Series battery high-precision electric quantity monitoring method and system based on adaptive algorithm

Abstract

The invention relates to the technical field of battery health management, in particular to a series battery high-precision electric quantity monitoring method and system based on an adaptive algorithm, and the method comprises the steps: obtaining multi-dimensional parameter data of a series battery pack; operating a main state estimator based on an electrothermal coupling equivalent circuit model, and performing online estimation according to the multi-dimensional parameter data to obtain a baseline state-of-charge value; predicting an estimation error of the baseline state-of-charge value based on the multi-dimensional parameter data by adopting a probability error prediction model, and outputting error probability distribution; generating a final high-fidelity state-of-charge estimation value by using the error probability distribution through a dual-channel self-adaptive collaborative correction process; and constructing and driving a battery protection model by using the final high-fidelity state-of-charge estimation value and the multi-dimensional parameter data, and outputting a safety control instruction for controlling and protecting the series battery pack. According to the invention, accurate monitoring and active safety protection of electric quantity are realized by using a self-adaptive algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of battery health management, and specifically to a high-precision power monitoring method and system for series-connected batteries based on an adaptive algorithm. Background Art

[0002] The battery state of charge is a key core parameter in the battery management system. Its accurate estimation is crucial to ensuring the safety of the battery system and extending its service life.

[0003] Currently, mainstream battery SOC estimation methods include the ampere-hour integration method and model-based filtering algorithms (such as Kalman filtering). While the ampere-hour integration method is simple to implement, its open-loop calculation method is susceptible to current measurement noise and initial SOC value errors, resulting in estimation errors accumulating over time. Traditional modeling methods, when establishing a battery equivalent model, struggle to fully and accurately reflect the complex electrochemical and thermodynamic dynamic characteristics of batteries in actual operation. Therefore, when faced with variable dynamic operating conditions and significant temperature fluctuations, the accuracy and robustness of existing battery SOC estimation methods still need to be improved, and they are unable to fully meet the application requirements of high-precision real-time monitoring of battery charge.

[0004] Therefore, a high-precision power monitoring method and system for series batteries based on an adaptive algorithm is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a high-precision power monitoring method and system for series-connected batteries based on an adaptive algorithm, which realizes accurate power monitoring and active safety protection by utilizing an adaptive algorithm. The method comprises the following steps: obtaining multi-dimensional parameter data of a series-connected battery pack; operating a main state estimator based on an electrothermal coupling equivalent circuit model, and online estimating a baseline state of charge value based on the multi-dimensional parameter data; using a probabilistic error prediction model, predicting the estimation error of the baseline state of charge value based on the multi-dimensional parameter data, and outputting an error probability distribution; generating a final high-fidelity state of charge estimate using the error probability distribution through a dual-channel adaptive collaborative correction process; and constructing and driving a battery protection model using the final high-fidelity state of charge estimate and the multi-dimensional parameter data, outputting safety control instructions for controlling and protecting the series-connected battery pack.

[0006] To achieve the above object, the present invention provides the following technical solutions: A high-precision battery capacity monitoring method for series batteries based on an adaptive algorithm includes: Acquiring multi-dimensional parameter data of the series-connected battery pack, the multi-dimensional parameter data including the terminal voltage, current, and temperature of each single battery; Running a main state estimator based on an electrothermal coupling equivalent circuit model, wherein the main state estimator uses a Kalman filter algorithm to obtain a baseline state of charge value based on the multi-dimensional parameter data online; A probabilistic error prediction model is used to predict the estimated error of the baseline state of charge value based on the multidimensional parameter data, and an error probability distribution is output including an error mean and a variance used to characterize the prediction confidence; and a final high-fidelity state of charge estimate is generated using the error probability distribution through a dual-channel adaptive collaborative correction process; The final high-fidelity state of charge estimation value and multi-dimensional parameter data are used to construct and drive a battery protection model, and output safety control instructions, which are used to control and protect the series battery pack.

[0007] Preferably, the electrothermal coupling equivalent circuit model includes: The temperature-sensitive parameter adaptation layer contains a parameter library that stores preset multi-dimensional mapping relationships between circuit parameters and temperature. It receives real-time temperature as a query input, retrieves and outputs the complete circuit parameters corresponding to the current temperature from the parameter library; The electrical state calculation layer receives circuit parameters and updates state variables through integration operations based on the real-time current input, and calculates the predicted terminal voltage value according to Kirchhoff's voltage law; The coupled heat generation calculation layer is used to calculate the total heat power generated by electrical activities; it receives the real-time current and the internal resistance value and entropy heat coefficient corresponding to the current temperature, calculates the irreversible Joule heat and reversible entropy heat respectively, and sums the two to output the total heat generation power value.

[0008] Preferably, the main state estimator adopts a Kalman filter algorithm, and the process of online estimating the baseline state of charge value based on the multi-dimensional parameter data includes: obtaining the state prior prediction value output by the electrothermal coupling equivalent circuit model after forward operation, the state prior prediction value contains the prior state of charge and the terminal voltage value predicted by the model; comparing the terminal voltage predicted by the electrothermal coupling equivalent circuit model with the actual battery terminal voltage synchronously sampled by the hardware and calculating the measurement residual, calculating the optimal Kalman gain based on the measurement residual and the error covariance matrix, and using the optimal Kalman gain to perform weighted correction on the state prior prediction value to generate the baseline state of charge value.

[0009] Preferably, the probability error prediction model includes: The state feedback and working condition fusion perception layer is used to receive real-time multi-dimensional parameter data synchronized with the main state estimator, and at the same time receive the diagonal elements of the optimal Kalman gain and error covariance matrix output by the main state estimator at the previous moment, and fuse them into a high-dimensional feature vector; the deep coupling error inference layer is used to model and predict the estimation error of the main state estimator at the current moment based on the high-dimensional feature vector; the confidence distribution generation layer is used to parse and encapsulate the prediction results of the deep coupling error inference layer, and output an error probability distribution containing the error mean and the variance used to characterize the prediction confidence.

[0010] Preferably, the process of generating a final high-fidelity state of charge estimation value using the error probability distribution through a dual-channel adaptive collaborative correction process includes: Based on the error probability distribution, which includes error variance and error mean, calculations of two correction channels are performed; in the output-end gated correction channel, the error variance is converted into a confidence fusion weight through a preset inverse function, the fusion weight is multiplied by the error mean to obtain a weighted error correction, and the weighted error correction is compensated to the baseline state of charge value to generate a final high-fidelity state of charge estimation value; in the model-end online adaptive channel, the error variance is used as a dynamic indicator of the mismatch degree of the electrothermal coupling equivalent circuit model, and is mapped to a process noise adjustment value through a preset monotonically increasing function, and the noise adjustment value is superimposed on the noise covariance matrix of the Kalman filter algorithm.

[0011] Preferably, the process of using the final high-fidelity state of charge estimation value and the multi-dimensional parameter data to construct and drive a battery protection model and output a safety control instruction includes: The final high-fidelity state of charge estimate and the real-time temperature and battery health status information in the multi-dimensional parameter data are input as a combined feature vector into a neural network model for inferring safety boundaries to generate a safe operating boundary for the battery in the current state. The safe operating boundary includes a charging cutoff SOC threshold, a discharging cutoff SOC threshold, a maximum charge and discharge current allowed under the current operating conditions, and a maximum / minimum operating temperature that are dynamically adjusted according to the state. The final high-fidelity state of charge estimate and the real-time current and temperature are compared with the safe operating boundary. If it is detected that the real-time parameters have a tendency to cross the corresponding safety boundary, the corresponding protection logic is triggered, and a safety control instruction is generated to directly control the charging and discharging circuit.

[0012] A high-precision battery capacity monitoring system for series-connected batteries based on an adaptive algorithm, including: A data acquisition module, which acquires multi-dimensional parameter data of the series-connected battery pack, wherein the multi-dimensional parameter data includes the terminal voltage, current and temperature of each single battery; a state estimation module, running a main state estimator based on an electrothermal coupling equivalent circuit model, wherein the main state estimator uses a Kalman filter algorithm to online estimate a baseline state of charge value based on the multi-dimensional parameter data; an error correction module, employing a probabilistic error prediction model to predict an estimated error of the baseline state of charge value based on the multi-dimensional parameter data, and outputting an error probability distribution including an error mean and a variance used to characterize the prediction confidence; and utilizing the error probability distribution to generate a final high-fidelity state of charge estimate through a dual-channel adaptive collaborative correction process; The safety control module uses the final high-fidelity state of charge estimation value and multi-dimensional parameter data to build and drive a battery protection model and output safety control instructions, wherein the safety control instructions are used to control and protect the series battery pack.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. By adopting the electrothermal coupling equivalent circuit model, the effect of temperature on the internal characteristics of the battery is more accurately described, which helps to reduce the risk of model mismatch caused by temperature changes and provides a more reliable model basis for subsequent accurate estimation.

[0014] 2. By establishing a probabilistic error prediction model, the system has the ability to actively evaluate its own estimation uncertainty, providing a quantitative reference for the reliability of the current SOC estimation results, thereby providing a clear decision-making basis for subsequent adaptive corrections.

[0015] 3. By executing a dual-channel adaptive collaborative correction process, the predicted error information is used to synchronously correct the output value and adjust the model parameters online, which helps to suppress the long-term accumulation of errors and improves the ability of the power monitoring system to maintain high accuracy in different environments.

[0016] 4. By building and driving a battery protection model linked to high-fidelity SOC estimation values, a mechanism for dynamically generating safety boundaries is provided to replace the traditional fixed threshold protection method. This helps to better utilize the battery's available capacity and performance while ensuring safety, thereby having a positive effect on improving the safety and operating efficiency of the battery throughout its life cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of a method for high-precision power monitoring of series-connected batteries based on an adaptive algorithm provided in an embodiment of the present invention; Figure 2 A schematic diagram of a dual-channel adaptive collaborative correction process provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of a high-precision power monitoring system for series-connected batteries based on an adaptive algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] See also Figures 1 to 3 The present invention provides a high-precision battery capacity monitoring method and system based on an adaptive algorithm. The technical solution is as follows: Example 1: A high-precision battery capacity monitoring method based on an adaptive algorithm for series batteries. The specific process is as follows: Figure 1 Shown, including: Acquiring multi-dimensional parameter data of the series-connected battery pack, the multi-dimensional parameter data including the terminal voltage, current, and temperature of each single battery; Running a main state estimator based on an electrothermal coupling equivalent circuit model, wherein the main state estimator uses a Kalman filter algorithm to obtain a baseline state of charge value based on the multi-dimensional parameter data online; A probabilistic error prediction model is used to predict the estimated error of the baseline state of charge value based on the multidimensional parameter data, and an error probability distribution is output including an error mean and a variance used to characterize the prediction confidence; and a final high-fidelity state of charge estimate is generated using the error probability distribution through a dual-channel adaptive collaborative correction process; The final high-fidelity state of charge estimation value and multi-dimensional parameter data are used to construct and drive a battery protection model, and output safety control instructions, which are used to control and protect the series battery pack.

[0020] Furthermore, the electrothermal coupling equivalent circuit model includes: The temperature-sensitive parameter adaptation layer is used to provide circuit parameters for the electrical state calculation layer based on the measured battery temperature. It contains a parameter library that stores preset multi-dimensional mapping relationships between circuit parameters and temperature. It receives real-time temperature as query input, and through table lookup and interpolation calculation, retrieves and outputs the complete circuit parameters corresponding to the current temperature from the parameter library, including ohmic internal resistance, polarization resistance, polarization capacitance, and open circuit voltage state of charge curve.

[0021] The electrical state calculation layer is used to solve the battery's state space equation based on the current circuit parameters and real-time current, and output the terminal voltage predicted by the model; the topological structure of the second-order RC equivalent circuit is solidified internally, and the state variables include the state of charge and the polarization voltage of the two RC networks; it receives the circuit parameters provided by the temperature-sensitive parameter adaptation layer, and updates the state variables through integral operations based on the input real-time current, and calculates the predicted terminal voltage value according to Kirchhoff's voltage law.

[0022] The coupled heat generation calculation layer is used to calculate the total heat power generated by electrical activities; it internally integrates calculation logic based on Joule's law and thermodynamic principles; it receives real-time current and the internal resistance value and entropy heat coefficient corresponding to the current temperature, calculates irreversible Joule heat and reversible entropy heat respectively, and sums the two to output the total heat generation power value.

[0023] Specifically, the temperature-sensitive parameter adaptation layer provides the entire model with precise circuit parameters that dynamically change with temperature. The internal parameters of a battery (such as internal resistance and capacitance) are not constant but are closely related to temperature. To capture this characteristic, this method pre-calibrates experimentally and establishes a multidimensional mapping table, or parameter library, that stores the relationship between circuit parameters and temperature.

[0024] During system operation, this layer receives real-time battery temperature measurements from external sensors. Using this temperature as a query index, the system performs interpolation calculations when the temperature is between two calibration points, accurately retrieving and outputting a complete set of circuit parameters at the current temperature from the parameter library. This set of parameters includes: Ohmic internal resistance: characterizes the instantaneous voltage response inside the battery; Polarization resistance: the impedance that describes the electrochemical polarization and concentration polarization processes; Polarization capacitance: describes the dynamic characteristics and relaxation time of the polarization process; Open circuit voltage state of charge curve: describes the nonlinear relationship between open circuit voltage and state of charge of the battery when it is in equilibrium.

[0025] This level ensures that subsequent calculations of the model are always based on the parameters that best match the current thermodynamic state and is the basis of the model's adaptive capabilities.

[0026] The electrical state calculation layer is the core of the model, responsible for simulating the battery's electrical dynamic response. This layer internally implements the topology of a second-order RC equivalent circuit model, which effectively balances model accuracy and computational complexity. Its state variables include the battery's state of charge (SOC) and the polarization voltages on two RC networks, representing the fast and slow polarization processes, respectively. This layer's workflow involves first receiving real-time circuit parameter updates from the temperature-sensitive parameter adaptation layer and acquiring real-time charge and discharge currents from the current sensor. Based on the real-time input current, the battery's state of charge (SOC) is updated by integrating the current. Simultaneously, the two polarization voltage predictions are updated based on the RC network's state equations. Following Kirchhoff's voltage law, the model-predicted battery terminal voltage is calculated by algebraically summing the open-circuit voltage corresponding to the current SOC, the voltage drop across the ohmic internal resistance, and the two polarization voltages. Finally, the predicted terminal voltage output by this layer is used in subsequent state estimators (such as Kalman filters) to compare with the actual measured values ​​to correct and optimize the state estimate.

[0027] The coupled heat generation calculation layer is responsible for quantifying the heat generated by the internal electrical activity of the battery during operation. The heat generation of the battery mainly consists of two parts: irreversible Joule heat and reversible entropy heat. This layer integrates calculation logic based on Joule's law and thermodynamic principles, receives real-time current, and obtains the ohmic internal resistance, polarization resistance, and entropy heat coefficient corresponding to the current temperature from the temperature-sensitive parameter adaptation layer. Then, Joule heat is calculated. Joule heat is the heat generated when current flows through the internal resistance and is the main part of heat generation. Further, entropy heat is calculated. Entropy heat is the reversible endothermic or exothermic phenomenon accompanying the electrochemical reaction itself, which is related to the current direction, temperature, and entropy heat coefficient. Finally, the total thermal power is output. The calculated Joule heat and entropy heat are added to obtain the total heating power value of the battery at the current moment. This output value can not only be used to assess the thermal safety status of the battery, but also serve as input for a more complex fully coupled electrothermal model to further improve the model accuracy.

[0028] An electrothermal coupling equivalent circuit model is constructed, and a temperature-sensitive parameter adaptation layer is used to adjust internal parameters according to the real-time temperature to address the problem of model parameter drift in a wide temperature range working environment. The second-order RC network structure adopted can more carefully describe the dynamic responses such as battery polarization. The coupled heat generation calculation layer in the model quantifies the heat generated by the battery, providing data support for evaluating the thermal state of the battery and realizing active thermal management, which makes up for the lag of traditional methods that rely solely on temperature measurement.

[0029] Furthermore, the main state estimator adopts a Kalman filter algorithm, and the process of online estimating the baseline state of charge value based on the multi-dimensional parameter data includes: obtaining the state prior prediction value output by the electrothermal coupling equivalent circuit model after forward operation, the state prior prediction value contains the prior state of charge and the terminal voltage predicted by the model; comparing the terminal voltage predicted by the electrothermal coupling equivalent circuit model with the actual battery terminal voltage synchronously sampled by the hardware to calculate the measurement residual, calculating the optimal Kalman gain based on the measurement residual and the error covariance matrix, and using the optimal Kalman gain to perform weighted correction on the state prior prediction value to generate the baseline state of charge value.

[0030] Specifically, the main state estimator receives the output of the electrothermal coupling equivalent circuit model and uses the Kalman filter algorithm to optimally fuse the model prediction with the hardware measured data through a periodically iterative prediction and correction closed-loop process, thereby online estimating a baseline state of charge that is closer to the actual value.

[0031] The first stage of the process is the prediction stage. In this stage, the estimator first performs state prediction. It uses the battery state space equation defined in the electrical state calculation layer to infer the prior prediction value of the state at the current moment (k) based on the optimal state estimate value (including SOC and polarization voltage) at the previous moment and the real-time current input at the current moment (k).

[0032] Next, in the prediction phase, the estimator performs covariance prediction to update the error covariance matrix, which describes the degree of uncertainty in the state prediction value. As time goes by, the uncertainty of the prediction increases. After completing the prediction phase, a state prior prediction value including the prior state of charge and the model-predicted terminal voltage, as well as the corresponding uncertainty description, is obtained.

[0033] The second stage of the process is the correction stage. In this stage, the estimator uses the actual hardware measurement value to correct the prior prediction value given in the prediction stage to obtain the optimal estimate at the current moment. The first step of this stage is to calculate the measurement residual. The difference between the model-predicted terminal voltage contained in the state prior prediction value and the actual battery terminal voltage obtained by synchronous sampling of the hardware circuit is calculated to obtain the measurement residual. The measurement residual intuitively reflects the deviation between the model prediction and the actual situation.

[0034] Subsequently, the optimal Kalman gain is calculated based on the measurement residuals. The Kalman gain is a weight coefficient used to determine how much the measurement residuals should be trusted when correcting the state. The calculation of this gain integrates two aspects of uncertainty: one is the state prediction uncertainty calculated in the prediction phase (represented by the error covariance matrix P), and the other is the noise uncertainty inherent in the hardware measurement system itself (represented by the measurement noise covariance matrix R). When the model prediction uncertainty is large and the measurement noise is small, the Kalman gain will become larger, which means that the measurement value is more trusted; otherwise, the model prediction is more trusted.

[0035] After obtaining the optimal Kalman gain, a state weighted correction is performed. The measurement residual is multiplied by the calculated optimal Kalman gain to obtain a correction value. This correction value is then added to the prior state of charge in the state prior prediction value, and the final output is the baseline state of charge value.

[0036] Finally, at the end of the correction phase, the error covariance is updated. Based on the Kalman gain and this correction process, the error covariance matrix is ​​updated to reduce its uncertainty and provide a more accurate initial value for the prediction phase at the next moment. Specifically, the updated error covariance has a built-in Kalman filter standard algorithm and a Joseph morphology algorithm. Under normal stable working conditions, the system defaults to the standard algorithm to save computing resources and continuously monitors the size of the measurement residual during this correction process. Once it is detected that the residual exceeds the preset normal range, indicating an external shock or a sudden and drastic deviation in the model, it switches to the Joseph morphology algorithm for this covariance update to ensure the positive definiteness and convergence of the error covariance matrix. This method effectively prevents the risk of the filter diverging due to numerical calculation problems and greatly improves the robustness of the estimator under dynamic shocks.

[0037] Through the prediction and correction closed loop of Kalman filtering, the measured voltage is used to correct the cumulative error of the electrothermal coupling model caused by simplification or aging. At the same time, the dynamic characteristics of the model are used to suppress the interference of sensor noise on the estimation results. The core of this method is to dynamically weigh the confidence of the model and the measured value through the Kalman gain to obtain a baseline state of charge value that is statistically better, more stable and reliable than a single information source, providing a solid foundation for subsequent precise correction and protection strategies.

[0038] Furthermore, the probabilistic error prediction model includes: a state feedback and working condition fusion perception layer, which is used to receive real-time multi-dimensional parameter data synchronized with the main state estimator, and at the same time receive the diagonal elements of the Kalman gain and error covariance matrix P output by the main state estimator at the previous moment, and fuse them into a high-dimensional feature vector; a deep coupling error inference layer, which models and predicts the estimation error of the main state estimator at the current moment based on the high-dimensional feature vector; a confidence distribution generation layer, which is used to parse and encapsulate the prediction results of the deep coupling error inference layer, and output an error probability distribution containing the error mean and the variance used to characterize the prediction confidence.

[0039] Specifically, the function of the state feedback and operating condition fusion perception layer is to construct a high-dimensional feature vector for error prediction. First, the state feedback and operating condition fusion perception layer is synchronized with the main state estimator to receive real-time multi-dimensional parameter data (terminal voltage, current, temperature). At the same time, it obtains the Kalman gain and diagonal elements of the error covariance matrix P output by the main state estimator at the previous calculation moment. Finally, the above real-time operating condition data is fused with the internal state feedback data to form a high-dimensional feature vector.

[0040] Furthermore, the deep coupling error inference layer models and predicts the estimation error of the main state estimator at the current moment based on the input high-dimensional feature vector, receives the high-dimensional feature vector passed in by the previous layer, performs forward propagation calculation through the internal nonlinear mapping model, and outputs the predicted value of the estimation error at the current moment.

[0041] Furthermore, the confidence distribution generation layer parses and encapsulates the raw prediction results from the error inference layer, outputting them as a probability distribution. Specifically, it receives the prediction results output by the deeply coupled error inference layer and parses them into two independent values: the error mean and the error variance. Finally, this layer encapsulates the error mean and error variance together, outputting a complete error probability distribution, which includes a prediction of the error magnitude (mean) and a confidence representation of the prediction result (variance). The deeply coupled error inference layer is specifically implemented as an attention network enhanced with temporal features. The internal structure of the attention network includes a temporal convolution module that captures the dynamic changing trends of the input high-dimensional feature vector within a short time window and generates local operating condition features containing temporal information. In this embodiment, the high-dimensional feature vectors include voltage, current, and temperature. Subsequently, the high-dimensional feature vectors are fed into an attention weighting module, which learns the feature weight distribution and dynamically adjusts the attention weights of different features based on the criticality of the operating condition. For example, when the battery is nearing full charge or discharge, the attention module automatically increases the weights of features related to the terminal voltage and the internal polarization voltage of the model. In this way, the model achieves a transition from passively receiving all information to actively focusing on key information, making its prediction logic closer to the actual physical and chemical processes of the battery, thereby significantly improving the accuracy and robustness of error prediction under complex and changeable working conditions.

[0042] By constructing a probabilistic error prediction model, the real-time working conditions and the internal state of the estimator are integrated, and nonlinear mapping is used to actively predict the potential errors of the main estimator. It not only outputs the error prediction value, but also provides a credibility quantification of the prediction, providing a key decision-making basis for the subsequent correction process, enabling it to perform intelligent compensation and adaptive adjustment according to the prediction credibility, thereby improving the effectiveness of the correction and the overall robustness of the system.

[0043] Furthermore, the dual-channel adaptive collaborative correction process is as follows: Figure 2As shown, the process of generating a final high-fidelity state of charge estimation value using the error probability distribution includes: obtaining the error probability distribution output by the probabilistic error prediction model, which includes an error mean and an error variance, and synchronously starting and executing calculations of two correction channels based on the error probability distribution; in the output-end gated correction channel, converting the error variance into a confidence fusion weight through a preset inverse function, multiplying the fusion weight by the error mean to obtain a weighted error correction, compensating the weighted error correction to the baseline state of charge value, and generating a corrected state of charge value; in the model-end online adaptive channel, using the same error variance as a dynamic indicator characterizing the current mismatch degree of the electrothermal coupling equivalent circuit model, mapping it to a process noise adjustment value through a preset monotonically increasing function, and superimposing the noise adjustment value on the basic process noise covariance matrix Q of the Kalman filter algorithm.

[0044] Specifically, the core of the dual-channel adaptive collaborative correction process is to synchronously start and execute two parallel and complementary correction channels based on the error probability distribution, including the output-end gated correction channel and the model-end online adaptive channel.

[0045] The function of the output-end gated correction channel is to directly compensate and correct the baseline state of charge value output by the main state estimator to eliminate the prediction error. The channel first obtains the baseline state of charge value output by the main state estimator and the error probability distribution output by the probability error prediction model. Then, the error variance value in the error probability distribution is mapped through a preset inverse function and converted into a confidence fusion weight ranging from 0 to 1. The characteristic of the inverse function is that the smaller the input error variance, the higher the credibility of the error prediction and the closer the output weight value is to 1; conversely, the larger the error variance, the closer the weight value is to 0. Subsequently, the calculated confidence fusion weight is multiplied by the error mean to obtain the final weighted error correction amount. Finally, the weighted error correction amount is compensated to the input baseline state of charge value, and the final output is a high-fidelity state of charge estimation value. The inverse function is specifically implemented as a nonlinear mapping function with adjustable threshold and sensitivity, specifically a parameterized tuned Sigmoid function, and the specific formula is: ; in, Represents the final generated confidence fusion weight, represents the Sigmoid function, Representative sensitivity, obtained by combining offline experimental analysis and manual engineering adjustment, represents the error variance value, It represents the maximum acceptable error variance of the system and is pre-calibrated based on a large amount of offline experimental data.

[0046] This design allows developers to precisely define the extent to which they trust error predictions based on battery type and application scenarios, making the compensation behavior of the correction amount no longer fixed but configurable and optimizable, making the entire adaptive correction process more intelligent.

[0047] The purpose of the model-side online adaptive channel is not to directly correct the output results, but to indirectly optimize the estimation performance of the main state estimator at the next moment by adjusting its internal parameters in real time. The model-side online adaptive channel first obtains the error variance value in the error probability distribution and uses it as a dynamic indicator to characterize the current mismatch degree of the electrothermal coupling equivalent circuit model. Then, using a preset monotonically increasing function, specifically the Softplus function in this embodiment, the error variance value is mapped to a process noise adjustment value. The larger the error variance, the more severe the model mismatch, and the larger the output process noise adjustment value. The obtained process noise adjustment value is then superimposed on the basic process noise covariance matrix Q of the Kalman filter algorithm. In this way, the process noise covariance matrix Q is increased, causing the Kalman filter to reduce its trust in the model prediction in the next calculation cycle and instead increase its trust in the actual hardware measurement value, thereby achieving online adaptive adjustment of the main state estimator.

[0048] By establishing a dual-channel adaptive collaborative correction process, the output-side gated correction channel uses the confidence of the error prediction to intelligently and non-blindly compensate for the current state of charge value, ensuring the immediate accuracy of each output. At the same time, the model-side online adaptive channel uses the error variance as an indicator of model mismatch and dynamically adjusts the process noise of the Kalman filter, allowing the estimator to trust the measured data more when the model performs poorly, significantly improving the system's robustness and estimation accuracy throughout the battery life cycle and under complex operating conditions.

[0049] Furthermore, the process of using the final high-fidelity state of charge estimate and the multi-dimensional parameter data to construct and drive a battery protection model and output safety control instructions includes: inputting the final high-fidelity state of charge estimate and the real-time temperature and battery health status information in the multi-dimensional parameter data as a combined feature vector into a neural network model for inferring safety boundaries, and the neural network dynamically generates in real time the safe operating boundaries of the battery in the current state through forward propagation calculation. The safe operating boundaries include a charging cutoff SOC threshold, a discharging cutoff SOC threshold, a maximum charge and discharge current allowed under the current operating conditions, and a maximum / minimum operating temperature that are dynamically adjusted with the state; comparing the final high-fidelity state of charge estimate and the real-time current and temperature with the dynamic safe operating boundaries generated in real time. If it is detected that the real-time parameters have a tendency to cross the corresponding safety boundaries, the corresponding protection logic is triggered, and a safety control instruction is generated to directly control the charging and discharging circuits.

[0050] Specifically, the battery protection model structure includes three layers: a hardware protection layer, a state-aware software protection layer, and a dynamic safety boundary prediction protection layer.

[0051] The hardware protection layer monitors the total current in real time through hardware for overcurrent or short-circuit protection. Once the current exceeds a preset hardware threshold (for example, a short-circuit detection threshold that is much larger than the normal operating current), the protection circuit will be triggered instantaneously, directly disconnecting the MOSFET switch of the main circuit without waiting for software intervention.

[0052] In terms of overvoltage or undervoltage protection, the hardware also monitors the battery voltage. When extreme overcharge or overdischarge occurs, causing the voltage to reach the limit voltage threshold set by the hardware, the same hardware instantaneous cut-off action is performed.

[0053] The state-aware software protection layer is executed by the main controller software and no longer relies solely on fixed thresholds. Instead, it uses intermediate data such as high-fidelity state of charge estimates output by the previous process to achieve more intelligent protection that is aware of the battery state.

[0054] For overcharge protection, if the real-time voltage of any single battery is monitored to exceed the preset charging cut-off voltage, or the high-fidelity state of charge estimation value output by the dual-channel adaptive collaborative correction process reaches or exceeds its upper limit threshold, if any condition is met, the protection is triggered and charging is stopped.

[0055] For over-discharge protection, if the real-time voltage of any single battery is monitored to be lower than the preset discharge cut-off voltage, or the high-fidelity state of charge estimation value reaches or falls below its lower limit threshold, if any of the conditions are met, the protection is triggered and discharge is stopped.

[0056] For overcurrent protection, state-based charging and discharging overcurrent thresholds are set at the software level. The size of the threshold will be dynamically adjusted according to the current high-fidelity state of charge estimate and the real-time temperature. For example, in the high or low SOC range, or when the temperature is high, the allowable continuous current threshold will be lowered accordingly. When the real-time current continues to exceed the dynamically adjusted threshold for a certain period of time, protection is triggered.

[0057] For temperature protection, when the battery temperature is detected to exceed the set maximum operating temperature or be lower than the minimum operating temperature, the system will further combine the current high-fidelity state of charge estimate to make a judgment. For example, in a low temperature environment (such as below 5°C), if the SOC is at a high level, it will trigger stricter charging restrictions or prohibit charging protection to prevent low-temperature lithium deposition.

[0058] The dynamic security boundary prediction protection layer is the core intelligent protection layer of this method. It uses the high-precision state cognition obtained to achieve predictive protection against security risks rather than simple threshold triggering.

[0059] First, a dynamic safety margin is generated. The final high-fidelity state of charge estimate output by the dual-channel adaptive collaborative correction process, the real-time temperature in the multi-dimensional parameter data, and the estimated battery state of health (SOH) information are combined into a combined feature vector and input into a pre-trained neural network model for inferring the safety margin.

[0060] Through forward propagation calculations, this neural network model dynamically and in real time generates the complete safe operating boundaries of the battery under the current state. These boundaries are a multi-dimensional set of thresholds that adjust dynamically with the battery's state and operating conditions. These include: charge cutoff SOC threshold, discharge cutoff SOC threshold, maximum allowable charge and discharge current under the current operating conditions, and maximum and minimum allowable operating temperatures. These dynamic boundaries are more precise and intelligent than the fixed thresholds in the second stage.

[0061] Next, this layer triggers predictive protection, continuously comparing the final high-fidelity SOC estimate, along with real-time current and temperature, with the dynamic safe operating boundaries generated in real time in the previous step. If any real-time parameter is detected to be trending across its corresponding safety boundary (for example, the system predicts that the dynamic charge cutoff SOC threshold will be reached within a short period of time based on the current charging rate), the system triggers the corresponding protection logic in advance. Finally, after triggering the protection logic, the system generates a clear safety control instruction. This instruction is sent directly to the actuator to control the charge and discharge circuit. For example, by disconnecting relays or adjusting the power of the charge and discharge modules, it can force the battery back into the safe operating range, thereby achieving precise and proactive battery protection.

[0062] Through the coordinated work of the above three-level protection strategies, this method not only achieves comprehensive protection against abnormal conditions such as overcharging, over-discharging, overcurrent, short circuit and temperature, but also greatly improves the timeliness, accuracy and intelligence of protection through the combination of software and hardware and dynamic prediction.

[0063] By building a complete adaptive algorithm framework, the battery monitoring and protection performance is comprehensively improved. The self-optimizing closed-loop system can adapt to battery aging and temperature changes, achieving high-precision estimation throughout the entire life cycle, effectively solving the problems of insufficient accuracy and error accumulation in traditional solutions. At the same time, the dynamic safety boundary based on neural networks upgrades the protection strategy from passive threshold triggering to active risk prediction, significantly enhancing safety. The combination of multi-level software and hardware protection and the online adaptive capabilities of the algorithm ensures the robustness of the system under complex working conditions, providing a high-precision and high-reliability management solution for series battery packs.

[0064] Example 2: In order to better realize the power supply monitoring of the battery pack of 8 lithium-ion batteries in series at the outdoor monitoring station, the high-precision power monitoring system of series batteries based on the adaptive algorithm provided by the present invention is introduced. The specific system structure is as follows: Figure 3 shown.

[0065] Initial state of charge (SOC) of the monitoring station's battery pack, at an early morning field temperature of -15°C. The monitoring station is in low-power data acquisition mode.

[0066] The system began real-time monitoring of the terminal voltage, charging current, and surface temperature of each battery cell in the battery pack. After charging began, the battery temperature slowly rose from -15°C. When the temperature reached -12°C, the temperature-sensitive parameter adaptation layer detected this change. Using this temperature as an index, the system retrieved and updated the circuit parameters of the entire electrothermal coupling model from its internal parameter library through table lookup and interpolation calculations. This provided subsequent calculations with internal resistance, capacitance, and OCV-SOC curves that precisely matched the current extremely low temperature environment.

[0067] The solar charging management module outputs the charging current. The electrical state calculation layer receives this current value and the circuit parameters at -12°C provided by the previous layer. It performs forward operations using the state-space equations, outputs a model-predicted terminal voltage, and derives an a priori SOC estimate.

[0068] This priori SOC and predicted terminal voltage are passed to the main state estimator. The estimator compares the model-predicted terminal voltage with the actual terminal voltage sampled by the hardware and detects a deviation due to low-temperature model errors. The Kalman filter algorithm calculates the optimal Kalman gain based on this deviation (measurement residual) and system uncertainty, and corrects the priori SOC, ultimately outputting a more reliable baseline state-of-charge value.

[0069] Furthermore, the probabilistic error prediction model is started, and its state feedback and operating condition fusion perception layer receives the current real-time operating condition data (voltage, charging current, temperature), and simultaneously obtains the Kalman gain and diagonal elements of the error covariance matrix P output by the main state estimator at the previous moment, and fuses these data into a high-dimensional feature vector.

[0070] This feature vector is fed into the Deep Coupled Error Inference layer, where a neural network model calculates and predicts the possible estimation error of the current main state estimator. For example, a medium-sized error variance indicates that the prediction result has a certain degree of uncertainty.

[0071] The confidence distribution generation layer encapsulates the above results into error probability distribution and passes it to the dual-channel adaptive collaborative correction process.

[0072] Output-side gated correction channel: Based on a medium-sized error variance, an inverse function is used to calculate a medium-sized confidence fusion weight, such as 0.6. This weight is then multiplied by the error mean to obtain a weighted error correction. Finally, this correction is applied to the baseline SOC value. This results in a high-fidelity SOC estimate for the current output.

[0073] Model-side online adaptive channel: Simultaneously, based on the same medium-sized error variance, a corresponding process noise adjustment value is mapped using a monotonically increasing function. This adjustment value is superimposed on the Kalman filter's process noise covariance matrix Q, causing the filter to appropriately reduce its confidence in the current low-temperature model in the next cycle and instead rely more on hardware measurements.

[0074] Furthermore, the battery has multiple protection features. For example, if a short circuit occurs at the battery output due to aging circuits or animal bites, the hardware protection layer's hardware circuitry detects a significant short-circuit current far exceeding the normal range. Within microseconds, its hardware comparator immediately triggers and disconnects the MOSFET switch in the main circuit, cutting off power and preventing the battery from catching fire or exploding due to the short circuit. This entire process requires no software intervention, resulting in an extremely fast response.

[0075] Regarding the second level of protection provided by the state-aware software protection layer, for example, after prolonged sunlight exposure, the battery charge reaches a high level. If one cell, due to rapid aging, reaches its preset software charge cutoff voltage first, the state-aware software protection layer detects the voltage exceeding the limit for that cell and immediately triggers overcharge protection (OVP), generating a command to stop charging. Simultaneously, it initiates cell balancing to discharge that cell to reduce inconsistencies within the battery pack. For another example, during charging, due to factors such as low temperature or battery aging, no cell voltage may reach the absolute charge cutoff voltage. However, the high-fidelity state of charge estimate output by the method of the present invention accurately reaches 100%. In this case, the state-aware software protection layer determines that the battery is fully charged based on this high-precision SOC data, rather than relying solely on voltage. This triggers overcharge protection (OVP) and generates a command to stop charging, preventing overcharging caused by model bias or inaccurate voltage measurement, thereby improving protection accuracy.

[0076] For the third level of protection of the dynamic safety boundary prediction protection layer, for example, the ambient temperature of the monitoring station is as high as 40°C, and it is exposed to the scorching sun for a long time, causing the battery temperature to quickly rise to 58°C during the charging process.

[0077] The neural network model in the dynamic safety margin prediction protection layer receives information such as the current high-fidelity SOC (e.g., 85%), high temperature (58°C), and battery state of health (SOH, e.g., 88%), and immediately recalculates and outputs a new dynamic safety margin. This margin indicates that, under the current conditions, to prevent thermal runaway, the maximum allowable operating temperature should be lowered from the conventional 60°C to 55°C. The system compares this dynamic margin with the real-time monitored battery temperature of 58°C and discovers that the real-time temperature has exceeded the dynamic safety margin. The system determines that the battery is at risk of overheating, triggers the protection logic in advance, and generates safety control instructions. On the one hand, it sends a high-temperature warning to the backend management center via the wireless module, and on the other hand, it directly controls the charging module to reduce the charging current or even suspend charging until the battery temperature drops back to within the new dynamic safety margin.

[0078] By building a self-optimizing closed-loop estimation system, high-precision status perception of the battery throughout its entire life cycle is achieved, and the protection strategy is upgraded from passive response to active prediction, thereby significantly enhancing the reliability and safety of the system under complex working conditions.

[0079] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A high-precision power monitoring method for series-connected batteries based on an adaptive algorithm, characterized in that: include: Acquiring multi-dimensional parameter data of the series-connected battery pack, the multi-dimensional parameter data including the terminal voltage, current, and temperature of each single battery; Running a main state estimator based on an electrothermal coupling equivalent circuit model, wherein the main state estimator uses a Kalman filter algorithm to obtain a baseline state of charge value based on the multi-dimensional parameter data online; Using a probabilistic error prediction model, predict an estimation error of the baseline state of charge value based on the multidimensional parameter data, and output an error probability distribution including an error mean and a variance used to characterize the prediction confidence; The error probability distribution is used to generate a final high-fidelity state of charge estimate through a dual-channel adaptive collaborative correction process; The final high-fidelity state of charge estimation value and multi-dimensional parameter data are used to construct and drive a battery protection model, and output safety control instructions, which are used to control and protect the series battery pack.

2. The method for high-precision power monitoring of series-connected batteries based on an adaptive algorithm according to claim 1, characterized in that: The electrothermal coupling equivalent circuit model includes: The temperature-sensitive parameter adaptation layer contains a parameter library that stores preset multi-dimensional mapping relationships between circuit parameters and temperature. It receives real-time temperature as a query input, retrieves and outputs the complete circuit parameters corresponding to the current temperature from the parameter library; The electrical state calculation layer receives circuit parameters and updates state variables through integration operations based on the real-time current input, and calculates the predicted terminal voltage value according to Kirchhoff's voltage law; The coupled heat generation calculation layer is used to calculate the total heat power generated by electrical activities; it receives the real-time current and the internal resistance value and entropy heat coefficient corresponding to the current temperature, calculates the irreversible Joule heat and reversible entropy heat respectively, and sums the two to output the total heat generation power value.

3. The method for high-precision power monitoring of series-connected batteries based on an adaptive algorithm according to claim 1, characterized in that: The main state estimator adopts a Kalman filter algorithm, and the process of online estimating the baseline state of charge value based on the multi-dimensional parameter data includes: obtaining the state prior prediction value output by the electrothermal coupling equivalent circuit model after forward operation, the state prior prediction value contains the prior state of charge and the terminal voltage value predicted by the model; comparing the terminal voltage predicted by the electrothermal coupling equivalent circuit model with the actual battery terminal voltage synchronously sampled by the hardware and calculating the measurement residual, calculating the optimal Kalman gain based on the measurement residual and the error covariance matrix, and using the optimal Kalman gain to perform weighted correction on the state prior prediction value to generate the baseline state of charge value.

4. The method for high-precision power monitoring of series-connected batteries based on an adaptive algorithm according to claim 1, characterized in that: The probability error prediction model includes: The state feedback and working condition fusion perception layer is used to receive real-time multi-dimensional parameter data synchronized with the main state estimator, and at the same time receive the diagonal elements of the optimal Kalman gain and error covariance matrix output by the main state estimator at the previous moment, and fuse them into a high-dimensional feature vector; the deep coupling error inference layer is used to model and predict the estimation error of the main state estimator at the current moment based on the high-dimensional feature vector; the confidence distribution generation layer is used to parse and encapsulate the prediction results of the deep coupling error inference layer, and output an error probability distribution containing the error mean and the variance used to characterize the prediction confidence.

5. The method for high-precision power monitoring of series-connected batteries based on an adaptive algorithm according to claim 1, characterized in that: The process of generating a final high-fidelity state of charge estimation value using the error probability distribution through the dual-channel adaptive collaborative correction process includes: Based on the error probability distribution, which includes error variance and error mean, calculations of two correction channels are performed; in the output-end gated correction channel, the error variance is converted into a confidence fusion weight through a preset inverse function, the fusion weight is multiplied by the error mean to obtain a weighted error correction, and the weighted error correction is compensated to the baseline state of charge value to generate a final high-fidelity state of charge estimation value; in the model-end online adaptive channel, the error variance is used as a dynamic indicator of the mismatch degree of the electrothermal coupling equivalent circuit model, and is mapped to a process noise adjustment value through a preset monotonically increasing function, and the noise adjustment value is superimposed on the noise covariance matrix of the Kalman filter algorithm.

6. The method for high-precision power monitoring of series-connected batteries based on an adaptive algorithm according to claim 1, characterized in that: The process of using the final high-fidelity state of charge estimation value and the multi-dimensional parameter data to construct and drive the battery protection model and output safety control instructions includes: The final high-fidelity state of charge estimate and the real-time temperature and battery health status information in the multi-dimensional parameter data are input as a combined feature vector into a neural network model for inferring safety boundaries to generate a safe operating boundary for the battery in the current state. The safe operating boundary includes a charging cutoff SOC threshold, a discharging cutoff SOC threshold, a maximum charge and discharge current allowed under the current operating conditions, and a maximum / minimum operating temperature that are dynamically adjusted according to the state. The final high-fidelity state of charge estimate and the real-time current and temperature are compared with the safe operating boundary. If it is detected that the real-time parameters have a tendency to cross the corresponding safety boundary, the corresponding protection logic is triggered, and a safety control instruction is generated to directly control the charging and discharging circuit.

7. A high-precision battery power monitoring system based on an adaptive algorithm is characterized by: include: A data acquisition module, which acquires multi-dimensional parameter data of the series-connected battery pack, wherein the multi-dimensional parameter data includes the terminal voltage, current and temperature of each single battery; a state estimation module, running a main state estimator based on an electrothermal coupling equivalent circuit model, wherein the main state estimator uses a Kalman filter algorithm to online estimate a baseline state of charge value based on the multi-dimensional parameter data; an error correction module, which uses a probabilistic error prediction model to predict an estimation error of the baseline state of charge value based on the multidimensional parameter data, and outputs an error probability distribution including an error mean and a variance used to characterize the prediction confidence; The error probability distribution is used to generate a final high-fidelity state of charge estimate through a dual-channel adaptive collaborative correction process; The safety control module uses the final high-fidelity state of charge estimation value and multi-dimensional parameter data to build and drive a battery protection model and output safety control instructions, wherein the safety control instructions are used to control and protect the series battery pack.

Images