A charger autonomous adjustment system based on an adaptive algorithm and a control method thereof

Abstract

This invention discloses a charger autonomous adjustment system and its control method based on an adaptive algorithm. The system includes a power conversion module, a data acquisition module, a state estimation module, an optimization control module, a PWM drive module, and a safety protection module. The power conversion module can be connected to an external power supply and converted to a high-frequency switching DC power supply. It collects the instantaneous values ​​of the signal generation terminal voltage, charging current, and temperature of the battery being charged in real time and updates the adaptive battery model parameter vector in real time. Using the adaptive battery model parameter vector as dynamic input, it solves to generate an optimal charging current reference sequence, where the first element serves as the current command and is compared using vector pulse width modulation to generate a pulse control signal to drive autonomous adjustment. It monitors the instantaneous terminal voltage, charging current, and temperature in real time and in parallel; if a preset threshold is exceeded, the charging circuit is immediately cut off. This invention enables flexible adaptation and autonomous adjustment of the charging power supply.

Description

Technical Field

[0001] This invention relates to the field of intelligent charging control technology, specifically to a charger autonomous adjustment system and control method based on an adaptive algorithm. Background Technology

[0002] With the increasing prominence of advantages such as high energy density, long cycle life, and low environmental pollution, lithium batteries have been widely used in multiple core fields such as consumer electronics, electric vehicles, and energy storage systems, becoming a key component in the modern energy storage and supply system. As the core supporting equipment in the lithium battery charging process, the charger's level of intelligence and adjustment precision directly determine the charging efficiency, cycle life, and safety of the lithium battery. Especially against the backdrop of ever-increasing demands for charging convenience and reliability in various scenarios, the limitations of traditional chargers are becoming increasingly apparent, necessitating the development of intelligent charging technology with autonomous adjustment capabilities to adapt to the dynamic usage needs of lithium batteries.

[0003] Currently, mainstream lithium battery charging equipment in the industry is mainly divided into two categories. One is the traditional non-communication charger (commonly known as "blind charging"). This type of charger uses a fixed voltage and current output mode. Its charging logic was originally designed for lead-acid batteries and includes a "balancing charging" stage that is unnecessary for lithium batteries, as well as a dangerous "float charging" stage. Due to the lack of adaptability to the characteristics of lithium batteries, directly using this type of non-communication charger to charge lithium batteries can easily lead to safety hazards such as overcharging, excessive temperature rise, and thermal runaway. At the same time, it cannot match the optimal charging curve of lithium batteries, which will significantly shorten the battery cycle life and fail to meet the safe charging requirements of lithium batteries.

[0004] To address the shortcomings of traditional non-communication chargers, the industry has proposed several smart charger solutions. Their core logic involves interacting with the battery's built-in Battery Management System (BMS) to obtain status information such as battery voltage, temperature, and remaining charge (SOC), thereby dynamically adjusting the charging strategy to adapt to lithium batteries. However, existing charging technologies, whether traditional non-communication chargers or BMS-dependent smart chargers, suffer from fixed charging strategies and lack dynamic adaptability. They cannot adjust the charging curve in real time based on dynamic factors such as battery aging and changes in ambient temperature, resulting in low charging efficiency and premature battery degradation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a charger autonomous adjustment system based on an adaptive algorithm, comprising the following modules: The power conversion module is used to connect to an external power source and convert the input AC or DC power into a high-frequency switching DC power source with adjustable amplitude and frequency, providing programmable charging voltage and charging current output to the battery being charged. The data acquisition module is used to acquire the terminal voltage signal of the battery being charged, the current signal in the charging circuit, and the battery surface temperature signal in real time. It performs anti-aliasing filtering, isolation amplification, and analog-to-digital conversion on the acquired signals to generate instantaneous terminal voltage, instantaneous charging current, and instantaneous temperature values. It also includes: Based on the instantaneous values ​​of terminal voltage and charging current, a recursive least squares algorithm with a forgetting factor is used to perform online real-time estimation and updating of the second-order RC equivalent circuit model constructed to characterize the internal dynamic characteristics of the battery, generating an adaptive battery model parameter vector containing ohmic internal resistance, electrochemical polarization internal resistance, concentration polarization internal resistance and their corresponding time constants. The state estimation module is used to take the adaptive battery model parameter vector as dynamic input, combine the instantaneous terminal voltage value and the instantaneous charging current value, and use the extended Kalman filter algorithm (EKF) to jointly estimate the battery's state of charge, electrochemical polarization voltage and concentration polarization voltage, and generate a joint battery state estimation vector, which includes the estimated value of the state of charge and its estimation error covariance. The optimization control module is used to construct a finite-time rolling optimization problem under the model predictive control algorithm MPC based on the joint estimation vector of battery state and the adaptive battery model parameter vector. The objective function of the optimization problem includes the tracking error of the target state of charge trajectory, the penalty term for the magnitude of the charging current, and the smoothing penalty term for the rate of change of current. The optimal charging current reference sequence in the next few control cycles is generated by solving the optimization problem by combining the instantaneous temperature value. The PWM drive module is used to receive the first element of the optimal charging current reference sequence as the current command for the current control cycle, compare the current command with the instantaneous value of the charging current measured from the charging circuit by vector pulse width modulation, and generate a pulse control signal to drive the switching power devices in the power conversion module to adjust autonomously. The safety protection module is used independently of the closed-loop control system composed of all the above modules. It monitors the instantaneous values ​​of terminal voltage, charging current, and temperature in real time and in parallel. Once any instantaneous value exceeds the preset hardware protection threshold, a fault lock signal is immediately generated, directly controlling the hardware protection circuit to cut off the charging circuit and lock the fault state.

[0006] Furthermore, the present invention also provides a charger autonomous adjustment control method based on an adaptive algorithm. This method is implemented based on the charger autonomous adjustment system based on the adaptive algorithm described above. The charger autonomous adjustment control method based on the adaptive algorithm includes the following steps: S01: The instantaneous values ​​of the terminal voltage, charging current, and temperature of the battery being charged are collected in real time through the data acquisition module; it also includes an online parameter correction step: during the charging process, the adaptive battery model parameter vector containing the ohmic internal resistance, electrochemical polarization internal resistance, concentration polarization internal resistance and their corresponding time constants of the second-order RC equivalent circuit model is periodically updated using a recursive least squares algorithm with a forgetting factor, and it is synchronized to S02 and S03; S02: The state estimation module is based on a second-order RC equivalent circuit model and uses an extended Kalman filter algorithm to estimate the state of charge and polarization voltage of the battery in real time. S03: The optimization control module dynamically generates the optimal charging current sequence based on the real-time status output by S02 using a model predictive control algorithm; S04: The PWM drive module receives the first element in the optimal charging current reference sequence as the current command for the current control cycle and adjusts the pulse control signal of the power conversion module. S05: The safety protection module independently monitors the instantaneous values ​​of terminal voltage, charging current, and temperature, and cuts off the charging circuit when the values ​​exceed the safety threshold. S06: Repeat S01 to S05 to form a closed-loop control.

[0007] The beneficial effects of this application are as follows: 1. The power conversion module connects to an external power supply, converting the input AC or DC power into a high-frequency switching DC power supply with adjustable amplitude and frequency. This provides programmable charging voltage and current output to the battery being charged. The module can flexibly switch between different types of input power, adapt to various power supply scenarios, and output programmable charging parameters. It can completely eliminate the unreasonable charging stages of the past, avoid unnecessary damage to the lithium battery, and adjust the charging voltage and current autonomously without relying on the BMS. This enables flexible adaptation of the charging power supply, provides basic support for the execution of subsequent dynamic charging strategies, thereby avoiding safety hazards, improving charging efficiency, and reducing battery life loss.

[0008] 2. The data acquisition module collects the terminal voltage signal of the battery being charged, the current signal in the charging circuit, and the battery surface temperature signal in real time. After anti-aliasing filtering, isolation amplification, and analog-to-digital conversion, instantaneous values ​​are generated. It also uses a recursive least squares algorithm with a forgetting factor to estimate and update the second-order RC equivalent circuit model online. This module first ensures the accuracy and stability of the collected data through signal processing, comprehensively captures the external state of the battery, and then uses algorithms to deeply explore the internal dynamic characteristics of the battery, updating the equivalent circuit model parameters in real time. This fully reflects the characteristic changes caused by battery aging and temperature changes. The generated adaptive battery model parameter vector provides comprehensive and realistic basic data for subsequent state estimation and optimized control, allowing the charging strategy to dynamically adjust according to the battery state, effectively solving the shortcomings of existing technologies.

[0009] 3. The state estimation module uses the adaptive battery model parameter vector as dynamic input, combined with the instantaneous values ​​of terminal voltage and charging current, to jointly estimate the battery's state of charge, electrochemical polarization voltage, and concentration polarization voltage using the Extended Kalman Filter (EKF) algorithm. This module relies on adaptively updated battery model parameters, breaking the limitations of traditional indirect inference. Through advanced algorithms, it achieves joint estimation of multiple state parameters, not only obtaining the estimated state of charge but also capturing the dynamic changes in polarization voltage, while providing the estimation error covariance to ensure the reliability of the state estimation. The generated joint battery state estimation vector comprehensively reflects the real-time battery state, providing a scientific basis for the optimization control module to formulate reasonable charging strategies. This allows charging parameter adjustments to closely match the current battery state, avoiding overcharging and undercharging, reducing damage to the battery from polarization, and ultimately improving charging efficiency and battery cycle life.

[0010] 4. The optimized control module, based on the joint estimation vector of battery state and the adaptive battery model parameter vector, constructs a finite-time rolling optimization problem under the Model Predictive Control (MPC) algorithm. It combines instantaneous temperature values ​​to generate an optimal charging current reference sequence. This module avoids the drawbacks of fixed charging curves by constructing an objective function that includes target state-of-charge trajectory tracking error, charging current amplitude penalty, and current change rate smoothing penalty, taking into account the real-time battery state and internal characteristics. This ensures charging speed while reducing battery loss and guaranteeing charging stability. Simultaneously, it dynamically adjusts the optimization direction based on instantaneous temperature values ​​to avoid damage to the battery from high-temperature charging. By solving the optimization problem, it generates an optimal charging current reference sequence for multiple future control cycles, enabling proactive adjustments to the charging strategy, addressing the lag issue of existing charging strategies, balancing charging efficiency and battery life, and adapting to the dynamic usage requirements of lithium batteries.

[0011] 5. The PWM drive module receives the first element of the optimal charging current reference sequence as the current command for the current control cycle. It generates a pulse control signal to drive the switching power devices of the power conversion module through vector pulse width modulation comparison. This module compares the optimal current command generated by the optimization control module with the measured instantaneous charging current value in real time. Using vector pulse width modulation technology, it quickly generates a pulse control signal to drive the switching power devices of the power conversion module, achieving autonomous adjustment of the charging current. This ensures that the actual charging current strictly follows the optimal reference sequence, reducing damage to the battery caused by current fluctuations. Its rapid response capability can promptly adapt to the strategy adjustments of the optimization control module, ensuring that the charging current is in an optimal state in each control cycle. This allows the dynamically optimized charging strategy to be truly implemented, further improving charging efficiency, ensuring battery safety, and reducing battery life loss caused by current fluctuations.

[0012] 6. The safety protection module is independent of the closed-loop control system. It monitors the instantaneous values ​​of terminal voltage, charging current, and temperature in real time and in parallel. When these values ​​exceed the limits, it immediately generates a fault lockout signal, cuts off the charging circuit, and latches the fault. Existing chargers often rely on the main control system for safety protection. Once the main control system fails, the protection function fails, and the response is delayed. Protection is often triggered only after dangerous situations such as overcharging, excessive temperature rise, or thermal runaway occur, failing to promptly curb the escalation of the fault and seriously threatening the safety of the battery and charger. This module adopts an independent design, achieving real-time parallel monitoring without relying on other modules. It avoids protection failures caused by main control system malfunctions, quickly detecting abnormalities such as excessive voltage, excessive current, and excessive temperature. It immediately triggers the protection mechanism without complex calculations, cutting off the charging circuit and latching the fault state to prevent further escalation of the fault and effectively avoid various safety hazards. Simultaneously, fault state latching facilitates subsequent troubleshooting and maintenance, further improving the safety and reliability of the lithium battery charging process, providing dual protection for battery use safety, and addressing the shortcomings of existing protection mechanisms such as delayed response and insufficient reliability. Attached Figure Description

[0013] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the module structure of the charger autonomous adjustment system based on the adaptive algorithm in this embodiment; Figure 2 This is a schematic diagram of the electrical principle of the second-order RC equivalent circuit model in this embodiment; Figure 3 for Figure 1 Functional flowchart of the optimization control module; Figure 4 This is a flowchart illustrating the steps of the charger autonomous adjustment control method based on an adaptive algorithm in this embodiment. Detailed Implementation

[0014] The following drawings disclose several embodiments of the present invention. For clarity, many practical details will be described in the following description. However, it should be understood that these practical details are not intended to limit the invention. That is, in some embodiments of the invention, these practical details are not essential. Furthermore, for the sake of simplicity, some conventional structures and components will be shown in the drawings in a simple schematic manner.

[0015] To further understand the invention's content, features, and effects, the following embodiments are provided, and detailed descriptions are given below in conjunction with the accompanying drawings: Reference Figure 1 , Figure 1 This is a schematic diagram of the module structure of the charger autonomous adjustment system based on the adaptive algorithm in this embodiment. The charger autonomous adjustment system based on the adaptive algorithm in this embodiment includes the following modules: The power conversion module is used to connect to an external power source and convert the input AC or DC power into a high-frequency switching DC power source with adjustable amplitude and frequency, providing programmable charging voltage and charging current output to the battery being charged. In this embodiment of the invention, the power conversion module integrates a rectifier unit, a filter unit, and a high-frequency inverter unit. One end is connected to an external power supply, and the other end is connected to the battery being charged. Its core function is to convert the input power supply into a high-frequency switching DC power supply that meets charging requirements. The external power supply can be AC ​​220V / 50Hz or DC 12V. When AC is input, it first undergoes full-bridge rectification by the rectifier unit. The rectification formula is as follows: ,in AC input voltage, The rectified DC voltage is filtered by an electrolytic capacitor with a capacitance of 1000μF to remove ripple and obtain a stable DC voltage. The high-frequency inverter unit adopts a full-bridge inverter topology, using insulated-gate bipolar transistors (IGBTs) as the switching power devices. The switching frequency is fixed at 20kHz. By adjusting the on and off times of the switching power devices, the stable DC voltage is converted into a high-frequency switching DC power supply with adjustable amplitude and frequency. The output voltage adjustment range is 3.0V~4.2V, and the output current adjustment range is 0~2A. It provides programmable charging voltage and charging current outputs for the charged battery. The output voltage and current are calibrated in real time through a feedback regulation mechanism to ensure output stability.

[0016] The full-bridge rectifier circuit is a single-phase uncontrolled rectifier, the diode forward voltage drop is ignored, and the AC input voltage is a sine wave. (in This is the peak value of the AC voltage. , =50Hz), the rectified output DC voltage is the average value, and the calculation process is as follows: In practical applications, the AC input voltage is labeled as the effective value, i.e. Substituting into (in This is the effective value of the AC input voltage. (This represents the average DC voltage after rectification). This formula is a theoretical derivation of single-phase full-bridge uncontrolled rectification. After rectification, the voltage is filtered by an electrolytic capacitor. The derivation of the filter capacitor capacity is as follows: The core function of the filter capacitor is to filter out the ripple after rectification. The allowable ripple voltage value is set as follows: (Lithium battery charging has high requirements for ripple; this allowable value is set based on the charging characteristics of lithium batteries to avoid excessive ripple affecting battery life); the formula for calculating the filter capacitor capacity is as follows: ,in Given the rectified DC current, and considering the system's maximum output current of 2A and a rectification efficiency of approximately 90%, we can obtain... ; =50Hz (AC input frequency) (Substituting the AC 220V input scenario), the calculation yields... Considering the actual capacitance deviation (typically ±10%) and the need to reserve a certain margin to cope with current fluctuations, the filter capacitor capacitance was set to 2200μF. This setting is based on the capacitance deviation margin plus current fluctuation allowance. In actual testing, the ripple voltage at this capacitance was 3.2V, which meets the requirements. To meet the requirements, a stable DC voltage is obtained after ripple filtering. A full-bridge inverter topology is adopted, and the switching power device is an insulated gate bipolar transistor (IGBT). The selection criteria are that IGBT combines the high input impedance of MOSFET and the low on-state voltage drop of GTR, which is suitable for the system's requirements of 20kHz switching frequency and 2A output current. Compared with MOSFET, IGBT has lower conduction loss in high current scenarios, and compared with GTR, it has a faster switching speed, which meets the performance requirements of high-frequency inverter. The switching frequency was set to 20kHz. Specifically, a higher switching frequency would increase the losses of the switching devices (switching losses are positively correlated with frequency), while a lower frequency would increase output ripple and make filtering more difficult. Considering the lithium battery charging ripple requirements (≤50mV) and the IGBT switching characteristics (the IGBT used in this system is IRF460, with a maximum switching frequency of 50kHz and a switching loss of approximately 1.2W at 20kHz, which is within a reasonable range), a loss-ripple balance calculation was performed: when the frequency is below 15kHz, the output ripple exceeds 80mV, failing to meet the requirements; when the frequency is above 25kHz, the switching loss exceeds 2W, increasing the system's heat dissipation pressure. Therefore, the switching frequency was determined to be 20kHz, at which the switching loss is 1.2W and the output ripple is 42mV, balancing loss and ripple performance. The output voltage adjustment range is 3.0V~4.2V, and the setting is based on the fact that this system is compatible with lithium batteries with a nominal voltage of 3.7V (the nominal voltage of mainstream consumer and industrial lithium batteries). The lithium battery charging termination voltage is 4.2V (industry standard, determined by the characteristics of lithium battery materials; too high a voltage will cause the battery to bulge due to overcharging, while too low a voltage will result in insufficient charging), and the discharge termination voltage is 3.0V (to avoid damage to the battery due to over-discharge). Therefore, the output voltage adjustment range matches the charging and discharging characteristics of lithium batteries and does not require additional manual setting, as it is determined based on lithium battery industry standards. The output current adjustment range is 0~2A. Specifically, this is based on the compatible lithium battery capacity (mainstream 1000mAh~5000mAh). Excessive charging current can easily lead to battery overheating and lifespan loss, while insufficient current results in low charging efficiency. According to the recommended lithium battery charging rate (0.2C~1C, where C is the battery capacity), taking a 5000mAh battery as an example, the 1C charging current is 5A. However, considering the system's heat dissipation capacity (the heat sink size of this system is 50mm×50mm×10mm, with a maximum heat dissipation power of 5W), when the current exceeds 2A, the system temperature rise exceeds 30℃ (ambient temperature 25℃), affecting device stability. When the current is below 0.2A, the charging efficiency is below 85%. Therefore, the output current adjustment range is determined to be 0~2A, balancing charging efficiency, heat dissipation, and battery life.

[0017] The data acquisition module is used to acquire the terminal voltage signal of the battery being charged, the current signal in the charging circuit, and the battery surface temperature signal in real time. It performs anti-aliasing filtering, isolation amplification, and analog-to-digital conversion on the acquired signals to generate instantaneous terminal voltage, instantaneous charging current, and instantaneous temperature values. It also includes: Based on the instantaneous values ​​of terminal voltage and charging current, a recursive least squares algorithm with a forgetting factor is used to perform online real-time estimation and updating of the second-order RC equivalent circuit model constructed to characterize the internal dynamic characteristics of the battery, generating an adaptive battery model parameter vector containing ohmic internal resistance, electrochemical polarization internal resistance, concentration polarization internal resistance and their corresponding time constants. In this embodiment of the invention, the data acquisition module incorporates a signal acquisition unit, an anti-aliasing filter unit, an isolation amplification unit, and an analog-to-digital conversion unit to acquire and process three types of signals from the battery being charged in real time. The acquisition unit acquires the battery's terminal voltage signal, the current signal in the charging circuit, and the battery surface temperature signal in real time. The acquisition frequency is set to 1kHz. The acquired raw signals first enter the anti-aliasing filter unit and undergo first-order low-pass filtering to generate the instantaneous digital terminal voltage value. Instantaneous value of charging current and instantaneous temperature value Meanwhile, based on the instantaneous value of the terminal voltage and instantaneous value of charging current Using a recursive least squares algorithm with a forgetting factor, the second-order RC equivalent circuit model is estimated and updated online in real time. The expression for the second-order RC equivalent circuit model is as follows: For example, the algorithm initializes the forgetting factor. =0.97, initial parameter vector =[100mΩ,300mΩ,1200mΩ,0.5s,50s], updated each time. ; ; ; Calculate and generate the internal resistance of the ohm. Electrochemical polarization internal resistance Concentration polarization internal resistance and corresponding time constant , Adaptive battery model parameter vector .

[0018] The sampling frequency is set to 1kHz. Specifically, according to the sampling theorem, the sampling frequency must be greater than twice the highest frequency of the signal (Nyquist's theorem). The highest frequency of the battery terminal voltage and current signals is determined by the battery polarization characteristics. In the second-order RC equivalent circuit, the concentration polarization time constant is at most 50s, corresponding to the highest frequency. However, during actual charging, the current adjustment frequency is 10ms / time (100Hz). In order to capture the signal changes after current adjustment, the sampling frequency needs to be 10 times higher than the adjustment frequency, i.e. ≥1000Hz. At the same time, considering the processing capability of the analog-to-digital conversion unit (the ADC chip ADS1115 selected in this system has a maximum sampling rate of 1kHz and a resolution of 12 bits), the sampling frequency is determined to be 1kHz. This satisfies the sampling theorem and matches the performance of the ADC chip, avoiding data redundancy due to excessively high sampling frequency and signal distortion due to excessively low sampling frequency.

[0019] The state estimation module is used to take the adaptive battery model parameter vector as dynamic input, combine the instantaneous terminal voltage value and the instantaneous charging current value, and use the extended Kalman filter algorithm (EKF) to jointly estimate the battery's state of charge, electrochemical polarization voltage and concentration polarization voltage, and generate a joint battery state estimation vector, which includes the estimated value of the state of charge and its estimation error covariance. In this embodiment of the invention, the state estimation module incorporates an extended Kalman filter operation unit to adaptively adjust the battery model parameter vector. For dynamic input, combined with instantaneous terminal voltage value With instantaneous value of charging current The battery state is jointly estimated. The extended Kalman filter consists of two steps: prediction and update. In the prediction stage, the discrete state equation is... , , , =1ms is the sampling period. =0.97, predicted covariance matrix , Let be the process noise covariance matrix. During the update phase, the Kalman gain... and set =1×10 -3 To observe the noise covariance, then through , , Update the estimates to finally generate estimates including the state of charge. Electrochemical polarization voltage Concentration polarization voltage The joint estimation vector of battery state and its estimation error covariance is calculated. The sampling period needs to match the acquisition frequency of the data acquisition module (acquisition frequency 1kHz, sampling period 1ms), while also considering the real-time performance and computational complexity of the EKF algorithm. The EKF algorithm's single iteration calculation time is approximately 0.2ms (based on an ARM Cortex-M4 core, main frequency 168MHz). A 1ms sampling period ensures sufficient time for the algorithm to complete iterations and track battery state changes in a timely manner; therefore, Z=1ms is set. Coulomb efficiency is the ratio of the actual charge input to the battery during charging. Affected by battery internal resistance and temperature, the coulomb efficiency of a single lithium-ion battery at 25℃ and 0.5C~1C charging conditions is approximately 0.95~0.98. Through actual measurement (25℃, 1A charging, charge input 1940mAh, input 2000mAh), the calculated efficiency is... =1940 / 2000=0.97, therefore set =0.97. Furthermore, based on the terminal voltage measurement error, the standard deviation of the voltage measurement error at the analog-to-digital converter back-end is approximately 0.008V, therefore the observation noise variance is... =(0.008) 2 =6.4×10 -5 Based on comprehensive engineering experience, set =1×10 -3 (A certain margin is left to avoid the impact of observation noise on the stability of the estimation).

[0020] The optimization control module is used to construct a finite-time rolling optimization problem under the model predictive control algorithm MPC based on the joint estimation vector of battery state and the adaptive battery model parameter vector. The objective function of the optimization problem includes the tracking error of the target state of charge trajectory, the penalty term for the magnitude of the charging current, and the smoothing penalty term for the rate of change of current. The optimal charging current reference sequence in the next few control cycles is generated by solving the optimization problem by combining the instantaneous temperature value. In this embodiment of the invention, the optimization control module has a built-in model predictive control computation unit, which is based on the joint battery state estimation vector and the adaptive battery model parameter vector. A finite-time rolling optimization problem is constructed. The prediction time domain is set to 10 control periods (100ms), the control period is 10ms, and the objective function is... ( =1 to 10), among which For the target charged state trajectory, This is the current amplitude penalty coefficient. The current change rate weighting coefficient. Maximum charging current, This represents the maximum permissible rate of change of charging voltage under different SOC and temperatures. The constraint is 0% ≤ ≤100%, 0℃≤ ≤55℃, 3.0V≤ ≤4.2V, 0≤ ≤2A, combined with instantaneous temperature value Correction The optimization problem is transformed into a standard quadratic programming problem. , The optimal charging current reference sequence for the next 10 control cycles is generated by solving the problem using an embedded quadratic programming solver. Among them, the sampling period is matched with that of the state estimation module (1ms), but considering the response speed of the power conversion module (switching frequency 20kHz, response time about 50μs), the control period can be appropriately extended to balance control accuracy and computational load. The control period is set to 10ms (10 sampling periods). The prediction time domain needs to cover the dynamic response time of the battery state. The electrochemical polarization response time constant is about 0.5s (50 control periods), and the concentration polarization response time constant is about 50s (5000 control periods). Considering the computational load and control effect, the prediction time domain is selected as 10 control periods (100ms). This can track the dynamic changes of polarization voltage and avoid the real-time insufficiency caused by excessive computation. In addition, the weight coefficients of the objective function are derived as follows: (1) SOC tracking error weight ( SOC tracking accuracy is affected by battery internal resistance; the higher the internal resistance, the greater the impact of SOC tracking error on battery life. Therefore, the weight is positively correlated with internal resistance. Through experimental comparison (using different weight ratios to test SOC tracking error and battery cycle life), the specific experimental parameters and results are shown in Table 1 below. Weight 0.4 Weight 0.3 With a weight of 0.3, the SOC tracking error is the smallest (≤2%) and the battery cycle life is the longest (15% higher than with a fixed weight), so this weight allocation method is determined.

[0021] Table 1 Weighted Comparison Experiment Table

[0022] (2) Current amplitude penalty coefficient : The maximum permissible rate of change of charging voltage at different SOC and temperatures. This is the maximum charging current (2A, derived from the battery capacity 1C). Its function is to prevent excessive charging current from causing the battery to overheat and increase internal resistance. Through temperature coefficient testing, when… =25℃ When = 50%, =0.05V / s, derived as follows =2 / 0.05=40, as the temperature increases, Increase Decrease Increase to achieve adaptive penalty for current amplitude. (3) Weighting coefficient of current change rate Excessive current change rate can cause sudden changes in polarization voltage, affecting battery life. The weighting coefficient is positively correlated with the polarization time constant (the larger the time constant, the slower the polarization response, requiring limitation of the current change rate). =0.5s =50s, derived as follows The value is 0.495, but the actual value is 0.5, balancing the ease of calculation with the control effect.

[0023] The PWM drive module is used to receive the first element of the optimal charging current reference sequence as the current command for the current control cycle, compare the current command with the instantaneous value of the charging current measured from the charging circuit by vector pulse width modulation, and generate a pulse control signal to drive the switching power devices in the power conversion module to adjust autonomously. In this embodiment of the invention, the PWM drive module integrates an adjustment unit, a comparison unit, and a pulse processing unit, and receives the first element of the optimal charging current reference sequence. The current command is used as the current command for the current control cycle (10ms), and the instantaneous value of the charging current measured in the charging circuit is acquired simultaneously. Calculate the amplitude error Input the digital proportional-integral-derivative controller, and then... Generate analog voltage control signal , =5、 =0.1、 =0.05, voltage range 0~5V. Based on the 20kHz switching frequency of the power conversion module, a triangular carrier wave with an amplitude of 0~5V and a slope of ±200V / ms is generated. Compared with the triangular carrier input comparator, If the signal is greater than the carrier wave, a high level is output; otherwise, a low level is output, generating the original square wave signal. A 2μs dead time is inserted into the original square wave to generate a pulse control signal with a protected dead time, which is output to the power conversion module's switching power device drive circuit. A digital proportional-integral-derivative (PID) regulator is used to eliminate the amplitude error between the current command and the measured current. Regulator output The parameters are derived through the engineering tuning method (attenuation curve method): (1) proportional coefficient First =0、 =0, gradually increasing Until the current response exhibits a 4:1 damped oscillation, at which point... =5, response time is about 1ms, which meets the control cycle requirement. (2) Integral coefficient :exist Starting from 5, gradually increase until the static error is eliminated, at this point =0.1, static error ≤0.01A, to avoid current overshoot caused by integral saturation. (3) Differential coefficients :exist =5、 Based on =0.1, add To suppress current overshoot, when When the value is 0.05, the overshoot is ≤5%, and the response time is not significantly prolonged; therefore, this parameter is determined. The frequency of the triangular carrier wave must be consistent with the switching frequency of the power conversion module (20kHz) to ensure the synchronization of the pulse width modulation; the amplitude must match the output voltage range of the PID controller (0~5V) to facilitate comparison and generation of square wave signals; the slope is derived from the switching speed of the switching devices. The maximum switching speed of the IGBT is approximately 200V / μs, and the carrier slope is set to ±200V / ms to avoid voltage spikes caused by excessively fast switching speeds. The dead time is used to prevent the upper and lower bridge arm switching devices in the full-bridge inverter circuit of the power conversion module from being simultaneously turned on, which could lead to short-circuit damage. The turn-off delay time of the IGBT is approximately 1μs, and the turn-on delay time is approximately 0.5μs. To ensure safety, the dead time must be greater than the turn-off delay time; therefore, the dead time is set to 2μs. After testing, no bridge arm short circuits were observed, and the switching devices operated stably.

[0024] The safety protection module is used independently of the closed-loop control system composed of all the above modules. It monitors the instantaneous values ​​of terminal voltage, charging current, and temperature in real time and in parallel. Once any instantaneous value exceeds the preset hardware protection threshold, a fault lock signal is immediately generated, directly controlling the hardware protection circuit to cut off the charging circuit and lock the fault state.

[0025] In this embodiment of the invention, the safety protection module integrates an independent monitoring unit, a threshold comparison unit, and a fault handling unit. It operates independently of the closed-loop control system, monitoring three types of instantaneous values ​​in real time and in parallel at a monitoring frequency of 1kHz. The preset hardware protection threshold is: terminal voltage. <2.8V or >4.3V, charging current >2.5A, temperature <-5℃ or >60℃. The threshold comparison unit compares the monitored value with the threshold every 1ms. When any instantaneous value exceeds the threshold, a fault lock signal is immediately generated. The relay is controlled to disconnect the charging circuit through the hardware trigger mechanism, cut off the electrical connection, and lock the fault state. The fault state can only be cleared by manual reset until the monitored value returns to the threshold range before charging can be restarted to ensure system safety. Among them, the monitoring frequency must be higher than the sampling frequency of the closed-loop control system to ensure that the fault signal can be captured in time. The closed-loop sampling frequency is 1kHz, so the monitoring frequency is set to 1kHz to synchronize with the acquisition module and avoid the increase in power consumption caused by excessively high frequency. Based on the battery safety characteristics and the rated parameters of the power device, the circuit can be cut off in time when a fault occurs to protect the battery and system devices: (1) Terminal voltage threshold: The undervoltage protection threshold of a single lithium-ion battery is usually 2.75V~2.8V, and the overvoltage protection threshold is 4.25V~4.3V. Combined with the output voltage adjustment range (3.0V~4.2V) of this embodiment, the undervoltage threshold is set. <2.8V, overvoltage threshold >4.3V, which avoids over-discharge / overcharge of the battery and leaves a certain margin. (2) Charging current threshold: The maximum output current of the power conversion module is 2A, and the rated current of the switching device IGBT40N120 is 40A. Considering the line loss and fault current impact, the overcurrent threshold is set. >2.5A (1.25 times the maximum output current) to ensure that fault current does not damage the device. (3) Temperature threshold: The extreme operating temperature of lithium-ion batteries is -10℃ to 60℃. Charging below -5℃ is prone to lithium dendrite formation, and charging above 60℃ is prone to thermal runaway. Therefore, a low temperature threshold is set. <-5℃, high temperature threshold >60℃. The fault lockout signal adopts a hardware triggering mechanism (independent of software control) to ensure that the charging circuit can be cut off in time even if the closed-loop control system fails; the fault state latch needs to be manually reset to avoid automatic restart of charging when the fault is not eliminated, which could lead to secondary damage and meet the reliability requirements of safety protection.

[0026] Furthermore, the generation of the adaptive battery model parameter vector in the data acquisition module includes: A second-order RC equivalent circuit model characterizing the internal dynamic characteristics of the battery is constructed based on the instantaneous values ​​of terminal voltage, charging current, and temperature. Design a recursive least squares algorithm with a forgetting factor. Initialize the initial values ​​of the parameter estimates, the parameter estimation error covariance matrix, and the forgetting factor required for the algorithm. The forgetting factor is used to balance the weights of new and old data in parameter updates. Generate the initialized recursive identification algorithm framework. At each sampling moment, the instantaneous values ​​of charging current, terminal voltage, and open-circuit voltage provided by the state estimation module of the previous moment are combined into the regression data vector of the current moment, and the regression data vector is input into the recursive identification algorithm framework. The recursive identification algorithm framework calculates the parameter estimation gain at the current time based on the new input regression data vector, and uses the parameter estimation gain to update the parameter estimation vector and the parameter estimation error covariance matrix to generate the updated adaptive battery model parameter vector. The updated adaptive battery model parameter vector is validated for rationality. The validation rules include the non-negativity of internal resistance, the positive range of time constant, and the smoothness constraints of parameters. Abnormal parameter jumps caused by measurement noise or model mismatch are eliminated. The validated adaptive battery model parameter vector is output to the state estimation module and the optimization control module. At the same time, the adaptive battery model parameter vector is stored in non-volatile memory for use in the next power-on initialization. Specifically, the recursive identification algorithm framework is as follows: ; ; ; ; ; ; in, for Parameter estimation gain at time step for The parameter estimation error covariance matrix at time t. for The regression data vector at time step, It is the transpose symbol. Forgetting factor, for The parameter estimation vector at time step [time]. for Output observations at time 10:00 for The prior estimation error at time , To estimate initial values ​​for the parameters, These represent the initial values ​​of the ohmic internal resistance, electrochemical polarization internal resistance, concentration polarization internal resistance, time constant of the electrochemical polarization stage, and time constant of the concentration polarization stage, respectively. Here are the initial values ​​for the covariance matrix. The initial covariance coefficient is 10. 3 -10 6 , It is an identity matrix.

[0027] In this embodiment of the invention, a second-order RC equivalent circuit model representing the internal dynamic characteristics of the battery is constructed based on the instantaneous values ​​of the terminal voltage, charging current, and temperature. This model is built upon real-time data collected by the data acquisition module, accurately representing the ohmic voltage drop, electrochemical polarization, and concentration polarization characteristics during battery charging, providing a foundation for subsequent parameter identification and state estimation. The data acquisition module collects the instantaneous values ​​of the terminal voltage, charging current, and temperature of the battery being charged in real time, with the acquisition frequency set to 1kHz and the acquisition accuracy controlled within ±0.01V, ±0.01A, and ±0.1℃. The collected data is filtered and then transmitted to the modeling unit. The second-order RC equivalent circuit model consists of an ideal voltage source, an ohmic internal resistance, and two parallel RC links. The ideal voltage source represents the nonlinear relationship between the open-circuit voltage and the state of charge (SOC). The ohmic internal resistance represents the instantaneous voltage drop generated when current flows through it. The first parallel RC link consists of the electrochemical polarization internal resistance and the time constant of the electrochemical polarization link, representing the electrochemical polarization process. The time constant is set to 0.1s~1s. The second parallel RC link consists of the concentration polarization internal resistance and the time constant of the concentration polarization link, representing the concentration polarization process. The time constant is set to 10s~100s. Based on the collected terminal voltage, charging current, and instantaneous temperature values, the initial parameter ranges were calibrated through offline experiments. The initial ranges for ohmic internal resistance were set to 50mΩ~200mΩ, electrochemical polarization internal resistance to 100mΩ~500mΩ, concentration polarization internal resistance to 500mΩ~2000mΩ, the initial ranges for the time constant of the electrochemical polarization circuit to 0.1s~1s, and the initial ranges for the time constant of the concentration polarization circuit to 10s~100s. A complete second-order RC equivalent circuit model was constructed, and its mathematical expression strictly followed the set formula, specifically: ; in, For time variables, In time The instantaneous value of the terminal voltage below, In time The instantaneous value of the charging current under the given conditions. In time The battery's state of charge (SOC) value at that time. This represents the initial state of charge (SOC) of the battery. As an ideal voltage source, For ohmic internal resistance, , In time The polarization voltage corresponding to the electrochemical or concentration polarization process under the given conditions. For electrochemical polarization internal resistance, The equivalent capacitance of the electrochemical polarization process. The time constant of the electrochemical polarization process. For concentration polarization internal resistance, This is the equivalent capacitance of the concentration polarization stage. The time constant of the concentration polarization process. For Coulomb efficiency, The nominal capacity of the battery is fixed according to the specifications of the battery being charged, ensuring that the model can reflect the dynamic characteristics inside the battery.

[0028] A recursive least squares algorithm with a forgetting factor is designed. The initial values ​​of the parameters, the parameter estimation error covariance matrix, and the forgetting factor are initialized. A recursive identification algorithm framework is constructed after initialization to achieve online adaptive updating of model parameters. The forgetting factor is used to balance the weights of old and new data in parameter updates. Its value range is set from 0.95 to 0.99; a larger value indicates a higher weight for old data, and a smaller value indicates a higher weight for new data. Considering the parameter change characteristics during battery charging, the forgetting factor is fixed at 0.97 to ensure that parameter updates can track battery state changes while avoiding parameter fluctuations caused by measurement noise. The initial parameter estimates are based on the offline calibration initial parameter settings of the second-order RC equivalent circuit model, corresponding to the initial values ​​of ohmic internal resistance, electrochemical polarization internal resistance, concentration polarization internal resistance, time constant of the electrochemical polarization stage, and time constant of the concentration polarization stage, forming the initial parameter estimation vector. The parameter estimation error covariance matrix is ​​initialized according to the formula, and the initial covariance coefficients are... The value is 5×10 4 The identity matrix is ​​a 5th order identity matrix. The initial parameter estimation error covariance matrix is ​​obtained by multiplying the initial covariance coefficients by the identity matrix. The diagonal elements of the matrix represent the initial variance of the estimation errors of each parameter, and the off-diagonal elements represent the initial covariance of the estimation errors of each parameter. This ensures that the initialized matrix accurately reflects the error level of the initial parameter estimation, thus completing the initialization of the recursive identification algorithm framework. The complete calculation formula for this framework is as follows: ; ; ; ; ; ; in, for Parameter estimation gain at time step for The parameter estimation error covariance matrix at time t. for The regression data vector at time step, It is the transpose symbol. Forgetting factor, for The parameter estimation vector at time step [time]. for Output observations at time 10:00 for The prior estimation error at time , To estimate initial values ​​for the parameters, These represent the initial values ​​of the ohmic internal resistance, electrochemical polarization internal resistance, concentration polarization internal resistance, time constant of the electrochemical polarization stage, and time constant of the concentration polarization stage, respectively. Here are the initial values ​​for the covariance matrix. The initial covariance coefficient is 10. 3 -10 6 , It is an identity matrix.

[0029] At each sampling moment, the instantaneous values ​​of the charging current and terminal voltage from the previous moment, along with the estimated open-circuit voltage provided by the state estimation module, are combined to form the regression data vector for the current moment. This regression data vector is then input into the recursive identification algorithm framework to provide real-time data support for parameter updates. The sampling time is consistent with the acquisition frequency of the data acquisition module, i.e., every 1ms, to ensure the real-time nature and continuity of the data. The instantaneous values ​​of the charging current and terminal voltage from the previous moment are extracted from the cache of the data acquisition module, with a cache capacity of 1024 sampling points to ensure rapid retrieval of historical data. The estimated open-circuit voltage is estimated in real time by the state estimation module using an extended Kalman filter algorithm. The update frequency of the state estimation module is 100Hz, synchronized with the sampling time, and the accuracy of the estimated open-circuit voltage is controlled within ±0.02V. The construction of the regression data vector strictly follows the requirements of the recursive identification algorithm framework. The instantaneous values ​​of the charging current, terminal voltage, and open-circuit voltage, as well as related derived data from the previous moment, are arranged in a fixed order to form a 5-dimensional regression data vector. Each dimension corresponds to an input parameter. After the vector is constructed, it is input into the recursive identification algorithm framework in real time through the data transmission interface to ensure that the algorithm can obtain the latest data and update the parameters in a timely manner.

[0030] The recursive identification algorithm framework calculates the parameter estimation gain at the current time step based on the newly input regression data vector, and uses this gain to update the parameter estimation vector and the parameter estimation error covariance matrix, generating an updated adaptive battery model parameter vector, thus achieving online adaptive correction of the model parameters. The calculation of the parameter estimation gain strictly follows the formula defined in the recursive identification algorithm framework. During calculation, first construct the matrix expression within the parentheses. This can remove the forgetting factor. With the identity matrix Multiply, and then add the transpose of the regression data vector. The covariance matrix of the parameter estimation error at the previous time step Regression data vector After multiplying the matrices to obtain a square matrix, invert the square matrix and then use the parameter estimation error covariance matrix from the previous time step. The regression data vector at the current moment Perform matrix multiplication, and finally multiply the product with the inverse of the matrix within the parentheses to obtain the parameter estimation gain at the current time step. Prior estimation error Through formula The calculation yielded, where The instantaneous value of the terminal voltage collected at the current moment, i.e. Output observations at time 10:00 This is the transpose of the regression data vector. This is the parameter estimation vector from the previous time step. The update is done via formula Calculation, i.e., the parameter estimation vector of the previous time step. With parameter estimation gain Prior estimation error The sum of the products. Parameter estimation error covariance matrix. The update is done via formula Calculation, i.e., forgetting factor The reciprocal of the parameter estimation error covariance matrix at the previous time step The product of the two, minus the parameter estimation gain. Regression data vector transpose Covariance matrix of parameter estimation error with the previous time step The product of these factors. The above calculation process is repeated at each sampling time point to generate the updated adaptive battery model parameter vector. This ensures that the parameters can track changes in battery status in real time.

[0031] The updated adaptive battery model parameter vector undergoes a rationality check. The check rules include the non-negativity of internal resistance, the positive range of the time constant, and parameter smoothness constraints. Abnormal parameter jumps caused by measurement noise or model mismatch are eliminated. The validated adaptive battery model parameter vector is then output to the state estimation module and the optimization control module. Simultaneously, the adaptive battery model parameter vector is stored in non-volatile memory for use during the next power-on initialization. The non-negativity of internal resistance is checked by determining whether the values ​​of ohmic internal resistance, electrochemical polarization internal resistance, and concentration polarization internal resistance are greater than or equal to 0. , , If a negative value appears, it is judged as an abnormal parameter; the positive range of the time constant is checked by judging whether the time constant of the electrochemical polarization link and the time constant of the concentration polarization link are greater than 0 and within the preset initial range, that is... , Parameters exceeding the specified range are considered abnormal. Parameter smoothness is constrained by calculating the difference between the current and previous parameter estimation vectors. If the absolute value of this difference exceeds a preset threshold (10% of the initial value of each parameter), it's considered an abnormal parameter jump. For parameters deemed abnormal, the previous parameter estimation vector is used to replace them, ensuring parameter continuity and rationality. The validated adaptive battery model parameter vector is synchronously output to the state estimation and optimization control modules via a data interface, providing accurate parameter support for extended Kalman filter state estimation and model predictive control optimization. Simultaneously, it is written to non-volatile memory with a capacity of 1MB and a storage frequency consistent with the parameter update frequency. Upon the next power-on, the parameters in memory are directly read as initialization parameters, reducing initialization time and improving system startup efficiency.

[0032] Furthermore, such as Figure 2 As shown, Figure 2 This is a schematic diagram of the electrical principle of the second-order RC equivalent circuit model in this embodiment. The mathematical expression of the second-order RC equivalent circuit model is as follows: ; in, For time variables, In time The instantaneous value of the terminal voltage below, In time The instantaneous value of the charging current under the given conditions. In time The battery's state of charge (SOC) value at that time. This represents the initial state of charge (SOC) of the battery. As an ideal voltage source, For ohmic internal resistance, , In time The polarization voltage corresponding to the electrochemical or concentration polarization process under the given conditions. For electrochemical polarization internal resistance, The equivalent capacitance of the electrochemical polarization process. The time constant of the electrochemical polarization process. For concentration polarization internal resistance, This is the equivalent capacitance of the concentration polarization stage. The time constant of the concentration polarization process. For Coulomb efficiency, This refers to the battery's nominal capacity.

[0033] The model uses two independent parallel RC circuits, each corresponding to electrochemical polarization (resulting from the electrochemical polarization internal resistance). With equivalent capacitance (Composition) and concentration polarization (consisting of concentration polarization internal resistance) With equivalent capacitance Composition), combined with ohmic internal resistance The instantaneous ohmic voltage drop characterization can comprehensively capture the dynamic changes of different types of polarization during battery charging, accurately reflect the internal electrochemical reaction mechanism of the battery, and avoid model mismatch problems caused by incomplete polarization characteristic characterization. The electrochemical polarization time constant is determined through electrochemical kinetic derivation, diffusion characteristic time formula calculation, and offline experimental calibration. (0.1s~1s), Concentration polarization time constant The initial ranges (10s~100s) and various internal resistance parameters conform to the typical magnitudes of the double-layer capacitance and solid-phase diffusion coefficient of lithium-ion batteries, and are consistent with the impedance characteristics of actual healthy batteries. This allows for compatibility with single-cell lithium-ion batteries of different specifications, eliminating the need to redesign the model structure for specific batteries and reducing system adaptation costs. The model's mathematical formulas strictly correspond to the physical functions of each component. The terminal voltage formula, polarization voltage differential equation, and SOC calculation formula are interconnected and logically rigorous. It can directly combine the instantaneous values ​​of terminal voltage and charging current acquired by the data acquisition module, and achieve online adaptive parameter updates through a recursive least squares algorithm with a forgetting factor. This eliminates the need for complex calculation iterations, reducing the computational load and ensuring the real-time performance and accuracy of parameter identification. The model includes an ideal voltage source. The model characterizes the nonlinear relationship between open-circuit voltage and state of charge (SOC), indirectly relating it to the influence of temperature on battery electrochemical characteristics. Combined with instantaneous temperature values ​​acquired by the data acquisition module, the model can accurately characterize battery dynamics at different temperatures (e.g., low, normal, high) and within different SOC ranges, avoiding model errors caused by temperature changes and SOC fluctuations. This provides accurate model support for subsequent extended Kalman filter state estimation and model predictive control optimization. Model parameters ( It can be updated in real time through recursive identification algorithm, which can track parameter changes caused by battery aging and increased charge and discharge cycle count, eliminate abnormal parameter jumps, and always maintain consistency with the actual state of the battery. This solves the problem that traditional fixed parameter models cannot adapt to the battery aging process and the accuracy decreases after long-term use, thus extending the service life of the system and improving the safety and reliability of charging control.

[0034] Furthermore, the dynamic adjustment of the forgetting factor in the recursive least squares algorithm includes: During the algorithm's operation, the trace of the parameter estimation error covariance matrix is ​​calculated in real time, and the value of the trace is compared with the preset upper and lower bound thresholds to generate a quantitative index of parameter estimation uncertainty. Simultaneously, the condition number of the autocorrelation matrix of the regression data vector within multiple consecutive sampling periods is calculated, and the activation degree of the regression data vector is evaluated by the condition number as a quantitative indicator. The parameter estimation uncertainty quantification index and the incentive degree quantification index are input into a preset fuzzy inference engine. The fuzzy inference engine is used to output the adjustment direction and adjustment magnitude of the forgetting factor and generate the preliminary forgetting factor adjustment amount. A first-order low-pass filter is applied to the initial forgetting factor adjustment to smooth out drastic fluctuations in the forgetting factor caused by data mutations, generating a smoothed forgetting factor adjustment sequence. This smoothed forgetting factor adjustment sequence is then superimposed with the previous time-instance value of the forgetting factor, ensuring the superposition result is always constrained within a preset closed interval formed by the minimum and maximum values. This generates a dynamic forgetting factor value for updating parameters at the current time, which is then applied to the gain and covariance update calculations of the recursive least squares algorithm. In this embodiment, during algorithm execution, the trace of the parameter estimation error covariance matrix is ​​calculated in real time. The parameter estimation error covariance matrix is ​​the real-time updated value in the recursive identification algorithm. The calculation of its trace follows the rules of matrix trace operation, specifically: ,in to They are respectively The five elements on the main diagonal of the matrix correspond to the estimation error variances of the five model parameters. The trace value directly characterizes the overall uncertainty of the parameter estimation; a larger value indicates greater instability in the parameter estimation. The trace value is compared with preset upper and lower bound thresholds, with the preset upper bound threshold set at 5 × 10⁻⁶. 5 The lower bound threshold is set to 5×10. 3 The uncertainty quantification index for parameter estimation is generated. When the trace is greater than the upper bound threshold, the quantification index is high; when the trace is less than the lower bound threshold, the quantification index is low; and when the trace is between the two, the quantification index is medium. Simultaneously, the condition number of the autocorrelation matrix of the regression data vectors within multiple consecutive sampling periods is calculated. The sampling period is selected as 10 consecutive periods, i.e., regression data vectors within 10ms. The formula for calculating the autocorrelation matrix is: ( =10), where The number of sampling periods. The regression data vector for each sampling time point, Transpose it and calculate the 5th-order autocorrelation matrix using this formula. The condition number is calculated by the ratio of the largest singular value to the smallest singular value of the matrix, using the following formula: ,in Autocorrelation matrix The maximum singular value, The condition number is the minimum singular value. A smaller condition number indicates a stronger incentive for the regression data vector, while a larger number indicates a weaker incentive. The condition number is used to quantify the incentive level of the regression data vector: less than 10 indicates high incentive, between 10 and 100 indicates medium incentive, and greater than 100 indicates low incentive. The parameter estimation uncertainty quantification index and the incentive level quantification index are input into a pre-defined fuzzy inference engine. Based on pre-defined rules, the fuzzy inference engine outputs the adjustment direction and magnitude of the forgetting factor, generating an initial forgetting factor adjustment. The fuzzy inference engine takes two quantification indices as input and outputs the adjustment amount. It has a pre-defined fixed inference logic and requires no additional training. It directly matches the corresponding adjustment strategy based on the level of the input indices. For example, when both the parameter estimation uncertainty quantification index and the incentive level quantification index are high, the output adjustment magnitude is positive and large, i.e., increasing the forgetting factor to enhance the weight of new data; when both the parameter estimation uncertainty quantification index and the incentive level quantification index are low, the output adjustment magnitude is negative and small, i.e., slightly decreasing the forgetting factor to balance the weights of old and new data, ensuring that the initial adjustment amount matches the current parameter estimation state. A first-order low-pass filter is applied to the initial forgetting factor adjustment. The filter calculation formula is as follows: ,in This is the smoothed adjustment amount at the current moment. This is the filter coefficient, with a value of 0.1. This is a preliminary adjustment for the current moment. This is the adjustment amount after smoothing from the previous time step. This filtering operation smooths out the drastic fluctuations in the forgetting factor caused by data mutations, generating a smoothed forgetting factor adjustment sequence. The smoothed forgetting factor adjustment sequence is then superimposed with the previous time step value of the forgetting factor using the following superposition formula: Simultaneously, it ensures that the superposition result is always constrained within a closed interval formed by a preset minimum and maximum value. The preset minimum value is 0.95, and the maximum value is 0.99. If the superposition result is greater than 0.99, it is clamped to 0.99; if it is less than 0.95, it is clamped to 0.95. This generates a dynamic forgetting factor value for updating parameters at the current time. And it is directly applied to the gain of the recursive least squares algorithm. With covariance In the update calculation, the original fixed forgetting factor is replaced to achieve adaptive adjustment of the algorithm parameters. Among these, the filter coefficients... Used for first-order low-pass filtering operations Its core function is to smooth out the initial adjustment amount. Fluctuations, avoiding forgetting factor Dramatic changes occur, thus ensuring the stability of parameter estimation. The derivation is based on the following: (1) Derivation premise: The smoothing effect of the first-order low-pass filter is due to Decide, The larger the value, the faster the filtering response, but the worse the smoothing effect; A smaller value results in better smoothing, but a slower response time. In this embodiment, the forgetting factor adjustment period is consistent with the sampling period (1ms), and the initial adjustment amount... The regression data vector may experience instantaneous jumps due to sudden changes (such as fluctuations in charging current or temperature). Therefore, it is necessary to suppress these fluctuations to the greatest extent possible while ensuring real-time adjustment response (response time ≤ 10 sampling periods, i.e., 10ms). Adjustment is stable. (2) Specific derivation: Time constant of first-order low-pass filter ( =1ms is the sampling period, time constant The goal of determining the filter response speed is to ensure that the response time of the adjusted value after filtering is ≤10ms. Substituting into the formula for derivation: Simplifying, we get ,Right now In this example The value is set to 0.1. Additionally, the derivation of the upper and lower bound thresholds for the trace of the parameter estimation error covariance matrix (upper bound 5×10⁻⁶) is provided. 5 The lower bound is 5×10. 3 ): Thresholds are used to quantify the uncertainty of parameter estimation. The core basis is the parameter estimation error characteristics of the recursive least squares algorithm, which is derived by combining the actual variation range of battery model parameters: (1) Lower bound threshold (5×10 3 : Minimum error derivation based on fresh battery parameter estimation, model parameters of fresh batteries (cycle count ≤ 50 times) , , , , The fluctuations are small, and the variance of the parameter estimation error is minimized. Through actual measurements, when estimating parameters for a fresh battery, Matrix main diagonal elements ~ The minimum values ​​are respectively: ≈1×10 3 ( (Estimation error variance) ≈1.2×10 3 ( (Estimation error variance) ≈1.5×10 3 ( (Estimation error variance) ≈0.8×10 3 ( (Estimation error variance) ≈0.5×10 3 ( (Estimation error variance), minimum value of the trace Therefore, the lower threshold is set to 5×10. 3 At this point, the parameter estimation uncertainty is the lowest, and the quantitative index is "low". (2) Upper bound threshold (5×10 5 The parameter estimation error derivation is based on aged batteries (≥1000 cycles) and harsh operating conditions (temperature -5℃~55℃, drastic current fluctuations of 0~2A). Under these conditions, the model parameters fluctuate greatly, resulting in the largest variance in parameter estimation error. Actual measurements show... Matrix main diagonal elements ~ The maximum values ​​are respectively: ≈1×10 5 , ≈1.2×10 5 , ≈1.5×10 5 , ≈0.8×10 5 , ≈0.5×10 5 The maximum value of the trace If the trace exceeds this value, the parameter estimation error will exceed 15%, which will not meet the design requirements (parameter estimation error ≤ 5%). Therefore, the upper threshold is set to 5 × 10. 5 At this point, the parameter estimation uncertainty is the highest, and the quantitative indicator is "high".

[0035] Furthermore, the rule construction and output process of the preset fuzzy inferencer includes: Two input linguistic variables are defined for the fuzzy inference engine: a parameter estimation uncertainty quantification index and an incentive degree quantification index. For each input linguistic variable, fuzzy subsets corresponding to "low", "medium", and "high" are defined, and the corresponding membership functions adopt triangular or Gaussian functions, while determining the corresponding membership degrees. The output linguistic variable of the fuzzy inference engine is defined as the forgetting factor adjustment, and fuzzy subsets corresponding to "negative large", "negative small", "zero", "positive small", and "positive large" are defined for it. A fuzzy rule base based on expert experience is established, which contains several fuzzy rules in the form of "if-then". At each sampling time, the input quantization index is converted into the membership degree of the corresponding input linguistic variable through a fuzzification interface. At the same time, fuzzy inference is performed based on the fuzzy rule base using the Mamdani inference method to obtain the fuzzy set corresponding to the output linguistic variable. The fuzzy set consists of multiple fuzzy subsets with non-zero membership degrees and their corresponding membership functions. The obtained fuzzy set is defuzzified by the centroid method to calculate the abscissa of the centroid of the geometric shape enclosed by the membership function curve of the fuzzy set. Specifically, it is the integral of the membership function over its domain divided by the area under the membership function curve. The calculated abscissa of the centroid is mapped to the actual physical range of the forgetting factor adjustment amount, which is preset by the minimum and maximum allowable adjustment steps of the forgetting factor allowed by the system, to generate the original forgetting factor adjustment amount without smoothing processing. Check the absolute value of the difference between the original forgetting factor adjustment amount and the actually adopted forgetting factor adjustment amount at the previous moment. If this absolute value exceeds the preset maximum allowable single-step change amount, clamp the current adjustment amount to the boundary of this maximum allowable single-step change amount to generate the clamped adjustment amount. Perform first-order lag filtering on the clamped adjustment amount. The filtering time constant is dynamically adjusted according to the current charging stage of the battery. A smaller time constant is used in the constant current charging stage for fast response, and a larger time constant is used in the constant voltage charging stage to enhance stability, and output the smoothed forgetting factor adjustment amount that can be used to update the recursive algorithm.

[0036] In the embodiment of the present invention, by defining two input language variables of the fuzzy inference engine, namely the parameter estimation uncertainty quantification index and the excitation degree quantification index, each input language variable defines three fuzzy subsets of "low", "medium", and "high". The membership function uses a triangular function. The "low" fuzzy subset of the parameter estimation uncertainty quantification index is ≤5×10 3 , and the vertex of the membership function is at =2.5×10 3 , "medium" is 5×10 3 < ≤5×10 5 , and the vertex is at =2.525×10 5 , "high" is >5×10 5 , and the vertex is at =7.5×10 5 ; the "low" of the excitation degree quantification index corresponds to >100, and the vertex is at =150, "medium" corresponds to 10< ≤100, and the vertex is at =55, "high" corresponds to ≤10, and the vertex is at =5. The membership degree corresponding to each input quantization index is calculated using the triangular membership function, with the membership degree ranging from 0 to 1. The output linguistic variable of the fuzzy inference engine is defined as the forgetting factor adjustment. Five fuzzy subsets are defined: "negative large", "negative small", "zero", "positive small", and "positive large". The adjustment range is -0.02 to 0.02. "Negative large" corresponds to -0.02 to -0.015, "negative small" corresponds to -0.015 to -0.005, "zero" corresponds to -0.005 to 0.005, "positive small" corresponds to 0.005 to 0.015, and "positive large" corresponds to 0.015 to 0.02. A fuzzy rule base based on expert experience is established, containing nine fuzzy rules in the form of "if-then": If the parameter estimation uncertainty is low and the incentive level is low, the adjustment is negative and small; if the parameter estimation uncertainty is low and the incentive level is medium, the adjustment is zero; if the parameter estimation uncertainty is low and the incentive level is high, the adjustment is positive and small; if the parameter estimation uncertainty is medium and the incentive level is low, the adjustment is negative and small; if the parameter estimation uncertainty is medium and the incentive level is medium, the adjustment is zero; if the parameter estimation uncertainty is medium and the incentive level is high, the adjustment is positive and small; if the parameter estimation uncertainty is high and the incentive level is low, the adjustment is zero; if the parameter estimation uncertainty is high and the incentive level is medium, the adjustment is positive and small; if the parameter estimation uncertainty is high and the incentive level is high, the adjustment is positive and large. At each sampling time, the input quantization index is converted into the membership degree of the corresponding input linguistic variable through a fuzzification interface. Fuzzy inference is performed using the Mamdani inference method based on the fuzzy rule base. The trigger strength of each rule is obtained by taking the minimum of the input membership degree and the membership degree of the rule's antecedent. Then, the trigger strength is taken as the minimum of the fuzzy subset of the rule's consequent. The results of all rules are superimposed to obtain the fuzzy set corresponding to the output linguistic variable. This fuzzy set consists of multiple fuzzy subsets with non-zero membership degrees and their corresponding membership functions. The fuzzy set obtained through inference is defuzzified using the centroid method, calculated using the following formula: ,in To output the membership function of the fuzzy set, The forgetting factor adjustment is calculated as follows: the numerator is the sum of the product of the integral of the membership function over its domain and the adjustment, and the denominator is the area under the membership function curve. The calculated centroid abscissa is the original adjustment after defuzzification. The calculated centroid abscissa is then mapped to the actual physical range of the forgetting factor adjustment, which is preset to -0.02 to 0.02. The mapping formula is as follows: ,in The maximum possible value of the centroid's x-coordinate is used to generate the original, unsmoothed forgetting factor adjustment. The absolute value of the difference between the original forgetting factor adjustment and the actual forgetting factor adjustment used in the previous time step is checked. The maximum allowable single-step change is preset to 0.005. If the absolute value of the difference exceeds 0.005, the current adjustment is clamped to ±0.005, generating a clamped adjustment. A first-order hysteresis filter is then applied to the clamped adjustment. The filter calculation formula is as follows: The corresponding time constant is also For example, to match the characteristics of different charging stages, a segmented coefficient (1) is used for the constant current charging stage: for example, setting... =0.05, time constant = 0.001 / (1 0.05)≈0.00105s, small time constant, weak filtering, fast response, can quickly track changes in internal resistance and polarization parameters, and ensure the real-time performance of parameter estimation; (2) Constant voltage charging stage: for example, setting =0.8, time constant = 0.001 / (1 0.8) = 0.005s, with a large time constant and strong filtering, which can significantly suppress measurement noise and current jitter, making parameter estimation stable and without divergence, and synchronously transmitting it to the forgetting factor update stage.

[0037] Furthermore, such as Figure 3 As shown, Figure 3 for Figure 1 A functional flowchart of the optimization control module is shown below. In this embodiment, the optimization control module includes the following functions: S401: Based on the state of charge (SOC) estimate, electrochemical polarization voltage (EPV) estimate, and concentration polarization voltage (CPV) estimate in the joint battery state estimation vector, and combined with the adaptive battery model parameter vector, the battery state trajectory within multiple future control cycles is predicted using the discrete state-space equation of the second-order RC equivalent circuit model, generating a battery state prediction sequence containing SOC, polarization voltage, and terminal voltage; the target SOC reference trajectory of the battery in the current charging stage is obtained; the SOC prediction subsequence in the battery state prediction sequence is compared with the target SOC reference trajectory; the difference between the two at each prediction time is calculated; and the difference is weighted based on the ohmic internal resistance and polarization internal resistance data in the adaptive battery model parameter vector to generate a weighted penalty term for SOC tracking error. S402: Calculate the maximum allowable lithium-ion insertion rate of the battery under different states of charge and temperatures based on the instantaneous temperature value, map the insertion rate to a dynamic current amplitude penalty coefficient related to temperature and state of charge, and generate a charging current amplitude penalty term by combining the instantaneous charging current value at the current moment. S403: Based on the polarization time constant in the parameter vector of the adaptive battery model, calculate the dynamic response characteristic frequency of the charge transfer process and diffusion process, use the reciprocal of the dynamic response characteristic frequency as the time-varying weight coefficient of the current change rate penalty term, and combine it with the charging current difference between adjacent prediction times to generate a current change rate smoothing penalty term. S404: The weighted penalty term for state of charge tracking error, the penalty term for charging current amplitude, and the penalty term for smoothing current change rate are superimposed to construct an objective function with the charging current sequence in the future control time domain as the decision variable. At the same time, the state of charge constraint, temperature constraint, terminal voltage constraint, and current constraint derived from the maximum allowable lithium-ion embedding rate are incorporated into the optimization problem in the form of inequalities. The constrained quadratic programming problem is solved by calling the embedded quadratic programming solver to generate the optimal charging current reference sequence in the future several control cycles.

[0038] In this embodiment of the invention, the state of charge estimate is based on the joint battery state estimation vector. Estimated value of electrochemical polarization voltage Concentration polarization voltage estimate Combined with adaptive battery model parameter vector (Include , , , , Using the discrete state-space equations of a second-order RC equivalent circuit model, the battery state trajectory over multiple control cycles is predicted. The control cycle is set to 10ms, and 10 prediction cycles are selected, representing the battery state over the next 100ms. The discrete state-space equations are obtained by discretizing the continuous equations of the second-order RC equivalent circuit model using the Euler method. The specific discrete state equations are as follows: , , , ,in =1,2,...,10 To control the period of 10ms, For the predicted charging current, , , , The first The predicted values ​​of state of charge (SOC), electrochemical polarization voltage (EPV), concentration polarization voltage (CPV), and terminal voltage at each moment are used to generate a battery state prediction sequence containing SOC, polarization voltage, and terminal voltage. The target SOC reference trajectory for the current charging stage is obtained. During constant current charging, the target reference trajectory is linearly increasing with a slope of 0.01% / ms; during constant voltage charging, the target reference trajectory is a constant value (set to 95% based on battery specifications). The SOC prediction subsequence from the battery state prediction sequence is then used. With reference trajectory of target state of charge Compare the two and calculate the difference between them at each prediction time. Based on the ohmic internal resistance in the parameter vector of the adaptive battery model With polarization internal resistance , The data is weighted by the differences, and the weighting coefficients are as follows: Weight 0.4 Weight 0.3 A weight of 0.3 is used to generate a weighted penalty term for the state of charge tracking error. ( =1 to 10). Based on instantaneous temperature values. The maximum permissible lithium-ion intercalation rate of the battery under different states of charge and temperatures was calculated. Instantaneous temperature values ​​were acquired by the data acquisition module at a frequency of 1 kHz, filtered, and then input into the optimization control module to calculate the maximum permissible lithium-ion intercalation rate. The fitting function is calculated using the temperature and state of charge fitting function, which is: ,in =0.001、 =0.002、 =0.05, in A / s, mapping the embedding rate to a dynamic current amplitude penalty coefficient related to temperature and state of charge. ,in The maximum allowable charging current of the battery (set to 2A according to battery specifications), combined with the instantaneous value of the charging current at the current moment. Generate a charging current amplitude penalty term. ( =1 to 10), when When the value is 0, the penalty term is 0; otherwise, a penalty is triggered. This is based on the polarization time constant in the parameter vector of the adaptive battery model. , Calculate the dynamic response characteristic frequencies of the charge transfer and diffusion processes. The reciprocal of the characteristic frequency of the dynamic response is used as the time-varying weighting coefficient of the current change rate penalty term. Combined with the charging current difference between adjacent prediction times Generate a current change rate smoothing penalty term ( =1 to 10). The charging current sequence in the future control time domain is constructed by superimposing the weighted penalty term for the state of charge tracking error, the penalty term for the charging current amplitude, and the smoothing penalty term for the rate of change of current. The objective function for decision variables Simultaneously, the state of charge constraint, temperature constraint, terminal voltage constraint, and current constraint derived from the maximum allowable lithium-ion insertion rate are incorporated into the optimization problem in the form of inequalities, including the state of charge constraint. ≤SOC_max, temperature constraint is 0℃≤ ≤55℃, terminal voltage constraint is 3.0V≤ ≤4.2V, current constraint is 0≤ ≤ In this example, the state of charge (SOC) can be set to a maximum of 100%. The core reason is that the physical definition of SOC is the ratio of the battery's current available capacity to its nominal capacity, and its physical range is naturally 0% to 100%. From the perspective of actual charging logic, 100% SOC corresponds to the battery being fully charged. At this point, the lithium-ion insertion inside the battery reaches saturation. If charging continues, it will lead to excessive lithium-ion insertion, causing increased battery polarization, a sudden rise in temperature, and even damage to the battery electrode structure. This aligns with the protection logic of the safety protection module (such as temperature and terminal voltage thresholds). By calling the embedded quadratic programming solver to solve this constrained quadratic programming problem, the optimal charging current reference sequence for the next 10 control cycles is generated.

[0039] Among them, the maximum allowable lithium-ion intercalation rate The coefficients corresponding to the fitting function of temperature and state of charge =0.001、 =0.002、 The specific derivation process of 0.05 is as follows: 1. Derivation premise: lithium-ion intercalation rate The core influencing factor that directly determines the maximum allowable charging current of a battery is temperature. ) and state of charge ( The higher the temperature, the faster the lithium-ion migration rate and the greater the insertion rate. The lower the temperature, the more vacancies there are for lithium ions to intercalate inside the battery, and the greater the intercalation rate; conversely, the higher the temperature, the smaller the intercalation rate. The fitting function must satisfy the following conditions: temperature, The inherent law of "positive correlation with embedding rate" is also consistent with the actual test characteristics of the battery (single-cell lithium-ion battery, 2000mAh, rated voltage 3.7V) in this embodiment, ensuring that the derived coefficients, when substituted, [are accurate / effective]. The calculated value and the measured value have an error of ≤5%. 2. Experimental data acquisition: Select different temperatures ( )and The lithium-ion intercalation rate was measured using a combination of methods under the following conditions: charging ambient temperatures of 0℃, 15℃, 25℃, 40℃, and 55℃ (covering the temperature constraints set by the safety protection module from 0℃ to 55℃). The SOC values ​​were 20%, 40%, 60%, 80%, and 95% (covering the typical SOC ranges of the constant current charging and constant voltage charging stages); each combination was tested three times, and the average embedding rate was taken as the measured value. The specific measured data are shown in Table 2 below: Table 2

[0040] 3. Establishment of the fitting function and solution of coefficients: Multiple linear regression was used, with temperature T and SOC as independent variables, and the measured embedding rate was determined. Construct a linear fitting model with the dependent variable as the dependent variable. ,in , , Let be the coefficients to be determined. The core of multiple linear regression is to minimize the sum of squared residuals between the fitted values ​​and the measured values. (n=25 sets of measured data), through analysis of... , , Taking the partial derivatives of each equation and setting them to zero, we obtain the system of equations: Substituting the above 25 sets of measured data into the system of equations, the preliminary coefficients are calculated as follows: ≈0.00098 ≈0.00197 ≈0.0502. 4. Coefficient calibration and determination: In accordance with the engineering application requirements of this embodiment, the preliminary coefficients are rounded (retaining three decimal places), and the accuracy of the fitted function after calibration is verified: (1) Calibration coefficients: ≈0.00098≈0.001, ≈0.00197≈0.002, ≈0.0502≈0.05, yielding the calibrated coefficient. =0.001、 =0.002、 =0.05, the fitting function is (2) Accuracy verification: all measured data corresponding to , Substitute into the fitting function and calculate the fitted value. And calculate the relative error Calculations showed that the relative error of all combinations was ≤4.8%, less than the preset error requirement of 5%, and the fitting accuracy met the design requirements. For example: =25℃ When = 60%, the fitted value =0.195, measured value 0.085, relative error 4.7%; =40℃ When = 80%, the fitted value =0.25, measured value 0.104, relative error 3.8%. (3) Constraint verification: ensure that the fitted function calculated Complies with battery safety characteristics, when =0℃、 =20% (minimum embedding rate scenario). =0.054A / s, corresponding to the maximum allowable charging current =2.16A, slightly larger than the set 2A, leaving a safety margin; when =55℃ =95% (maximum embedding rate scenario) =0.295A / s, corresponding to =2 / 0.295≈6.78, which can effectively punish excessive current and conforms to the design logic of current amplitude penalty coefficient.

[0041] Furthermore, the real-time solution process of the embedded quadratic programming solver includes: The constrained multi-objective optimization problem is transformed into a standard quadratic programming form, where the objective function is a quadratic form of the decision variables and the constraints are linear inequalities of the decision variables, generating the coefficient matrix and vector of the standard quadratic programming problem. Using the optimal solution sequence obtained from the previous control cycle, an initial feasible solution to the optimization problem of the current control cycle is constructed through a shift operation. This initial feasible solution and the corresponding Lagrange multiplier estimates are then loaded into the embedded quadratic programming solver to significantly reduce the number of iterations required to reach the optimum. The original dual interior point method suitable for embedded quadratic programming solvers is used for iterative solution. In each iteration, the search direction is calculated and the step size is determined. At the same time, a large-scale sparse linear system of equations is solved. During the iterative solution process, the dual gap and the original residual are monitored in real time. Once both are less than the preset convergence tolerance, or the number of iterations reaches the preset upper limit, the iteration is terminated immediately, and the current iteration point is output as the approximate optimal solution. The approximate optimal solution is post-processed, including fine-tuning the solution components that may be near the constraint boundary, and low-pass filtering the current command sequence to suppress numerical noise in the solution, generating the optimal charging current reference sequence for several future control cycles.

[0042] In this embodiment of the invention, the constrained multi-objective optimization problem is transformed into a standard quadratic programming form, the standard form being: The constraints are ,in The decision variable vector is 10-dimensional, corresponding to the charging current in the next 10 control cycles, that is, the decision variables correspond to the charging current. The coefficient matrix of the quadratic term of the objective function (order 10×10) is given by... The coefficients of the quadratic terms constitute, This is a vector of coefficients for the first-order terms (10-dimensional). The constraint coefficient matrix is ​​a 30×10 matrix, corresponding to the three types of constraints at 10 time points. Given a 30-dimensional vector of constraint constants, a coefficient matrix and vector are generated by rearranging the coefficients of the objective function and constraints. The optimal solution sequence obtained from the previous control cycle is then used. An initial feasible solution to the optimization problem of the current control cycle is constructed through a shift operation. The shift operation formula is as follows: That is, discard the first current value of the previous cycle and fill in the end with the last current value of the previous cycle to form this initial feasible solution. and the corresponding Lagrange multiplier estimates (Initially set as a zero vector) Load the embedded quadratic programming solver. The Lagrange multiplier estimates correspond one-to-one with the constraints. The initial feasible solution shortens the solution iteration path, significantly reducing the number of iterations required to reach the optimum. The primal-dual interior-point method, suitable for embedded quadratic programming solvers, is used for iterative solving. The initial iteration step size is set to 0.9. In each iteration, the search direction is calculated using the primal-dual equations. The formula for calculating the search direction is: ,in The obstacle parameter is initially set to 0.1, and after each iteration... = The decay rate is ×0.5, and the step size is determined through line search to ensure that the iteration points always satisfy the constraints. Simultaneously, a large-scale sparse linear system of equations (order 10) is solved using the Cholesky decomposition method to reduce computational complexity. During the iterative solution process, the duality gap is monitored in real time. Compared with the original residual The preset convergence tolerance is 1×10. -3The preset maximum number of iterations is 20. If both the maximum and minimum iteration counts are less than the convergence tolerance, or if the maximum number of iterations is reached, the iteration is terminated immediately, and the current iteration point is output as the approximate optimal solution. Post-processing is performed on the approximate optimal solution. Solution components that may be near the constraint boundaries are fine-tuned with an adjustment increment of 0.01A. If a solution component is close to the upper or lower constraint limit, it is adjusted to a position 0.01A away from the boundary to avoid exceeding the constraint range. The current command sequence is low-pass filtered to suppress numerical noise in the solution, generating the optimal charging current reference sequence for the next 10 control cycles. This sequence is synchronously transmitted to the PWM drive module to adjust the PWM signal of the power conversion module, enabling autonomous adjustment of the charging current.

[0043] Furthermore, the calculation of the search direction in the iterative solution using the original-dual interior point method includes: In each iteration, based on the values ​​of the decision variable, slack variable, and dual variable at the current iteration point, the Lagrange gradient of the optimization problem with respect to each variable is calculated, and the gradient vector is generated. Construct the correction equation required to solve the search direction. This equation is a series of linear equations. Its coefficient matrix is ​​determined by the Hessian matrix of the objective function, the Jacobian matrix of the constraints, the current dual variable, and the relaxation variable. Its right-hand side consists of the gradient vector and the complementary relaxation condition deviation. Sparsity analysis and symbolic decomposition are performed on the coefficient matrix of the modified equation. The solution of the large-scale sparse linear equation system is transformed into the back-substitution calculation of multiple small-scale triangular equation systems by utilizing its special structure of block diagonal and low-rank update, which greatly reduces the amount of computation. By performing forward and backward substitution operations, the multiple small-scale trigonometric equation systems are solved efficiently, and the increments of the search directions for decision variables, slack variables, and dual variables are obtained. The maximum allowable step size is calculated to ensure that the iteration point always stays near the center path. This step size is determined by the condition that the slack variable and the dual variable are strictly positive. The actual step size used to update the iteration point is generated, thereby completing the search direction calculation and update for one iteration.

[0044] In this embodiment of the invention, the decision variable value at the current iteration point is determined first in each iteration. slack variable values and dual variable values Among them, the values ​​of decision variables The slack variable values ​​are the 10-dimensional charging current sequence vector obtained in the current iteration. Used to transform inequality constraints, with the same dimension as the number of constraints (30 dimensions), and dual variable values. With slack variable values The same dimension is used to characterize the influence weights of the constraints. Based on the above variable values, the gradient of the Lagrangian function of the optimization problem with respect to each variable is calculated. The Lagrangian function is... ,in Let Hessian matrix be the objective function. Let be the coefficient vector of the first-order terms. The constraint coefficient matrix, Let the constraint constant vector be... , These are the transposes of the decision variable vector and the dual variable vector, respectively. (The last two sentences are incomplete and require further context.) slack variables Dual variables Find the partial derivative to generate the gradient vector. The formula for calculating the gradient vector is: ,in The Lagrange function with respect to the decision variables The gradient (10-dimensional). For slack variables The gradient (30-dimensional). For dual variables The gradient (30-dimensional). This is the transpose of the constraint coefficient matrix. A modified equation is constructed to solve for the search direction; this equation is a series of linear equations used to solve for the search direction of the decision variables. 2. Search direction for slack variables Dual variable search direction The corrected equation is constructed based on the gradient vector and the complementary relaxation condition bias, where the complementary relaxation condition bias is... ,in For obstacle parameters, It is a vector of all 1s, with the same dimension as the slack and dual variables. The matrix form of the modified equation is: The coefficient matrix is ​​a block matrix, with the first row and first column being the Hessian matrix. (10×10 order), first row, third column is (10×30 order), the second row and second column is a diagonal matrix. (30×30 order), diagonal elements are dual variables The components, with the second row and third column forming a diagonal matrix. (30×30 order), diagonal elements are slack variables The components, with the third row and first column being the constraint coefficient matrix. (30×10 order), the third row and second column is the identity matrix. (30×30 order), the remaining blank blocks are zero matrices of the corresponding dimensions; the right-hand side terms are... (10-dimensional) (30 dimensions) Composed of 30 dimensions, these equations together form a complete linear system of equations for the modified equation. Sparsity analysis and symbolic decomposition are performed on the coefficient matrix of the modified equation. Through matrix sparsity analysis, the distribution pattern of zero elements in the coefficient matrix is ​​determined, revealing a special structure of block diagonal updates with low rank. The Hessian matrix... diagonal matrix , and identity matrix All are block diagonal matrices, with non-zero elements distributed only on the diagonal blocks. The low-rank update part is... and This special structure transforms the solution of a large-scale sparse linear system of 40×40 order into a back-substitution calculation of multiple small-scale triangular equation systems, specifically two triangular equation systems of 10×10 and 30×30 order, significantly reducing computational load and adapting to the computing power of embedded solver units. Symbolic decomposition employs the Cholesky decomposition method, decomposing each diagonal matrix into the product of its lower triangular matrix and its transpose. The decomposition formula is as follows: ,in It is a block diagonal matrix. It is a lower triangular matrix. for The transpose of the matrix is ​​used to record the signs and positions of matrix elements during the decomposition process, providing support for subsequent back-substitution calculations. Through forward and backward substitution operations, the multiple small-scale triangular equation systems are solved efficiently. First, a forward substitution is performed on the decomposed lower triangular matrix; the calculation formula is as follows: ,in For intermediate variable vectors, Divide the right-hand side of the equation system into blocks, and calculate by substituting each line. The components of the matrix are then substituted backwards from the transpose of the lower triangular matrix, and the calculation formula is as follows: ,in The search direction increment is calculated by substituting each row in reverse order, ultimately yielding the search direction for the decision variable. 2. Search direction for slack variables and the search direction of dual variables The increments are calculated, with the three increment vectors corresponding to the dimensions of the corresponding variables. The longest allowable step size is calculated to ensure that the iteration point always remains near the central path. This step size is determined by the condition that the slack variable and the dual variable are strictly positive, avoiding the occurrence of zero or negative values ​​in variables during iteration, which would lead to solution failure. Step size calculation is divided into dual step size and relaxation step size. , =1 to 30, relaxation step size , = 1 to 30, take the smaller of the two values, and multiply by a safety factor of 0.9 to generate the actual step size used to update the iteration points. The actual step size range is: The update formula is obtained by updating the iteration points using the actual step size. This completes the calculation and update of the search direction for one iteration, and proceeds to the next iteration until the convergence condition is met.

[0045] Furthermore, the PWM drive module's comparison of the current command with the instantaneous current value measured from the charging circuit using vector pulse width modulation includes: The system receives the first element of the optimal charging current reference sequence as the current command for the current period, and simultaneously acquires the instantaneous value of the charging current output by the data acquisition module. It then calculates the real-time error vector between the current command and the instantaneous value of the charging current, wherein the real-time error vector contains two components: amplitude error and phase error. The current command and the instantaneous value of the charging current are input into a digital proportional-integral-derivative regulator. The regulator generates an analog voltage control signal for adjusting the charging current through a combination of proportional, integral and derivative operations. The switching frequency of the switching power device in the power conversion module is obtained, and a triangular carrier signal with corresponding frequency and amplitude is generated based on the switching frequency. The analog voltage control signal and the triangular carrier signal are input into a comparator, and the original pulse width modulation square wave signal is generated by comparing the amplitude of the two at each moment. Dead time insertion is performed on the original pulse width modulation square wave signal to prevent shoot-through short circuits from occurring in the switching power devices in the power conversion module during switching. A pulse control signal with a protection dead time is generated and output to the drive circuit of the switching power devices.

[0046] In this embodiment of the invention, the first element of the optimal charging current reference sequence is received as the current command for the current cycle. The current cycle is 10ms, current command. The value range is 0~2A, and the instantaneous value of the charging current output by the data acquisition module is also acquired. The data acquisition module collects the charging current at a frequency of 1kHz, and after filtering, transmits it to the PWM drive module to calculate the real-time error vector between the current command and the instantaneous value of the charging current. ,in The amplitude error is calculated using the following formula: The unit is A. The phase error is calculated using the following formula: (The formula is missing from the original text.) , The phase of the instantaneous value of the charging current (fixed at 0°) is therefore =0°, the real-time error vector fully reflects the deviation between the current command and the actual current. The current command... With instantaneous value of charging current The input is a digital proportional-integral-derivative (PID) controller. The controller has built-in proportional, integral, and derivative arithmetic units. Through a combination of proportional, integral, and derivative operations, it generates an analog voltage control signal *u* to regulate the charging current. The proportional operation formula is as follows: ,in The proportionality constant is 5, and the integral formula is: ,in Let be the integral coefficient, with a value of 0.1. The differential formula is: ,in The differential coefficient, with a value of 0.05, represents the analog voltage control signal. The calculation formula is The voltage range is 0~5V, used for subsequent comparison with the triangular carrier signal. The switching frequency of the switching power devices in the power conversion module is obtained, with the switching frequency fixed at 20kHz. Based on the switching frequency, a triangular carrier signal with corresponding frequency and amplitude is generated. The amplitude range of the triangular carrier signal is 0~5V, and it is compared with the analog voltage control signal. The amplitude range is consistent, and the frequency is the same as the switching frequency of the switching power device, which is 20kHz. The rising slope of the triangular carrier signal is (5V). 0V) / (1 / (2×20kHz))=200V / ms, the falling edge slope is 200V / ms, forming a symmetrical triangular wave. The analog voltage control signal u and the triangular carrier signal are input to a comparator. The comparator compares the amplitudes of the two signals in real time at each moment. When the analog voltage control signal... When the amplitude is greater than that of the triangular carrier signal, the comparator outputs a high level (5V); when the analog voltage control signal... When the amplitude of the triangular carrier signal is less than or equal to the amplitude of the signal, the comparator outputs a low level (0V). Through continuous comparison, the original pulse-width modulated square wave signal is generated. The frequency of the square wave signal is the same as that of the triangular carrier signal, which is 20kHz. The duty cycle is controlled by the analog voltage control signal. The signal dynamically adjusts according to changes in voltage. A dead-time insertion process is applied to the original pulse-width modulated square wave signal, with a dead time set to 2μs. This is achieved through a pulse delay unit. When the square wave signal switches from high to low, a 2μs delay is applied before outputting a low level; similarly, when it switches from low to high, a 2μs delay is applied before outputting a high level. This prevents shoot-through short circuits in the switching power devices during switching, avoiding device damage. After dead-time insertion, a pulse control signal with a protected dead time is generated. The frequency of the pulse control signal remains 20kHz, and the duty cycle is consistent with the original square wave signal, except for the addition of a 2μs dead time at level transitions. This pulse control signal is output to the drive circuit of the switching power devices, driving their on / off states, thereby regulating the output current of the power conversion module and achieving autonomous adjustment of the charging current. This aligns with the overall control logic of the charger's autonomous adjustment system based on an adaptive algorithm.

[0047] Furthermore, such as Figure 4 As shown, the present invention also provides a charger autonomous adjustment control method based on an adaptive algorithm. This method is implemented based on the charger autonomous adjustment system based on the adaptive algorithm described above. The charger autonomous adjustment control method based on the adaptive algorithm includes the following steps: S01: The instantaneous values ​​of the terminal voltage, charging current, and temperature of the battery being charged are collected in real time through the data acquisition module; it also includes an online parameter correction step: during the charging process, the adaptive battery model parameter vector containing the ohmic internal resistance, electrochemical polarization internal resistance, concentration polarization internal resistance and their corresponding time constants of the second-order RC equivalent circuit model is periodically updated using a recursive least squares algorithm with a forgetting factor, and it is synchronized to S02 and S03; In this embodiment of the invention, the data acquisition module has a built-in signal acquisition and filtering unit to acquire the instantaneous terminal voltage value of the battery being charged in real time. Instantaneous value of charging current and instantaneous temperature value The acquisition frequency was set to 1kHz, and the acquisition accuracy was controlled within ±0.01V, ±0.01A, and ±0.1℃. The acquired raw signal was processed by a first-order low-pass filter to ensure stable and noise-free data acquisition. Online parameter correction was performed simultaneously. Based on a preset second-order RC equivalent circuit model, a recursive least squares algorithm with a forgetting factor was used to periodically update the adaptive battery model parameter vector. The update period was consistent with the acquisition frequency, updating once every 1ms. The mathematical expression of the second-order RC equivalent circuit model is: , , The model parameters include ohmic internal resistance. Electrochemical polarization internal resistance Concentration polarization internal resistance and corresponding time constant , The parameter vector of the adaptive battery model is formed. Initialize the forgetting factor using the recursive least squares algorithm. =0.97, initial value vector of parameter estimation =[100mΩ,300mΩ,1200mΩ,0.5s,50s], Initial parameter estimation error covariance matrix ( (A 5th-order identity matrix), and on each update, construct the regression data vector. Through formula ; ; ; The updated parameter vector is calculated and then synchronously transmitted to the S02 state estimation module and the S03 optimization control module.

[0048] S02: The state estimation module is based on a second-order RC equivalent circuit model and uses an extended Kalman filter algorithm to estimate the state of charge and polarization voltage of the battery in real time. In this embodiment of the invention, the battery state of charge and polarization voltage are estimated in real time using the second-order RC equivalent circuit model and adaptive battery model parameter vector provided by S01. The extended Kalman filter algorithm consists of two steps: prediction and update. In the prediction stage, the predicted values ​​of the state of charge and polarization voltage are calculated based on the discrete state equation, which is: , , ,in =1ms is the sampling period. =0.97 is the Coulomb efficiency. This refers to the battery's nominal capacity. , , These are the predicted values. The formula for updating the predicted covariance matrix is: ,in Here is the state transition matrix. The process noise covariance matrix (5×5, with 1×10 elements on the diagonal) -4 During the update phase, the Kalman gain is calculated. ,in For the observation matrix, The observation noise covariance matrix (1×1 order, with a value of 1×10) -3 ), and then through , , Update the estimate. The terminal voltage is predicted, and the final output is a real-time state estimation vector. , , [To S03]

[0049] S03: The optimization control module dynamically generates the optimal charging current sequence based on the real-time status output by S02 using a model predictive control algorithm; In this embodiment of the invention, the model predictive control algorithm presets the prediction time domain to 10 control cycles (total duration 100ms), the control cycle to 10ms, and the objective function... ( =1 to 10), among which For the target charged state trajectory, This is the current amplitude penalty coefficient. The current change rate weighting coefficient. =2A is the maximum charging current. Constraints include 0% ≤ ≤100%, 0℃≤ ≤55℃, 3.0V≤ ≤4.2V, 0≤ If the value is less than or equal to 2A, the optimization problem is transformed into a standard quadratic programming problem. , The optimal charging current sequence for the next 10 control cycles is generated by solving the problem using an embedded quadratic programming solver. ], output to S04.

[0050] S04: The PWM drive module receives the first element in the optimal charging current reference sequence as the current command for the current control cycle and adjusts the pulse control signal of the power conversion module. In this embodiment of the invention, the first element of the optimal charging current reference sequence is received by the PWM drive module. The current command is used as the current command for the current control cycle (10ms), and the instantaneous value of the charging current output by the data acquisition module is also acquired. Calculate the amplitude error Input the digital proportional-integral-derivative controller, and then... Generate analog voltage control signal ( =5、 =0.1、 =0.05, voltage range 0~5V). Based on the power conversion module switching frequency of 20kHz, a triangular carrier signal with an amplitude of 0~5V and a slope of ±200V / ms is generated. Compared with the triangular carrier input comparator, When the voltage is greater than the carrier wave, a high level (5V) is output; otherwise, a low level (0V) is output, generating the original square wave signal. A 2μs dead time is inserted into the original square wave to avoid shoot-through short circuits in the switching power devices, generating a pulse control signal with a dead time. This signal is output to the switching power device drive circuit of the power conversion module to adjust the on and off durations of the switches, thereby regulating the charging current.

[0051] S05: The safety protection module independently monitors the instantaneous values ​​of terminal voltage, charging current, and temperature, and cuts off the charging circuit when the values ​​exceed the safety threshold. In this embodiment of the invention, the safety protection module monitors the instantaneous value of the terminal voltage in real time. Instantaneous value of charging current and instantaneous temperature value The monitoring frequency is 1kHz, and the preset safety threshold is the terminal voltage. <2.8V or >4.3V, charging current >2.5A, temperature <-5℃ or >60℃. The monitoring unit compares the monitored value with the safety threshold every 1ms. When any monitored value exceeds the threshold, the protection mechanism is immediately triggered. The charging circuit is disconnected by controlling the relay, cutting off the electrical connection between the battery and the charger. At the same time, a protection signal is output to the system, stopping the operation and control of all modules. Charging can only be restarted when the monitored value returns to the safety threshold range.

[0052] S06: Repeat S01 to S05 to form a closed-loop control.

[0053] In this embodiment of the invention, after completing all operations from S01 to S05, the system enters the next cycle, with the cycle frequency consistent with the data acquisition frequency, repeating once every 1ms. S01 re-acquires real-time data and updates model parameters; S02 re-estimates the battery state based on the new parameters and data; S03 generates a new optimal charging current sequence based on the updated state; S04 outputs a new pulse control signal to adjust the charging current; and S05 continuously and independently monitors the safety status, forming a complete closed-loop control. This closed-loop control ensures that during charging, the battery state, model parameters, and charging current are always in a dynamically adapted state, achieving both improved charging efficiency and ensuring a safe and reliable charging process through the safety protection module, perfectly meeting the core requirements of an adaptive algorithm-based charger autonomous adjustment system.

[0054] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A charger autonomous adjustment system based on an adaptive algorithm, characterized in that, Includes the following modules: The power conversion module is used to connect to an external power source and convert the input AC or DC power into a high-frequency switching DC power source with adjustable amplitude and frequency, providing programmable charging voltage and charging current output to the battery being charged. The data acquisition module is used to acquire the terminal voltage signal of the battery being charged, the current signal in the charging circuit, and the battery surface temperature signal in real time. It performs anti-aliasing filtering, isolation amplification, and analog-to-digital conversion on the acquired signals to generate instantaneous terminal voltage, instantaneous charging current, and instantaneous temperature values. It also includes: Based on the instantaneous values ​​of terminal voltage and charging current, a recursive least squares algorithm with a forgetting factor is used to perform online real-time estimation and updating of the second-order RC equivalent circuit model constructed to characterize the internal dynamic characteristics of the battery, generating an adaptive battery model parameter vector containing ohmic internal resistance, electrochemical polarization internal resistance, concentration polarization internal resistance and their corresponding time constants. The state estimation module is used to take the adaptive battery model parameter vector as dynamic input, combine the instantaneous terminal voltage value and the instantaneous charging current value, and use the extended Kalman filter algorithm (EKF) to jointly estimate the battery's state of charge, electrochemical polarization voltage and concentration polarization voltage, and generate a joint battery state estimation vector, which includes the estimated value of the state of charge and its estimation error covariance. The optimization control module is used to construct a finite-time rolling optimization problem under the model predictive control algorithm MPC based on the joint estimation vector of battery state and the adaptive battery model parameter vector. The objective function of the optimization problem includes the tracking error of the target state of charge trajectory, the penalty term for the magnitude of the charging current, and the smoothing penalty term for the rate of change of current. The optimal charging current reference sequence in the next few control cycles is generated by solving the optimization problem by combining the instantaneous temperature value. The PWM drive module is used to receive the first element of the optimal charging current reference sequence as the current command for the current control cycle, compare the current command with the instantaneous value of the charging current measured from the charging circuit by vector pulse width modulation, and generate a pulse control signal to drive the switching power devices in the power conversion module to adjust autonomously. The safety protection module is used independently of the closed-loop control system composed of all the above modules. It monitors the instantaneous values ​​of terminal voltage, charging current, and temperature in real time and in parallel. Once any instantaneous value exceeds the preset hardware protection threshold, a fault lock signal is immediately generated, directly controlling the hardware protection circuit to cut off the charging circuit and lock the fault state.

2. The charger autonomous adjustment system based on adaptive algorithm according to claim 1, characterized in that, The parameter vector for generating the adaptive battery model in the data acquisition module includes: A second-order RC equivalent circuit model characterizing the internal dynamic characteristics of the battery is constructed based on the instantaneous values ​​of terminal voltage, charging current, and temperature. Design a recursive least squares algorithm with a forgetting factor. Initialize the initial values ​​of the parameter estimates, the parameter estimation error covariance matrix, and the forgetting factor required for the algorithm. The forgetting factor is used to balance the weights of new and old data in parameter updates. Generate the initialized recursive identification algorithm framework. At each sampling moment, the instantaneous values ​​of charging current, terminal voltage, and open-circuit voltage provided by the state estimation module of the previous moment are combined into the regression data vector of the current moment, and the regression data vector is input into the recursive identification algorithm framework. The recursive identification algorithm framework calculates the parameter estimation gain at the current time based on the new input regression data vector, and uses the parameter estimation gain to update the parameter estimation vector and the parameter estimation error covariance matrix to generate the updated adaptive battery model parameter vector. The updated adaptive battery model parameter vector is validated for rationality. The validation rules include the non-negativity of internal resistance, the positive range of time constant, and the smoothness constraints of parameters. Abnormal parameter jumps caused by measurement noise or model mismatch are eliminated. The validated adaptive battery model parameter vector is output to the state estimation module and the optimization control module. At the same time, the adaptive battery model parameter vector is stored in non-volatile memory for use in the next power-on initialization. Specifically, the recursive identification algorithm framework is as follows: ； ； ； ； ； ； in, for Parameter estimation gain at time step for The parameter estimation error covariance matrix at time t. for The regression data vector at time step, It is the transpose symbol. Forgetting factor, for The parameter estimation vector at time step [time]. for Output observations at time 10:00 for The prior estimation error at time , To estimate initial values ​​for the parameters, These represent the initial values ​​of the ohmic internal resistance, electrochemical polarization internal resistance, concentration polarization internal resistance, time constant of the electrochemical polarization stage, and time constant of the concentration polarization stage, respectively. Here are the initial values ​​for the covariance matrix. The initial covariance coefficient is 10. 3 -10 6 , It is an identity matrix.

3. The charger autonomous adjustment system based on adaptive algorithm according to claim 2, characterized in that, The mathematical expression of the second-order RC equivalent circuit model is as follows: ； in, For time variables, In time The instantaneous value of the terminal voltage below, In time The instantaneous value of the charging current under the given conditions. In time The battery's state of charge (SOC) value at that time. This represents the initial state of charge (SOC) of the battery. As an ideal voltage source, For ohmic internal resistance, , In time The polarization voltage corresponding to the electrochemical or concentration polarization process under the given conditions. For electrochemical polarization internal resistance, The equivalent capacitance of the electrochemical polarization process. The time constant of the electrochemical polarization process. For concentration polarization internal resistance, This is the equivalent capacitance of the concentration polarization stage. The time constant of the concentration polarization process. For Coulomb efficiency, This refers to the battery's nominal capacity.

4. The charger autonomous adjustment system based on adaptive algorithm according to claim 2, characterized in that, The dynamic adjustment of the forgetting factor in the recursive least squares algorithm includes: During the algorithm's operation, the trace of the parameter estimation error covariance matrix is ​​calculated in real time, and the value of the trace is compared with the preset upper and lower bound thresholds to generate a quantitative index of parameter estimation uncertainty. Simultaneously, the condition number of the autocorrelation matrix of the regression data vector within multiple consecutive sampling periods is calculated, and the activation degree of the regression data vector is evaluated by the condition number as a quantitative indicator. The parameter estimation uncertainty quantification index and the incentive degree quantification index are input into a preset fuzzy inference engine. The fuzzy inference engine is used to output the adjustment direction and adjustment magnitude of the forgetting factor and generate the preliminary forgetting factor adjustment amount. A first-order low-pass filter is applied to the initial forgetting factor adjustment to smooth out the drastic fluctuations in the forgetting factor caused by data mutations, generating a smoothed forgetting factor adjustment sequence. The smoothed forgetting factor adjustment sequence is then superimposed with the previous value of the forgetting factor, ensuring that the superposition result is always constrained within the closed interval formed by the preset minimum and maximum values, generating a dynamic forgetting factor value for updating the parameters at the current time step. This value is then applied to the gain and covariance update calculation of the recursive least squares algorithm.

5. The charger autonomous adjustment system based on adaptive algorithm according to claim 4, characterized in that, The pre-defined fuzzy inferencer's rule construction and output process includes: Two input linguistic variables are defined for the fuzzy inference engine: a parameter estimation uncertainty quantification index and an incentive degree quantification index. For each input linguistic variable, fuzzy subsets corresponding to "low", "medium", and "high" are defined, and the corresponding membership functions adopt triangular or Gaussian functions, while determining the corresponding membership degrees. The output linguistic variable of the fuzzy inference engine is defined as the forgetting factor adjustment, and fuzzy subsets corresponding to "negative large", "negative small", "zero", "positive small", and "positive large" are defined for it. A fuzzy rule base based on expert experience is established, which contains several fuzzy rules in the form of "if-then". At each sampling time, the input quantization index is converted into the membership degree of the corresponding input linguistic variable through a fuzzification interface. At the same time, fuzzy inference is performed based on the fuzzy rule base using the Mamdani inference method to obtain the fuzzy set corresponding to the output linguistic variable. The fuzzy set consists of multiple fuzzy subsets with non-zero membership degrees and their corresponding membership functions. The fuzzy set obtained by reasoning is defuzzified using the centroid method to calculate the centroid abscissa of the geometric shape enclosed by the membership function curve of the fuzzy set. Specifically, the integral of the membership function over its domain is divided by the area under the membership function curve. The calculated centroid abscissa is then mapped to the actual physical range of the forgetting factor adjustment amount, where the minimum and maximum adjustment steps of the forgetting factor allowed by the system are preset to generate the original forgetting factor adjustment amount without smoothing. Check the absolute value of the difference between the original forgetting factor adjustment and the actual forgetting factor adjustment used at the previous moment. If the absolute value exceeds the preset maximum allowable single-step change, clamp the current adjustment to the boundary of the maximum allowable single-step change to generate the clamped adjustment. Perform first-order hysteresis filtering on the clamped adjustment. The filtering time constant is dynamically adjusted according to the current charging stage of the battery. A smaller time constant is used in the constant current charging stage for fast response, and a larger time constant is used in the constant voltage charging stage to enhance stability. Output a smoothed forgetting factor adjustment that can be used to update the recursive algorithm.

6. The charger autonomous adjustment system based on adaptive algorithm according to claim 1, characterized in that, The optimization control module includes the following functions: Based on the state of charge (SOC) estimate, electrochemical polarization voltage (EPV) estimate, and concentration polarization voltage (CPV) estimate in the joint battery state estimation vector, and combined with the adaptive battery model parameter vector, the battery state trajectory within multiple future control cycles is predicted using the discrete state-space equation of the second-order RC equivalent circuit model. This generates a battery state prediction sequence containing SOC, polarization voltage, and terminal voltage. The target SOC reference trajectory for the battery in the current charging stage is obtained. The SOC prediction subsequence in the battery state prediction sequence is compared with the target SOC reference trajectory. The difference between the two at each prediction time is calculated. Based on the ohmic internal resistance and polarization internal resistance data in the adaptive battery model parameter vector, the difference is weighted to generate a weighted penalty term for SOC tracking error. The maximum allowable lithium-ion insertion rate of the battery under different states of charge and temperatures is calculated based on the instantaneous temperature value. This insertion rate is then mapped to a dynamic current amplitude penalty coefficient related to temperature and state of charge. Combined with the instantaneous charging current value at the current moment, a charging current amplitude penalty term is generated. Based on the polarization time constant in the parameter vector of the adaptive battery model, the dynamic response characteristic frequency of the charge transfer process and diffusion process is calculated. The reciprocal of the dynamic response characteristic frequency is used as the time-varying weight coefficient of the current change rate penalty term. Combined with the charging current difference between adjacent prediction times, a current change rate smoothing penalty term is generated. The weighted penalty term for state of charge tracking error, the penalty term for charging current amplitude, and the penalty term for smoothing current change rate are superimposed to construct an objective function with the charging current sequence in the future control time domain as the decision variable. At the same time, the constraints of state of charge, temperature, terminal voltage, and current constraint derived from the maximum allowable lithium-ion embedding rate are incorporated into the optimization problem in the form of inequalities. The constrained quadratic programming problem is solved by calling the embedded quadratic programming solver to generate the optimal charging current reference sequence for several future control cycles.

7. The charger autonomous adjustment system based on adaptive algorithm according to claim 6, characterized in that, The real-time solution process of the embedded quadratic programming solver includes: The constrained multi-objective optimization problem is transformed into a standard quadratic programming form, where the objective function is a quadratic form of the decision variables and the constraints are linear inequalities of the decision variables, generating the coefficient matrix and vector of the standard quadratic programming problem. Using the optimal solution sequence obtained from the previous control cycle, an initial feasible solution to the optimization problem of the current control cycle is constructed through a shift operation. This initial feasible solution and the corresponding Lagrange multiplier estimates are then loaded into the embedded quadratic programming solver to significantly reduce the number of iterations required to reach the optimum. The original dual interior point method suitable for embedded quadratic programming solvers is used for iterative solution. In each iteration, the search direction is calculated and the step size is determined. At the same time, a large-scale sparse linear system of equations is solved. During the iterative solution process, the dual gap and the original residual are monitored in real time. Once both are less than the preset convergence tolerance, or the number of iterations reaches the preset upper limit, the iteration is terminated immediately, and the current iteration point is output as the approximate optimal solution. The approximate optimal solution is post-processed, including fine-tuning the solution components that may be near the constraint boundary, and low-pass filtering the current command sequence to suppress numerical noise in the solution, generating the optimal charging current reference sequence for several future control cycles.

8. The charger autonomous adjustment system based on adaptive algorithm according to claim 7, characterized in that, The calculation of the search direction in the iterative solution using the original-dual interior point method includes: In each iteration, based on the values ​​of the decision variable, slack variable, and dual variable at the current iteration point, the Lagrange gradient of the optimization problem with respect to each variable is calculated, and the gradient vector is generated. Construct the correction equation required to solve the search direction. This equation is a series of linear equations. Its coefficient matrix is ​​determined by the Hessian matrix of the objective function, the Jacobian matrix of the constraints, the current dual variable, and the relaxation variable. Its right-hand side consists of the gradient vector and the complementary relaxation condition deviation. Sparsity analysis and symbolic decomposition are performed on the coefficient matrix of the modified equation. The solution of the large-scale sparse linear equation system is transformed into the back-substitution calculation of multiple small-scale triangular equation systems by utilizing its special structure of block diagonal and low-rank update, which greatly reduces the amount of computation. By performing forward and backward substitution operations, the multiple small-scale trigonometric equation systems are solved efficiently, and the increments of the search directions for decision variables, slack variables, and dual variables are obtained. The maximum allowable step size is calculated to ensure that the iteration point always stays near the center path. This step size is determined by the condition that the slack variable and the dual variable are strictly positive. The actual step size used to update the iteration point is generated, thereby completing the search direction calculation and update for one iteration.

9. The charger autonomous adjustment system based on adaptive algorithm according to claim 1, characterized in that, The PWM drive module's comparison of the current command with the instantaneous current value measured from the charging circuit using vector pulse width modulation includes: The system receives the first element of the optimal charging current reference sequence as the current command for the current period, and simultaneously acquires the instantaneous value of the charging current output by the data acquisition module. It then calculates the real-time error vector between the current command and the instantaneous value of the charging current, wherein the real-time error vector contains two components: amplitude error and phase error. The current command and the instantaneous value of the charging current are input into a digital proportional-integral-derivative regulator. The regulator generates an analog voltage control signal for adjusting the charging current through a combination of proportional, integral and derivative operations. The switching frequency of the switching power device in the power conversion module is obtained, and a triangular carrier signal with corresponding frequency and amplitude is generated based on the switching frequency. The analog voltage control signal and the triangular carrier signal are input into a comparator, and the original pulse width modulation square wave signal is generated by comparing the amplitude of the two at each moment. Dead time insertion is performed on the original pulse width modulation square wave signal to prevent shoot-through short circuits from occurring in the switching power devices in the power conversion module during switching. A pulse control signal with a protection dead time is generated and output to the drive circuit of the switching power devices.

10. A charger autonomous adjustment control method based on an adaptive algorithm, characterized in that, The method is implemented based on the adaptive algorithm-based charger autonomous adjustment system according to any one of claims 1-9, and the adaptive algorithm-based charger autonomous adjustment control method includes the following steps: S01: The instantaneous values ​​of the terminal voltage, charging current, and temperature of the battery being charged are collected in real time through the data acquisition module; it also includes an online parameter correction step: during the charging process, the adaptive battery model parameter vector containing the ohmic internal resistance, electrochemical polarization internal resistance, concentration polarization internal resistance and their corresponding time constants of the second-order RC equivalent circuit model is periodically updated using a recursive least squares algorithm with a forgetting factor, and it is synchronized to S02 and S03; S02: The state estimation module is based on a second-order RC equivalent circuit model and uses an extended Kalman filter algorithm to estimate the state of charge and polarization voltage of the battery in real time. S03: The optimization control module dynamically generates the optimal charging current sequence based on the real-time status output by S02 using a model predictive control algorithm; S04: The PWM drive module receives the first element in the optimal charging current reference sequence as the current command for the current control cycle and adjusts the pulse control signal of the power conversion module. S05: The safety protection module independently monitors the instantaneous values ​​of terminal voltage, charging current, and temperature, and cuts off the charging circuit when the values ​​exceed the safety threshold. S06: Repeat S01 to S05 to form a closed-loop control.

Images