High-precision SOC monitoring method for lithium iron phosphate energy storage power station based on Kalman filtering

Abstract

The present invention provides a high-precision SOC monitoring method for lithium iron phosphate energy storage power stations based on Kalman filtering, which relates to the field of battery monitoring technology. The method collects battery module data in real time through a distributed sensor array to generate a real-time multi-dimensional dynamic parameter set aligned in time and space. The optimal state estimation model is dynamically selected, a nonlinear temperature drift compensation function is constructed based on three-dimensional temperature field data, and a radial basis function neural network is combined to perform real-time dynamic correction of the battery internal resistance. The aging compensation coefficient is generated by combining the time-varying attenuation characteristics with the stress-capacity attenuation correlation model. The Kalman filter parameters are updated online through a transfer learning mechanism, and the SOC estimation values ​​of each stage are comprehensively checked and corrected through a triple redundancy check module to output the final SOC estimation value. The present invention significantly improves the accuracy and reliability of SOC monitoring of lithium iron phosphate energy storage power stations through multi-dimensional data fusion, model dynamic optimization and transfer learning mechanism.

Description

Technical Field

[0001] The present invention relates to the field of battery detection technology, and in particular to a high-precision SOC monitoring method for a lithium iron phosphate energy storage power station based on Kalman filtering. Background Art

[0002] Accurately monitoring the battery's state of charge (SOC) is crucial for the operation and management of lithium iron phosphate (LiFePO4) energy storage power plants. However, existing technologies for SOC monitoring have numerous limitations. Traditional methods primarily rely on open-circuit voltage and current integration, but these methods are susceptible to battery aging, temperature fluctuations, and polarization effects, resulting in insufficient monitoring accuracy. With the expansion of energy storage power plants and the increasing complexity of their application scenarios, a single monitoring method can no longer meet the demand for high-precision SOC monitoring. Battery aging is a key factor affecting SOC monitoring accuracy. During long-term charge and discharge cycles, the activity of the electrode materials in LiFePO4 batteries gradually decreases, leading to irreversible capacity degradation. Existing monitoring methods often fail to fully account for the impact of battery aging on SOC, resulting in significant deviations between monitoring results and actual conditions. Furthermore, the impact of temperature changes on battery performance cannot be ignored. Temperature fluctuations alter the battery's internal resistance, polarization characteristics, and chemical reaction rates, thereby affecting accurate SOC estimation. However, many traditional monitoring methods rely solely on fixed temperature models for calibration, making them incapable of adapting to complex and changing operating environments. Polarization effect is also a difficult problem faced by existing technologies. During the battery charging and discharging process, electrochemical polarization and concentration polarization phenomena will cause the battery terminal voltage to deviate from the equilibrium voltage, resulting in errors in the voltage-based SOC estimation. Existing methods often use simple voltage compensation to deal with the polarization effect, but this method cannot capture the time-varying characteristics of the polarization voltage, resulting in limited compensation effect. In addition, existing monitoring systems also have shortcomings in data processing and model updating. Most systems rely only on limited monitoring parameters, such as voltage and current, and ignore the integration and utilization of multi-dimensional information such as temperature fields and stress. At the same time, the parameters of the monitoring model are usually pre-set in the offline stage and cannot be dynamically adjusted according to the actual operating status of the battery, causing the model to gradually fail during long-term operation.

[0003] In summary, existing SOC monitoring technology for lithium iron phosphate energy storage power stations has significant shortcomings in addressing battery aging, temperature fluctuations, polarization effects, and model adaptability, making it difficult to meet the requirements for efficient and safe operation of energy storage systems. Therefore, developing a high-precision SOC monitoring method that can comprehensively consider multi-dimensional influencing factors and has adaptive adjustment capabilities is of great practical significance. Summary of the Invention

[0004] In view of this, the object of the present invention is to provide a high-precision SOC monitoring method for a lithium iron phosphate energy storage power station based on Kalman filtering, so as to at least solve the above problems.

[0005] The technical solution adopted in the present invention is as follows:

[0006] A high-precision SOC monitoring method for lithium iron phosphate energy storage power stations based on Kalman filtering includes:

[0007] Step 1: Using a distributed sensor array to collect voltage, current, three-dimensional temperature field data, and surface stress signals of lithium iron phosphate battery modules in an energy storage power station in real time, and preprocessing them to generate a real-time multi-dimensional dynamic parameter set aligned in time and space;

[0008] Step 2: Input the real-time multi-dimensional dynamic parameter set into an improved multi-model adaptive Kalman filter architecture, which includes three sub-models: a basic equivalent circuit model, a temperature compensation model, and an aging correction model. The optimal state estimation model is dynamically selected from the three sub-models through a model probability weighting algorithm to output an initial SOC estimate.

[0009] Step 3: constructing a nonlinear temperature drift compensation function based on the three-dimensional temperature field data in the real-time multi-dimensional dynamic parameter set, performing real-time dynamic correction on the battery internal resistance based on a radial basis function neural network, and feeding the corrected battery internal resistance parameter back to the temperature compensation model to optimize the initial SOC estimate and generate a second SOC estimate;

[0010] Step 4: Using a pre-trained LSTM time series prediction network to extract the time-varying attenuation characteristics of the battery polarization voltage in the real-time multi-dimensional dynamic parameter set, combining it with a stress-capacity decay correlation model to generate an aging compensation coefficient, and feeding it back to the aging correction model to optimize the second SOC estimate to obtain a third SOC estimate;

[0011] In step 5, the Kalman filter parameters are updated online through the transfer learning mechanism, and the initial SOC estimation value, the second SOC estimation value, and the third SOC estimation value are comprehensively verified and corrected through the triple redundancy verification module to output the final SOC estimation value.

[0012] Furthermore, the construction process of the real-time multi-dimensional dynamic parameter set specifically includes: embedding a micro-thermocouple array inside the battery module, acquiring three-dimensional temperature field data at a sampling interval of 0.5 seconds, synchronously capturing the ripple characteristics of the charge and discharge current and the voltage transient response curve through a high-frequency data acquisition module, and using a MEMS piezoresistive sensor array to monitor the stress distribution on the battery shell surface; achieving spatiotemporal alignment of multi-source heterogeneous data through the ZigBee and CAN bus dual-channel transmission protocols to construct a structured parameter matrix; using a sliding window Fourier transform to identify current noise components, and combining the wavelet threshold denoising algorithm to reconstruct the signal to ensure the dynamic consistency of the input parameters.

[0013] Furthermore, the spatiotemporal alignment method includes: achieving nanosecond-level time synchronization based on the PTP protocol to eliminate multi-sensor time deviation; reconstructing the discrete temperature field through the Delaunay triangulation algorithm, and using the Kriging interpolation method to compensate for the signal loss area; starting the backup sensor array when the signal integrity is lower than 90%, and switching to the fiber optic channel when the CAN bus load exceeds 80%.

[0014] Furthermore, the basic equivalent circuit model, temperature compensation model, and aging correction model work together to generate an initial SOC estimate. The basic equivalent circuit model characterizes the polarization effect through a third-order RC network. The temperature compensation model constructs a temperature-internal resistance transfer function based on the corrected battery internal resistance parameters. The aging correction model integrates stress data and the number of charge and discharge cycles based on the aging compensation coefficient. The model probability weighted algorithm calculates the confidence level based on the residual covariance matrix and switches to a dynamic response priority model when the voltage mutation exceeds 5mV / ms. A thermal runaway warning is triggered when the temperature gradient exceeds 3°C / cm³, freezing the SOC output until the parameters converge.

[0015] Furthermore, the implementation of the dynamic response priority model includes: constructing a fifth-order RC equivalent circuit to characterize the concentration polarization transient process; increasing the sampling frequency to 1kHz when the current suddenly changes, and introducing a feedforward compensation mechanism to predict the charge and discharge direction; dynamically adjusting the filter gain coefficient through Jacobian matrix eigenvalue analysis to accelerate model convergence.

[0016] Furthermore, the method for compensating for nonlinear temperature drift includes: inputting three-dimensional temperature field data in a real-time multi-dimensional dynamic parameter set into a spatiotemporal convolutional network to extract the heat transfer path characteristics of the hot spot area inside the battery; constructing an exponential transfer function of temperature-capacity attenuation, combining a radial basis function neural network to perform real-time dynamic correction of the battery internal resistance, and feeding back the corrected battery internal resistance parameters to the temperature compensation model; when the ambient temperature change rate exceeds 2°C / min, starting the Q-learning algorithm to optimize the temperature sensitivity parameters.

[0017] Furthermore, in step 4, the training process of the LSTM time series prediction network uses the capacity decay trajectory in the historical cycle data as a training set to predict the time-varying decay slope of the polarization voltage, and the historical cycle data includes the voltage, current, three-dimensional temperature field data and surface stress signal in the historical multi-dimensional dynamic parameter set, as well as the corresponding SOC change trajectory; the method of generating an aging compensation coefficient in combination with the stress-capacity decay correlation model includes: inputting the time-varying decay characteristics extracted by the LSTM time series prediction network into the stress-capacity decay correlation model to calculate the loss rate of the electrode active material; weightedly fusing the electrochemical impedance spectroscopy characteristics of the lithium iron phosphate battery module with the number of charge and discharge cycles to generate a comprehensive health factor; in the Kalman filter update stage, calculating the aging compensation coefficient based on the electrode active material loss rate, the comprehensive health factor and the aging compensation matrix.

[0018] Furthermore, the specific implementation process of the transfer learning mechanism includes: constructing a parameter migration mapping table across battery models, and achieving cross-platform adaptation of model parameters through a feature space alignment algorithm; freezing the basic network layer and fine-tuning the top-level regressor when a new charging and discharging mode is detected; synchronizing the optimal filtering parameters of multiple sites through a cloud-based collaborative update mechanism, and rolling back to the stable version when the new parameters cause the SOC error to exceed 2%.

[0019] Furthermore, the triple redundancy check module includes an open circuit voltage check unit, an electrochemical impedance check unit, a coulomb counting check unit and a fuzzy logic arbiter; the open circuit voltage check unit is used to collect the equilibrium voltage during the static stage and obtain the benchmark SOC reference value through a table lookup method; the electrochemical impedance check unit is used to apply a 1kHz~10mHz sweep frequency signal and infer the available capacity through the relaxation time constant; the coulomb counting check unit is used to accumulate the net charge and discharge amount and calculate the SOC in combination with the rated capacity after temperature compensation; the fuzzy logic arbiter is used to dynamically assign a weight coefficient to the benchmark SOC reference value obtained by the open circuit voltage check unit, the available capacity inferred by the electrochemical impedance check unit and the SOC calculated by the coulomb counting check unit according to the current operating conditions, and trigger a manual review instruction when the deviation of the three exceeds 1%.

[0020] Furthermore, the implementation of the electrochemical impedance verification unit includes: designing a multi-frequency parallel excitation strategy to shorten the impedance spectrum acquisition time through orthogonal sweep frequency signals; constructing a capacity inversion model based on relaxation time distribution to extract the characteristic combination of charge transfer resistance and double-layer capacitance; eliminating the influence of ambient temperature on the relaxation time constant based on an impedance-temperature joint analytical algorithm; using a complex domain Kalman filter to collaboratively estimate the real and imaginary parts of the impedance to improve parameter identification accuracy; and prohibiting high-current charging and discharging operations when a characteristic frequency band of electrolyte decomposition is detected.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] A distributed sensor array collects multi-dimensional data such as voltage, current, three-dimensional temperature field, and surface stress in real time, and performs time-space alignment processing to provide a more comprehensive and accurate information basis for SOC estimation. This multi-dimensional data fusion method effectively overcomes the monitoring errors caused by traditional methods that rely on only a single parameter.

[0023] Adopting an improved multi-model adaptive Kalman filter architecture, including a basic equivalent circuit model, a temperature compensation model, and an aging correction model, it dynamically selects the optimal state estimation model through a model probability weighted algorithm to ensure high-precision initial SOC estimation under different operating conditions.

[0024] A nonlinear temperature drift compensation function is constructed based on three-dimensional temperature field data, and combined with a radial basis function neural network to dynamically correct the battery internal resistance in real time. This temperature compensation mechanism can effectively adapt to complex and changing operating environments and significantly reduce the impact of temperature fluctuations on SOC monitoring.

[0025] A pre-trained LSTM time series prediction network is used to extract the time-varying attenuation characteristics of the battery polarization voltage. This is then combined with a stress-capacity attenuation correlation model to generate an aging compensation coefficient. This is then fed back into the aging correction model to accurately compensate for the battery aging effect and ensure the reliability of long-term monitoring.

[0026] A transfer learning mechanism is introduced to iteratively update the Kalman filter parameters online, enabling the monitoring model to dynamically adjust according to the actual operating status of the battery, maintaining long-term effectiveness. At the same time, a triple redundancy check module is used to comprehensively check and correct the SOC estimation values ​​at each stage, further improving monitoring accuracy.

[0027] Through the synergistic effect of multi-model architecture, temperature compensation, aging correction and transfer learning mechanism, the present invention can significantly improve the accuracy and reliability of SOC monitoring of lithium iron phosphate energy storage power stations, providing strong support for the efficient operation and safe management of energy storage systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0029] Figure 1 It is a flow chart of the high-precision SOC monitoring method for a lithium iron phosphate energy storage power station described in the embodiment. DETAILED DESCRIPTION

[0030] The principles and features of the present invention are described below with reference to the accompanying drawings. The enumerated embodiments are only used to explain the present invention and are not used to limit the scope of the present invention.

[0031] Please refer to Figure 1 The embodiment of the present invention provides a high-precision SOC monitoring method for a lithium iron phosphate energy storage power station based on Kalman filtering, which is as follows:

[0032] Step 1: Use a distributed sensor array to collect the voltage, current, three-dimensional temperature field data and surface stress signals of the lithium iron phosphate battery module in the energy storage power station in real time, and preprocess them to generate a real-time multi-dimensional dynamic parameter set aligned in time and space.

[0033] (1) Data collection

[0034] We embed a micro-thermocouple array inside the battery module to acquire three-dimensional temperature field data at a sampling interval of 0.5 seconds. We use a high-frequency data acquisition module to synchronously capture the ripple characteristics of the charging and discharging current and the voltage transient response curve, and use a MEMS piezoresistive sensor array to monitor the stress distribution on the battery casing surface.

[0035] (2) Data preprocessing

[0036] We use ZigBee and CAN bus dual-channel transmission protocols to achieve spatiotemporal alignment of multi-source heterogeneous data and construct a structured parameter matrix.

[0037] Among them, the methods of spatiotemporal alignment include: achieving nanosecond-level time synchronization based on the PTP protocol to eliminate multi-sensor time deviation and ensure the synchronization of data acquisition; reconstructing the discrete temperature field through the Delaunay triangulation algorithm, and using the Kriging interpolation method to compensate for signal loss areas to improve the integrity of the temperature field data; starting the backup sensor array when the signal integrity is lower than 90%, and switching to the fiber optic channel when the CAN bus load exceeds 80%, to ensure the reliability of data acquisition.

[0038] The Kriging interpolation formula is

[0039]

[0040] in, Represents the internal temperature field value of the lithium iron phosphate battery module after interpolation; represents the weight coefficient; represents the temperature value of the i-th sensor; Indicates the number of temperature sensors distributed inside the battery module.

[0041] The data is then optimized, and the sliding window Fourier transform is used to identify the current noise component. The signal is reconstructed in combination with the wavelet threshold denoising algorithm to ensure the dynamic consistency of the input parameters. In this way, a real-time multi-dimensional dynamic parameter set that meets the requirements is obtained.

[0042] In this embodiment, the sliding window Fourier transform formula is:

[0043]

[0044] in, Represents the Fourier transform result, which is used to extract the current ripple characteristics; Represents the input current signal, i.e. the charge and discharge current of the lithium iron phosphate battery module; Represents a sliding window function, used to control the window position; Indicates the number of sampling points; represents the frequency index; Indicates the window position index; Represents an imaginary unit.

[0045] In step 2, the real-time multi-dimensional dynamic parameter set is input into the improved multi-model adaptive Kalman filter architecture. The architecture includes three types of sub-models: basic equivalent circuit model, temperature compensation model and aging correction model. The optimal state estimation model is dynamically selected from the three types of sub-models through the model probability weighted algorithm to output the initial SOC estimation value.

[0046] The basic equivalent circuit model, temperature compensation model, and aging correction model work together to generate an initial SOC estimate. The basic equivalent circuit model uses a third-order RC network to characterize polarization effects, improving the model's adaptability to dynamic operating conditions. The temperature compensation model optimizes temperature compensation by constructing a temperature-resistance transfer function based on corrected battery internal resistance parameters. The aging correction model integrates stress data and the number of charge and discharge cycles based on an aging compensation coefficient to dynamically correct the impact of battery aging on SOC. The model's probability weighted algorithm calculates confidence based on the residual covariance matrix. When the voltage mutation exceeds 5mV / ms, the model switches to a dynamic response priority model. When the temperature gradient exceeds 3°C / cm³, a thermal runaway warning is triggered, freezing the SOC output until the parameters converge.

[0047] Specifically, the implementation of the dynamic response priority model includes: constructing a fifth-order RC equivalent circuit to characterize the concentration polarization transient process, improving the model's adaptability to rapid charging and discharging; increasing the sampling frequency to 1kHz when the current suddenly changes, and introducing a feedforward compensation mechanism to predict the charging and discharging direction to accelerate the model response; dynamically adjusting the filter gain coefficient through Jacobian matrix eigenvalue analysis to accelerate model convergence.

[0048] The sampling frequency adjustment formula of the dynamic response priority model is:

[0049]

[0050] in, Indicates the adjusted sampling frequency; Indicates the basic sampling frequency; Indicates the adjustment coefficient determined according to the dynamic response characteristics of the battery; Indicates the current mutation of the lithium iron phosphate battery module; Indicates the maximum battery current.

[0051] The voltage response formula of a fifth-order RC network is:

[0052]

[0053] in, Indicates the open circuit voltage, i.e. the battery terminal voltage; represents the equilibrium voltage, that is, the voltage when there is no polarization; Indicates the charge and discharge current; Indicates the The resistance of each RC branch; Indicates the The time constant of each RC branch; is a time variable, which indicates the time change during the charging and discharging process.

[0054] Step 3: Construct a nonlinear temperature drift compensation function based on the three-dimensional temperature field data in the real-time multi-dimensional dynamic parameter set, perform real-time dynamic correction of the battery internal resistance based on the radial basis function neural network, and feed the corrected battery internal resistance parameters back to the temperature compensation model to optimize the initial SOC estimation value and generate a second SOC estimation value.

[0055] Specifically, the method for compensating for nonlinear temperature drift includes: inputting the three-dimensional temperature field data in the real-time multi-dimensional dynamic parameter set into the spatiotemporal convolutional network, extracting the heat transfer path characteristics of the hot spot area inside the battery, and identifying the key areas of temperature drift; constructing an exponential transfer function of temperature-capacity attenuation, combining the radial basis function neural network to perform real-time dynamic correction of the battery internal resistance, and feeding the corrected battery internal resistance parameters back to the temperature compensation model; when the ambient temperature change rate exceeds 2°C / min, starting the Q-learning algorithm to optimize the temperature sensitivity parameters to ensure the stability of the compensation effect.

[0056] Among them, the exponential transfer function of temperature-capacity decay is:

[0057]

[0058] in, Indicates the battery capacity at the current temperature T; Indicates the reference temperature Battery capacity under Indicates the battery temperature sensitivity coefficient.

[0059] The battery internal resistance correction formula is:

[0060]

[0061] in, Indicates the corrected internal resistance of the battery; Indicates the internal resistance of the battery at the reference temperature; Indicates the temperature coefficient of internal resistance.

[0062] In step 4, the pre-trained LSTM time series prediction network is used to extract the time-varying attenuation characteristics of the battery polarization voltage in the real-time multi-dimensional dynamic parameter set, and the aging compensation coefficient is generated by combining the stress-capacity attenuation correlation model. The coefficient is fed back to the aging correction model to dynamically optimize the second SOC estimate and obtain the third SOC estimate.

[0063] In this embodiment, the training process of the LSTM time series prediction network uses the capacity decay trajectory in the historical cycle data as the training set to predict the time-varying decay slope of the polarization voltage and extract the battery aging characteristics. The historical cycle data includes the voltage, current, three-dimensional temperature field data and surface stress signals in the historical multi-dimensional dynamic parameter set, as well as the corresponding SOC change trajectory.

[0064] In this embodiment, the method for generating an aging compensation coefficient in combination with a stress-capacity decay correlation model includes: inputting the time-varying attenuation characteristics extracted by the LSTM time series prediction network into the stress-capacity decay correlation model to calculate the loss rate of the electrode active material; weightedly fusing the electrochemical impedance spectroscopy characteristics of the lithium iron phosphate battery module with the number of charge and discharge cycles to generate a comprehensive health factor; and in the Kalman filter update stage, calculating the aging compensation coefficient based on the electrode active material loss rate, the comprehensive health factor, and the aging compensation matrix.

[0065] In this embodiment, the calculation formula of the aging compensation coefficient is:

[0066]

[0067] in, Indicates the aging compensation coefficient, which is used to correct the SOC estimation value; Represents the loss rate of electrode active material predicted based on LSTM; represents the comprehensive health factor; represents the aging compensation matrix related to the battery aging model; Represents the surface stress distribution data of the battery shell.

[0068] In step 5, the Kalman filter parameters are updated online through the transfer learning mechanism, and the initial SOC estimation value, the second SOC estimation value, and the third SOC estimation value are comprehensively verified and corrected through the triple redundancy verification module to output the final SOC estimation value.

[0069] Among them, the specific implementation process of the transfer learning mechanism includes: building a parameter migration mapping table across battery models, achieving cross-platform adaptation of model parameters through a feature space alignment algorithm to improve the versatility of the model; freezing the basic network layer when a new charging and discharging mode is detected, and fine-tuning the top-level regressor to ensure the model's adaptability to new working conditions; synchronizing the optimal filtering parameters of multiple sites through a cloud-based collaborative update mechanism, and rolling back to a stable version when the new parameters cause the SOC error to exceed 2% to ensure system stability.

[0070] The triple redundancy verification module consists of an open-circuit voltage verification unit, an electrochemical impedance spectroscopy unit, a coulomb counting verification unit, and a fuzzy logic arbiter. The open-circuit voltage verification unit collects the equilibrium voltage during the static phase and uses a table lookup to obtain a baseline SOC reference value. The electrochemical impedance spectroscopy unit applies a 1kHz to 10mHz sweep frequency signal to infer the available capacity based on the relaxation time constant. The coulomb counting verification unit accumulates the net charge and discharge charge and calculates the SOC based on the temperature-compensated rated capacity. The accuracy of the SOC estimate is verified from different perspectives using open-circuit voltage verification, electrochemical impedance spectroscopy, and coulomb counting verification. The fuzzy logic arbiter dynamically assigns weighting coefficients to the baseline SOC reference value obtained by the open-circuit voltage verification unit, the available capacity inferred by the electrochemical impedance spectroscopy unit, and the SOC calculated by the coulomb counting verification unit based on the current operating conditions. A manual review is triggered when the three values ​​deviate by more than 1%, ensuring the high reliability of the final SOC estimate.

[0071] The fuzzy logic arbitration weight distribution formula is:

[0072]

[0073] in, Indicates the The weight of each verification unit; Indicates the The confidence level of each verification unit; Indicates the The sensitivity of each calibration unit; Indicates the The confidence level of each verification unit; Indicates the The sensitivity of each calibration unit.

[0074] The implementation of the electrochemical impedance spectroscopy unit includes: designing a multi-frequency parallel excitation strategy, shortening the impedance spectrum acquisition time through orthogonal sweep signals, and improving parameter identification efficiency; constructing a capacity inversion model based on relaxation time distribution, extracting the characteristic combination of charge transfer resistance and double-layer capacitance, and inferring the available capacity of the battery; eliminating the influence of ambient temperature on the relaxation time constant based on the impedance-temperature joint analytical algorithm to ensure the accuracy of the impedance parameters; using complex domain Kalman filtering to collaboratively estimate the real and imaginary parts of the impedance to improve parameter identification accuracy; prohibiting high current charging and discharging operations when the characteristic frequency band of electrolyte decomposition is detected.

[0075] The relaxation time distribution model is:

[0076]

[0077] in, Represents the electrochemical impedance of the battery; Represents the charge transfer resistance, that is, the resistance of the chemical reaction inside the battery; Represents the double-layer capacitance formed on the battery surface; Indicates the The resistance corresponding to the relaxation time; Indicates the relaxation time; represents the angular frequency; Represents an imaginary unit.

[0078] In this embodiment, the Kalman filter state update formula is:

[0079]

[0080] in, Indicates the The estimated SOC value at the moment is the state of charge of the lithium iron phosphate battery module; Indicates the The prior estimate of the moment is a prediction based on the previous moment; Represents the Kalman filter gain, which determines the degree of influence of the observation value on the estimated value; Indicates the Observed values ​​at a given moment, such as voltage, current, etc.; represents the observation matrix.

[0081] We conducted an error experiment comparison analysis between the method of the present invention (OAF) and the existing technologies (open circuit voltage method, extended Kalman filter, and ampere-hour integration method) under two working conditions (working condition A: constant current charge and discharge; working condition B: dynamic pulse charge and discharge). The details are as follows:

[0082] (1) Analysis of working condition A

[0083] The root mean square error of the method (OAF) of the present invention is only 0.6%, which is significantly lower than that of the existing method 1 (open circuit voltage method (OCV), 4.5%), the existing method 2 (extended Kalman filter (EKF), 1.8%) and the existing method 3 (ampere-hour integration method (Ah), 3.2%).

[0084] The open-circuit voltage method relies on a static voltage lookup table and cannot compensate for polarization effects. The ampere-hour integration method suffers from errors due to the cumulative error of current integration. While the extended Kalman filter is superior to the previous two, it is limited by the lack of adaptability of a single model to dynamic characteristics. The method proposed in this paper combines voltage, current, temperature field, and stress data to reduce the error of a single parameter. It uses an LSTM network to predict the polarization voltage decay characteristics, thus avoiding capacity misjudgments caused by long-term charging and discharging.

[0085] (2) Analysis of working condition B

[0086] The maximum absolute error of the method of the present invention is only 1.8%, which is much lower than existing method 1 (20.0%), existing method 2 (6.5%) and existing method 3 (12.0%).

[0087] The open-circuit voltage method completely fails when current changes suddenly. The ampere-hour integration method leads to error accumulation due to high-frequency current fluctuations. The extended Kalman filter has a delayed response due to its high model complexity. However, the method proposed in this paper increases the sampling frequency to 1kHz when current changes suddenly and introduces feedforward compensation to accelerate the response. It also constructs a nonlinear compensation function based on the three-dimensional temperature field, reducing the impact of temperature fluctuations on SOC.

[0088] (3) Comprehensive performance comparison

[0089] The root mean square error of the method of the present invention is the lowest under both working conditions (A: 0.6%, B: 1.0%).

[0090] The root mean square error of the existing method 1 under dynamic working conditions is as high as 15.0%; the complex model of the existing method 2 leads to a root mean square error of 4.5% under dynamic working conditions; due to the cumulative error of the existing method 3, the root mean square error under dynamic working conditions is 8.0%.

[0091] The maximum absolute error of the proposed method under dynamic conditions is only 1.8%, demonstrating its strong robustness. The maximum absolute error of the existing method 1 under dynamic conditions is as high as 20%, which may cause false alarms in the system.

[0092] The error information tables of the method of the present invention, existing method 1, existing method 2, and existing method 3 under two working conditions A and B are as follows:

[0093]

[0094] (4) Conclusion

[0095] Experimental data show that the method of the present invention exhibits significant advantages under both constant current charging and discharging and dynamic pulse charging and discharging typical operating conditions: the root mean square error under static conditions is 87% lower than that of the open-circuit voltage method and 67% lower than that of the extended Kalman filter; the root mean square error under dynamic conditions is 88% lower than that of the ampere-hour integral method; through dynamic model switching and real-time compensation mechanisms, the maximum absolute error fluctuation during current mutations is controlled within 1.8%; the three-dimensional temperature field compensation and redundancy check module minimize the impact of temperature fluctuations and signal noise. The present invention's multi-model adaptive fusion, real-time compensation for aging and temperature, and dynamic response optimization jointly address the limitations of traditional methods in complex scenarios and provide reliable technical support for the efficient management of energy storage power stations.

[0096] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A high-precision SOC monitoring method for lithium iron phosphate energy storage power stations based on Kalman filtering, characterized in that: include: Step 1: Using a distributed sensor array to collect voltage, current, three-dimensional temperature field data, and surface stress signals of lithium iron phosphate battery modules in an energy storage power station in real time, and preprocessing them to generate a real-time multi-dimensional dynamic parameter set aligned in time and space; Step 2: Input the real-time multi-dimensional dynamic parameter set into an improved multi-model adaptive Kalman filter architecture, which includes three sub-models: a basic equivalent circuit model, a temperature compensation model, and an aging correction model. The optimal state estimation model is dynamically selected from the three sub-models through a model probability weighting algorithm to output an initial SOC estimate. Step 3: constructing a nonlinear temperature drift compensation function based on the three-dimensional temperature field data in the real-time multi-dimensional dynamic parameter set, performing real-time dynamic correction on the battery internal resistance based on a radial basis function neural network, and feeding the corrected battery internal resistance parameter back to the temperature compensation model to optimize the initial SOC estimate and generate a second SOC estimate; Step 4: Using a pre-trained LSTM time series prediction network to extract the time-varying attenuation characteristics of the battery polarization voltage in the real-time multi-dimensional dynamic parameter set, combining it with a stress-capacity decay correlation model to generate an aging compensation coefficient, and feeding it back to the aging correction model to optimize the second SOC estimate to obtain a third SOC estimate; The LSTM time series prediction network is trained using the capacity decay trajectory in historical cycle data as a training set to predict the time-varying decay slope of the polarization voltage. The historical cycle data includes voltage, current, three-dimensional temperature field data, and surface stress signals in the historical multi-dimensional dynamic parameter set, as well as the corresponding SOC change trajectory. The method for generating an aging compensation coefficient by combining a stress-capacity decay correlation model includes: inputting the time-varying decay characteristics extracted by the LSTM time series prediction network into the stress-capacity decay correlation model to calculate the electrode active material loss rate; weightedly fusing the electrochemical impedance spectroscopy characteristics of the lithium iron phosphate battery module with the number of charge and discharge cycles to generate a comprehensive health factor; and calculating the aging compensation coefficient based on the electrode active material loss rate, the comprehensive health factor, and the aging compensation matrix during the Kalman filter update phase. Step 5: Use the transfer learning mechanism to iteratively update the Kalman filter parameters online, and use the triple redundancy check module to comprehensively check and correct the initial SOC estimate, the second SOC estimate, and the third SOC estimate to output the final SOC estimate; The triple redundancy check module includes an open-circuit voltage check unit, an electrochemical impedance check unit, a coulomb counting check unit and a fuzzy logic arbiter; the open-circuit voltage check unit is used to collect the equilibrium voltage during the static stage and obtain the benchmark SOC reference value through a table lookup method; the electrochemical impedance check unit is used to apply a 1kHz~10mHz sweep frequency signal and inversely infer the available capacity through the relaxation time constant; the coulomb counting check unit is used to accumulate the net charge and discharge amount and calculate the SOC in combination with the rated capacity after temperature compensation; the fuzzy logic arbiter is used to dynamically assign weight coefficients to the benchmark SOC reference value obtained by the open-circuit voltage check unit, the available capacity inversely inferred by the electrochemical impedance check unit, and the SOC calculated by the coulomb counting check unit according to the current operating conditions, and trigger a manual review instruction when the deviation of the three exceeds 1%.

2. The high-precision SOC monitoring method for a lithium iron phosphate energy storage power station based on Kalman filtering according to claim 1 is characterized in that: The construction process of the real-time multi-dimensional dynamic parameter set specifically includes: embedding a micro-thermocouple array inside the battery module to obtain three-dimensional temperature field data at a sampling interval of 0.5 seconds, synchronously capturing the ripple characteristics of the charge and discharge current and the voltage transient response curve through a high-frequency data acquisition module, and using a MEMS piezoresistive sensor array to monitor the stress distribution on the battery casing surface; achieving spatiotemporal alignment of multi-source heterogeneous data through the ZigBee and CAN bus dual-channel transmission protocols to construct a structured parameter matrix; using a sliding window Fourier transform to identify current noise components, and combining a wavelet threshold denoising algorithm to reconstruct the signal to ensure the dynamic consistency of the input parameters.

3. The high-precision SOC monitoring method for a lithium iron phosphate energy storage power station based on Kalman filtering according to claim 2 is characterized in that: The spatiotemporal alignment method includes: achieving nanosecond-level time synchronization based on the PTP protocol to eliminate multi-sensor time deviation; reconstructing the discrete temperature field through the Delaunay triangulation algorithm and compensating for signal loss areas using the Kriging interpolation method; activating the backup sensor array when the signal integrity is less than 90%, and switching to the fiber optic channel when the CAN bus load exceeds 80%.

4. The high-precision SOC monitoring method for a lithium iron phosphate energy storage power station based on Kalman filtering according to claim 1 is characterized in that: The basic equivalent circuit model, temperature compensation model and aging correction model work together to generate an initial SOC estimate; the basic equivalent circuit model characterizes the polarization effect through a third-order RC network, the temperature compensation model constructs a temperature-internal resistance transfer function based on the corrected battery internal resistance parameters, and the aging correction model integrates stress data and charge and discharge cycle times based on the aging compensation coefficient; the model probability weighted algorithm calculates the confidence level based on the residual covariance matrix, and switches to a dynamic response priority model when the voltage mutation exceeds 5mV / ms; when the temperature gradient exceeds 3℃ / cm 3 The thermal runaway warning is triggered and the SOC output is frozen until the parameters converge.

5. The high-precision SOC monitoring method for a lithium iron phosphate energy storage power station based on Kalman filtering according to claim 4 is characterized in that: The implementation of the dynamic response priority model includes: constructing a fifth-order RC equivalent circuit to characterize the concentration polarization transient process; increasing the sampling frequency to 1kHz when the current changes suddenly, and introducing a feedforward compensation mechanism to predict the charge and discharge direction; and dynamically adjusting the filter gain coefficient through Jacobian matrix eigenvalue analysis to accelerate model convergence.

6. The high-precision SOC monitoring method for a lithium iron phosphate energy storage power station based on Kalman filtering according to claim 1 is characterized in that: The method for compensating for nonlinear temperature drift includes: inputting three-dimensional temperature field data in a real-time multi-dimensional dynamic parameter set into a spatiotemporal convolutional network to extract the heat transfer path characteristics of the hot spot area inside the battery; constructing an exponential transfer function for temperature-capacity decay, combining it with a radial basis function neural network to perform real-time dynamic correction of the battery internal resistance, and feeding the corrected battery internal resistance parameter back to the temperature compensation model; when the ambient temperature change rate exceeds 2°C / min, starting a Q-learning algorithm to optimize the temperature sensitivity parameter.

7. The high-precision SOC monitoring method for a lithium iron phosphate energy storage power station based on Kalman filtering according to claim 1 is characterized in that: The specific implementation process of the transfer learning mechanism includes: building a parameter migration mapping table across battery models, achieving cross-platform adaptation of model parameters through a feature space alignment algorithm; freezing the base network layer and fine-tuning the top-level regressor when a new charging and discharging mode is detected; synchronizing the optimal filtering parameters of multiple sites through a cloud-based collaborative update mechanism, and rolling back to the stable version when the new parameters cause the SOC error to exceed 2%.

8. The high-precision SOC monitoring method for a lithium iron phosphate energy storage power station based on Kalman filtering according to claim 1 is characterized in that: The implementation of the electrochemical impedance verification unit includes: designing a multi-frequency parallel excitation strategy to shorten the impedance spectrum acquisition time through orthogonal sweep frequency signals; constructing a capacity inversion model based on relaxation time distribution to extract the characteristic combination of charge transfer resistance and double-layer capacitance; eliminating the influence of ambient temperature on the relaxation time constant based on an impedance-temperature joint analytical algorithm; using a complex domain Kalman filter to collaboratively estimate the real and imaginary parts of the impedance to improve parameter identification accuracy; and prohibiting high-current charging and discharging operations when a characteristic frequency band of electrolyte decomposition is detected.

Images