyte splashing or drastic temperature changes due to excessive injection rate.

[0118] During the distilled water injection process, the level sensor continuously monitors the changes in the electrolyte level and feeds back the real-time level data to the control unit. When the electrolyte level reaches the target volume, the control unit sends a closing command to the solenoid valve assembly. In this embodiment, the target volume is set to 95% of the total capacity of the standard electrolyte in the battery.

[0119] Upon receiving a shutdown command, the solenoid valve assembly closes the corresponding channel, stops the replenishment of distilled water, and the water replenishment process ends.

[0120] Through the above hardware configuration and closed-loop control process, the automatic liquid level compensation module realizes the automatic monitoring and precise replenishment of electrolyte level.

[0121] Finally, the model optimization module is responsible for continuously evaluating and iteratively optimizing the comprehensive fault prediction model.

[0122] The model optimization module initiates the model performance evaluation process after each fault prediction and repair execution cycle.

[0123] The model performance evaluation process first compares the actual fault types with the fault prediction results. The actual fault types are determined based on post-event manual inspection results or special testing results. The comparison results are used to calculate the prediction accuracy of the comprehensive fault prediction model within the current prediction period. The prediction accuracy is defined as the ratio of the number of correctly predicted samples to the total number of samples.

[0124] The model performance evaluation process simultaneously compares the degree of battery performance recovery after repair with a preset repair effect threshold. Quantitative indicators of performance recovery include the battery's capacity recovery rate, internal resistance recovery rate, and terminal voltage recovery rate after repair. The comparison results are used to calculate a repair effect evaluation index, which comprehensively reflects the effectiveness of the repair measures.

[0125] The above prediction accuracy calculation and repair effect evaluation quantify the overall performance of this cycle from two dimensions: fault prediction accuracy and repair execution effect.

[0126] Based on prediction accuracy and repair effectiveness evaluation metrics, the model optimization module calculates a comprehensive evaluation score. The comprehensive evaluation score is calculated using a weighted average method, with the weighting coefficients for prediction accuracy and repair effectiveness evaluation metrics configured according to the emphasis of the application scenario. In this embodiment, the weighting coefficient for prediction accuracy is set to 0.6, and the weighting coefficient for the repair effectiveness evaluation metrics is set to 0.4.

[0127] When the overall evaluation score falls below a preset optimization threshold, the model optimization module initiates an iterative optimization process for model parameters. This process employs an incremental learning algorithm to adjust the parameters of the comprehensive fault prediction model. The incremental learning algorithm fine-tunes the model parameters using newly collected fault sample data while maintaining the model's existing knowledge. In this embodiment, the incremental learning algorithm is implemented as follows: the incremental learning process is triggered when the accumulated fault sample data reaches a preset threshold of 50 samples; the learning rate during the incremental learning process is set to 10% of the initial training learning rate to avoid excessive impact on the model's existing knowledge.

[0128] After verification and testing, the updated comprehensive fault prediction model is deployed to the fault prediction module by the model optimization module, replacing the original comprehensive fault prediction model. After the new model is put into operation, the system continues to collect battery operating data through the multi-dimensional sensor module and repeats the above-mentioned cycle of fault prediction, self-healing repair, and model optimization.

[0129] Through the aforementioned continuous evaluation and iterative optimization mechanism, the model optimization module enables the predictive performance of the comprehensive fault prediction model to gradually improve with the increase of running time.

[0130] In specific application scenarios, the IoT-based maintenance-free lead-acid battery charge and discharge adaptive control method provided in this embodiment can be deployed in various types of lead-acid battery energy storage systems. Taking a communication base station backup power system as an example, this system is typically equipped with a large number of lead-acid battery banks to ensure continuous power supply to communication equipment during mains power outages. In traditional technical solutions, maintenance personnel need to regularly inspect and maintain the battery banks, including measuring the voltage and internal resistance of each individual battery, adding distilled water, and performing equalization charging.

[0131] The method in this embodiment achieves real-time online monitoring of the operating status of each individual battery cell by deploying multi-dimensional sensor modules in the battery pack. Data such as voltage ripple, current fluctuation, temperature gradient, and internal resistance change rate collected by the sensors are transmitted to a cloud server via an Internet of Things (IoT) communication network.

[0132] The comprehensive fault prediction model running on the cloud server evaluates the degree of sulfation, plate softening, and liquid level status of each individual battery cell based on this data. When the model detects signs of sulfation fault in a single battery cell and the predicted probability of sulfation fault exceeds the first sulfation threshold, the cloud server sends a repair command to the local controller in the area where that battery cell is located.

[0133] Upon receiving the instruction, the local controller activates the adaptive repair pulse generation circuit to generate a corresponding low-frequency pulse charging sequence to repair and charge the individual battery cell. During the repair process, the local controller continuously monitors the terminal voltage and internal resistance changes of the individual battery cell and uploads the monitoring data to the cloud server. The cloud server evaluates the repair effect based on the monitoring data and terminates the repair charging process once the sulfides have been removed to a preset level.

[0134] Similarly, when the model detects signs of plate softening, the system automatically adjusts the depth of discharge limit of that individual battery cell; when the model detects signs of low liquid level, the system automatically triggers a distilled water replenishment process. Through these automated processes, the battery pack can automatically complete early fault prediction, self-healing repair, and performance recovery with reduced manual intervention, extending the battery pack's lifespan and reducing maintenance costs.

[0135] Taking a solar photovoltaic energy storage system as an example, this system is typically equipped with lead-acid battery banks to store the electrical energy generated by the photovoltaic modules during periods of ample sunlight, and to release the energy to power the load at night or during cloudy / rainy periods. In solar photovoltaic energy storage systems, lead-acid batteries operate under a prolonged cycle of shallow charging and discharging, leading to significant water decomposition and evaporation of the electrolyte, which can easily result in low electrolyte levels. Furthermore, due to the intermittent and fluctuating nature of photovoltaic power generation, the batteries frequently experience pulsed high-current charging during the charging process, which can easily induce sulfation and softening of the electrode plates.

[0136] The method in this embodiment continuously monitors the multi-dimensional characteristic parameters of the battery, enabling the detection of signs of problems such as sulfation, plate softening, and low electrolyte level in the early stages of a fault. When the predicted probability of a sulfation fault exceeds the first sulfation threshold, the system automatically triggers a low-frequency pulse charging repair process. This process decomposes sulfide crystals on the plate surface, restoring the battery's active materials' ability to participate in reactions. When the predicted probability of a plate softening fault exceeds the first plate softening threshold, the system automatically adjusts the discharge depth limit to control the mechanical stress on the plate structure during deep discharge. When the predicted probability of a low electrolyte level fault exceeds the first electrolyte level threshold, the system automatically executes a distilled water replenishment process to maintain the electrolyte level within the normal range.

[0137] Through the aforementioned self-healing control mechanism, the lead-acid batteries in the solar photovoltaic energy storage system can maintain their working state in the operating environment, thereby improving the overall service life and reliability of the system.

[0138] On the other hand, this embodiment provides an Internet of Things-based maintenance-free lead-acid battery charge and discharge adaptive control system, including:

[0139] The multi-dimensional sensor module is used to collect voltage ripple signals, current fluctuation signals, temperature gradient distribution data, and internal resistance change rate data of lead-acid batteries in real time, so as to obtain multi-dimensional characteristic parameters that characterize the battery's operating status.

[0140] The fault prediction module has a built-in pre-trained comprehensive fault prediction model, which is used to receive multi-dimensional feature parameters and construct fault feature vectors, and output the probability of sulfurization fault prediction, plate softening fault prediction, and low liquid level fault prediction.

[0141] The adaptive repair pulse module is used to perform low-frequency pulse charging repair when the predicted probability of sulfidation failure exceeds a preset threshold.

[0142] The power descent control module is used to limit the depth of discharge of the battery when the predicted probability of plate softening failure exceeds a preset threshold.

[0143] The automatic liquid level compensation module is used to replenish distilled water to restore the electrolyte level when the probability of a low liquid level fault exceeds a preset threshold.

[0144] The model optimization module is used to evaluate the prediction accuracy of the comprehensive fault prediction model based on the fault prediction results, repair execution status, and real-time operating parameters of the battery, and to iteratively optimize the model parameters based on the evaluation results.

[0145] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0146] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for adaptive charge and discharge control of maintenance-free lead-acid batteries based on the Internet of Things, characterized in that, include: By using a multi-dimensional sensor module to collect real-time voltage ripple signals, current fluctuation signals, temperature gradient distribution data, and internal resistance change rate data of lead-acid batteries, multi-dimensional characteristic parameters characterizing the battery's operating status are obtained. Based on the multidimensional feature parameters, a fault feature vector is constructed, and the fault feature vector is input into a pre-trained comprehensive fault prediction model to obtain the predicted probability of sulfidation fault, the predicted probability of plate softening fault, and the predicted probability of low liquid level fault. When the probability of each of the aforementioned fault predictions exceeds the corresponding preset threshold, the adaptive repair pulse module is triggered to perform sulfidation fault repair, the power descent control module is triggered to perform discharge depth limitation, and the automatic liquid level compensation module is triggered to perform electrolyte replenishment. Based on the fault prediction results, repair execution status, and real-time operating parameters of the battery, the prediction accuracy of the comprehensive fault prediction model is evaluated, and the model parameters of the comprehensive fault prediction model are iteratively optimized according to the evaluation results.

2. The method according to claim 1, characterized in that, The comprehensive fault prediction model includes a feature preprocessing layer, a feature fusion layer, a feature extraction layer, and a classification output layer connected in sequence. The feature fusion layer fuses the standardized feature vectors based on a weighted fusion algorithm. The weight coefficients in the weighted fusion algorithm are determined based on historical fault sample data during the model training phase. Different combinations of weight coefficients correspond to different fault type sensitivities. The feature extraction layer employs a deep neural network containing multiple stacked fully connected hidden layers, which outputs intermediate feature representations after multi-level nonlinear transformation of the fused feature vectors. The classification output layer calculates the predicted probabilities of sulfidation fault, plate softening fault, and low liquid level fault, which are all within the range of 0 to 1, based on the intermediate feature representation.

3. The method according to claim 1, characterized in that, The acquisition of the multidimensional feature parameters includes: Separate the AC ripple component from the voltage ripple signal, extract the fundamental frequency, harmonic amplitude, and energy ratio of each harmonic to the fundamental frequency, and calculate the ripple waveform distortion rate. The amplitude, frequency, and waveform distortion rate of the AC fluctuation component are extracted from the current fluctuation signal. Based on the multi-point temperature data collected by the temperature sensor array, the three-dimensional temperature field inside the battery is reconstructed by interpolation to obtain the temperature gradient of the plate area, the temperature gradient of the electrolyte area, and the temperature gradient of the shell surface. For the internal resistance change rate data, by applying a low-frequency AC excitation signal to the battery circuit, the real part of the dynamic complex impedance is calculated as the ohmic internal resistance, and the absolute value of the internal resistance, the rate of change of the internal resistance, and the acceleration of the change of the internal resistance are extracted.

4. The method according to claim 1, characterized in that, The adaptive repair pulse module performs the sulfurization fault repair process as follows: The degree of sulfide formation is determined based on the predicted probability of sulfidation failure. Based on the degree of sulfide formation, the corresponding pulse frequency, pulse amplitude, and pulse duty cycle are retrieved from a preset pulse parameter library. The pulse frequency is adjustable from 0.5 kHz to 5 kHz, the pulse amplitude is adjustable from 0.05 to 0.3 times the rated voltage of the battery, and the duty cycle is adjustable from 5% to 40%. The corresponding pulse sequence signal is generated, amplified by power, and applied between the positive and negative plates of the battery through the output isolation unit; During pulse application, the battery's terminal voltage and internal resistance are continuously monitored. When the internal resistance drops below 95% of the value before pulse charging starts and the terminal voltage recovers to above 98% of the battery's rated voltage, pulse charging is stopped.

5. The method according to claim 1, characterized in that, The process by which the power descent control module performs the depth of discharge limitation includes: The degree of plate softening is determined based on the predicted probability of plate softening failure. According to the degree of softening of the electrode plate, the corresponding target discharge depth limit is retrieved from the preset discharge depth mapping table, wherein the mild softening level corresponds to the 80% discharge depth limit, the moderate softening level corresponds to the 60% discharge depth limit, and the severe softening level corresponds to the 40% discharge depth limit. Obtain the target discharge depth parameter from the current discharge demand command. When the target discharge depth parameter is greater than the target discharge depth limit, forcibly adjust the target discharge depth parameter to the target discharge depth limit and then output the discharge control command.

6. The method according to claim 1, characterized in that, The automatic liquid level compensation module performs the electrolyte replenishment process as follows: The electrolyte level is monitored in real time by a liquid level sensor. When the current liquid level is lower than a preset liquid level threshold, the amount of distilled water that needs to be added is calculated. The amount of distilled water is the difference between the total capacity of the standard electrolyte of the battery and the actual capacity of the electrolyte corresponding to the current liquid level. The solenoid valve group is opened to allow distilled water in the distilled water storage tank to be injected into the electrolyte chamber of the battery through the pipeline system, and the injection flow rate is regulated by the flow control valve. When the electrolyte level reaches 95% of the total capacity of the standard electrolyte in the battery, close the solenoid valve assembly and stop adding water.

7. The method according to claim 1, characterized in that, The evaluation of the prediction accuracy of the comprehensive fault prediction model and the iterative optimization of the model parameters based on the evaluation results include: After each fault prediction and repair execution cycle, the actual fault type is compared with the fault prediction result to calculate the prediction accuracy. The battery capacity recovery rate, internal resistance recovery rate, and terminal voltage recovery rate after the repair are compared with the preset repair effect threshold to calculate the repair effect evaluation index. The comprehensive evaluation score is calculated based on a weighted average of the prediction accuracy and the repair effect evaluation indicators. When the comprehensive evaluation score is lower than the preset optimization threshold, the parameters of the comprehensive fault prediction model are iteratively optimized using an incremental learning algorithm. Incremental learning is triggered when the newly collected fault sample data accumulates to a preset number threshold, and the learning rate of incremental learning is set to 10% of the learning rate in the initial training phase.

8. The method according to claim 1, characterized in that, The training process of the comprehensive fault prediction model includes: Collect multidimensional historical feature data and corresponding historical fault labels of historical sample batteries. The historical fault labels are three-dimensional binary vectors, which correspond to sulfation fault, plate softening fault and low liquid level fault, respectively. The multidimensional historical feature data is input into the comprehensive fault prediction model to obtain the predicted probability of each fault type. With the goal of minimizing the cross-entropy loss between the predicted probability and the historical fault label, the model parameters are updated using the backpropagation algorithm and the adaptive moment estimation optimization algorithm, and the training is iterated until the loss converges.

9. The method according to claim 1, characterized in that, The multidimensional feature parameters are transmitted to the cloud server through the Internet of Things communication network, and the comprehensive fault prediction model runs on the cloud server. The cloud server generates a repair instruction based on the output of the comprehensive fault prediction model and sends the repair instruction to the local controller, which then controls the adaptive repair pulse module, power descent control module, and automatic liquid level compensation module to perform corresponding actions.

10. An Internet of Things-based maintenance-free lead-acid battery charge-discharge adaptive control system, applied to the method described in any one of claims 1-9, characterized in that, include: The multi-dimensional sensor module is used to collect voltage ripple signals, current fluctuation signals, temperature gradient distribution data, and internal resistance change rate data of lead-acid batteries in real time, so as to obtain multi-dimensional characteristic parameters that characterize the battery's operating status. The fault prediction module has a built-in pre-trained comprehensive fault prediction model, which is used to receive the multi-dimensional feature parameters and construct the fault feature vector, and output the probability of sulfurization fault prediction, the probability of plate softening fault prediction, and the probability of low liquid level fault prediction. An adaptive repair pulse module is used to perform low-frequency pulse charging repair when the predicted probability of the sulfidation fault exceeds a preset threshold. The power derating control module is used to limit the depth of discharge of the battery when the predicted probability of plate softening failure exceeds a preset threshold. An automatic liquid level compensation module is used to replenish distilled water to restore the electrolyte level when the probability of a low liquid level fault exceeds a preset threshold. The model optimization module is used to evaluate the prediction accuracy of the comprehensive fault prediction model based on the fault prediction results, repair execution status, and real-time operating parameters of the battery, and to iteratively optimize the model parameters based on the evaluation results.

Images