A method and system for coordinated control of a generator lithium battery

Abstract

The application discloses a generator lithium battery cooperative control method and system, and belongs to the technical field of electric control of hybrid electric vehicles. The method collects lithium battery current, SOC state and temperature data in real time, establishes a dynamic Imax threshold after standardization processing, constructs a multi-physical field coupling model, accurately calculates the high-power duration of the generator and outputs optimal intermittent control parameters, realizes intelligent cutting and recovery of the charging loop through MOSFET switches, collects operation data to feedback and optimize model parameters, and forms a closed-loop learning system. The application solves the technical problems of response lag and split electric heat management of the traditional method, realizes double-target optimization of the extension of the service life of the generator and the improvement of the charging efficiency of the lithium battery, reduces the daily average high-load time of the generator, and improves the cycle life of the lithium battery.

Description

Technical Field

[0001] This invention belongs to the technical field of electric control systems for hybrid vehicles, and particularly relates to a method and system for coordinated control of a generator and lithium battery. Background Technology

[0002] In commercial vehicles such as trucks, the generator is the core equipment providing power to the vehicle's electrical system and also charging the battery to maintain its charge. However, traditional lead-acid batteries suffer from low charging efficiency and short lifespan, while lithium batteries, due to their high energy density and long cycle life, are gradually becoming the ideal replacement for lead-acid batteries. However, the charging process of lithium batteries requires more precise management, especially under the high-load and frequent start-stop conditions of trucks. The generator operates at high power output for extended periods, which can easily lead to overload, overheating, and aging, severely impacting its lifespan. While the application of lithium batteries has gradually become widespread, research on control strategies for truck generators remains insufficient. Most systems lack intelligent management of the generator load, causing the generator to operate at high power for extended periods, accelerating its aging.

[0003] The premature aging of generators is a common problem in the commercial vehicle industry. The root cause lies in the inability of traditional control strategies to effectively coordinate the interaction characteristics between the generator and the lithium battery. Existing technologies mainly face the following technical bottlenecks:

[0004] In terms of generator control, the existing method uses a fixed temperature threshold trigger protection mechanism, but this method has two inherent defects: first, the temperature sensor response delay causes overload protection to lag (typically 30-60 seconds); second, it does not consider the nonlinear effect of current fluctuations on winding heat accumulation.

[0005] Regarding lithium battery coordination, existing charging control strategies, while incorporating SOC parameters, only address simple charging phase switching and fail to establish a dynamic correlation model between SOC and generator thermal stress. Particularly under frequent truck start-stop conditions, this approach causes the generator to experience unnecessary current surges.

[0006] In the existing technology, the scheme of adjusting the generator output power through PID algorithm has three key shortcomings: ① The use of a linearized model to deal with the highly nonlinear thermoelectric coupling process results in a temperature prediction error of ±15℃; ② The feedback effect of changes in the internal resistance of the lithium battery on the generator load is not considered; ③ The control parameters are fixed and cannot adapt to different operating conditions.

[0007] At the system integration level, existing commercial vehicle electronic control systems generally design the generator controller (GCU) and battery management system (BMS) as independent units, lacking a collaborative optimization mechanism. This architectural flaw leads to conflicts between the two systems in the following aspects: ① conflicting power allocation between generator cooling requirements and lithium battery fast charging requirements; ② conflicting parameter settings between current fluctuation suppression and charging efficiency optimization; ③ lack of balance between short-term performance indicators and long-term lifespan goals. Summary of the Invention

[0008] This invention provides a generator-lithium battery collaborative control method and system to solve the problem of how to dynamically optimize the generator intermittent operation strategy through a multi-physics collaborative control algorithm based on the coupled analysis of lithium battery SOC state, dynamic fluctuation characteristics of charging current and generator temperature rise model, so as to achieve the dual objective of extending generator life and improving lithium battery charging efficiency.

[0009] To address the aforementioned technical problems, this invention provides a generator-lithium battery coordinated control method, comprising:

[0010] Collect lithium battery current, SOC status and temperature data, and after filtering, normalization and timestamp alignment, obtain standardized current waveform data and a sensing dataset with time-series labels.

[0011] Based on the generator rated current and standardized current waveform data, the dynamic Imax threshold is calculated, and a buffered dynamic Imax threshold is generated by combining the fluctuation envelope.

[0012] Compare the cumulative high-power time of the standardized current waveform data with the dynamic Imax threshold, call the temperature rise model to output suggested Tmax and Trest parameters, trigger intermittent commands and generate the target Trest value;

[0013] The expression for the cumulative high-power time of comparing the standardized current waveform data with the dynamic Imax threshold is as follows:

[0014]

[0015] Among them, T a S represents the cumulative value during high power duration; H(·) is the Heaviside step function; S p (t) is the power state function; β d For dynamic determination of threshold; t s t is the start time of high-power state. e This is the end time of the high-power state;

[0016] The expression for triggering the intermittent command and generating the target Trest value is:

[0017]

[0018] in, The optimal interval time decision value; For expectation value operators;

[0019] argmax is the parameter corresponding to the maximum probability; f c (T a () is a function of cooling efficiency;

[0020] P(T a ) represents the probability density of high-power state triggering;

[0021] The system analyzes the trigger intermittent command to generate a MOSFET cutoff control signal, synchronizes the target Trest value to initialize the timer, and generates a power execution control package.

[0022] Receive power execution control package, execute charging circuit cut-off and intermittent timing, collect temperature and current recovery data and generate execution status feedback report;

[0023] Analyze the temperature drop data in the execution status feedback report, correct the temperature rise model parameters, update the Tmax and Trest calculation rules and send them back to the collaborative decision engine to complete the closed-loop learning.

[0024] Furthermore, the collected lithium battery current, SOC status, and temperature data, after filtering, normalization, and timestamp alignment processing, include:

[0025] The lithium battery charging current signal and battery SOC status data are acquired, and signal filtering and AD conversion are performed to obtain digital current signal and SOC value.

[0026] Fluctuation characteristics are extracted from digital current signals, and combined with generator temperature sensor data, signal normalization processing is performed to obtain standardized current waveform data.

[0027] The standardized current waveform data and SOC values ​​are timestamped to generate a sensing dataset with time-series labels.

[0028] Furthermore, the calculation of the dynamic Imax threshold, combined with the fluctuation envelope to generate a buffered dynamic Imax threshold, includes:

[0029] Based on the generator rated current parameters and standardized current waveform data, dynamic threshold calculation is performed to obtain the initial Imax threshold.

[0030] The pulse fluctuation characteristics are analyzed from the standardized current waveform data, and the fluctuation envelope is fitted to obtain the current fluctuation compensation coefficient.

[0031] The initial Imax threshold is adjusted based on the current fluctuation compensation coefficient to generate a dynamic Imax threshold with a buffer.

[0032] Furthermore, the proposed Tmax and Trest parameters output by the temperature rise model include:

[0033] By comparing the standardized current waveform data with the dynamic Imax threshold, high-power state marking is performed to obtain the cumulative value of high-power duration;

[0034] The generator temperature rise model is called to process the cumulative value of high power duration, and combined with the SOC value, the suggested Tmax and Trest parameters are calculated.

[0035] When the cumulative high-power duration exceeds the recommended Tmax parameter, an intermittent command is triggered and a target Trest value is generated.

[0036] Further, the generation of the MOSFET cutoff control signal includes:

[0037] The trigger intermittent command is parsed, and the MOSFET drive signal of the charging circuit is encoded to obtain the cut-off control signal.

[0038] Further, the synchronization target Trest value initializes the timer, including:

[0039] Set the intermittent timer parameters according to the target Trest value, perform countdown initialization, and generate the timer start command.

[0040] Furthermore, the generation of the power execution control package includes:

[0041] Synchronously cut off the control signal and the timer start command to generate a power execution control package.

[0042] Furthermore, the execution of the charging circuit disconnection and intermittent timing includes:

[0043] Upon receiving the power execution control package, the MOSFET switch is driven to cut off the charging circuit and enter intermittent mode;

[0044] Start an intermittent timer to monitor the intermittent duration. When the target Trest value is reached, generate a resume charging command.

[0045] Furthermore, the generation of the execution status feedback report includes:

[0046] Collect generator temperature change data and current recovery response data during intermittent mode, and generate an execution status feedback report.

[0047] A generator-lithium battery collaborative control system, used to implement the generator-lithium battery collaborative control method described in any one of the above claims, comprising:

[0048] The multi-source sensing module is used to collect lithium battery current, SOC status and temperature data, and output standardized current waveform data and sensing dataset with time-series labels.

[0049] The dynamic parameter quantization module is used to calculate the dynamic Imax threshold and generate an adjusted threshold with a buffer.

[0050] The collaborative decision-making engine is used to call the temperature rise model to output suggested parameters and trigger intermittent commands.

[0051] The control strategy generation module is used to generate MOSFET cutoff control signals and power execution control packages;

[0052] The power execution module is used to perform charging circuit disconnection and intermittent timing operations.

[0053] The dynamic learning feedback module is used to correct the temperature rise model parameters and update the calculation rules.

[0054] The key innovations of this invention include:

[0055] (1) The introduction of a velocity field vector to characterize the rate of change of current solves the problem that the traditional RMS detection method cannot reflect transient characteristics. This technology is the core foundation for achieving millisecond-level response speed.

[0056] (2) Solve for the optimal operating point using canonical equations. Real-time collaborative optimization of multiple energy domains is achieved in automotive systems, breaking through the design constraints of "electrothermal separation" in existing technologies.

[0057] (3) A Monte Carlo probabilistic decision layer is superimposed on the traditional deterministic control framework, and the impact of system uncertainty on the control strategy is quantified by the probability density function.

[0058] The following are its main beneficial effects:

[0059] (1) By modeling current fluctuations as fluid motion, the problem of delayed response to transient conditions in the traditional fixed threshold method is solved. Compared with the protection mechanism that relies solely on temperature thresholds in the prior art, the present invention adopts a dynamic judgment model with multi-parameter coupling, which can predict high-power risk states in advance, reduce the temperature fluctuation amplitude of generator winding hot spots, and effectively avoid aging of insulation materials caused by thermal stress concentration.

[0060] (2) An innovative unified energy representation model for electro-thermal-mechanical energy is established, breaking through the limitation of the separation of electrical system and thermal management system in traditional control strategies. Specifically, the impact of lithium battery SOC state on heat dissipation efficiency is considered simultaneously when the generator is running at high power, and the cooling strategy is optimized by combining temperature gradient distribution during the charging interval, thereby improving the overall energy efficiency of the system.

[0061] (3) The system uncertainty is handled by a probabilistic decision algorithm, so that the life decay curves of the generator and the lithium battery tend to be synchronized. Attached Figure Description

[0062] Figure 1 A schematic flowchart illustrating a generator-lithium battery collaborative control method provided in an embodiment of this application;

[0063] Figure 2 This is a structural block diagram of a generator-lithium battery collaborative control system provided in an embodiment of this application. Detailed Implementation

[0064] Example 1: Refer to Figure 1 This is a flowchart illustrating a generator-lithium battery coordinated control method provided in an embodiment of the present invention. The process may include at least steps S100-S600:

[0065] S100 collects lithium battery current, SOC status and temperature data. After filtering, normalization and timestamp alignment, standardized current waveform data and a sensing dataset with time-series labels are obtained.

[0066] S200: Based on the generator rated current and standardized current waveform data, calculate the dynamic Imax threshold, and generate a buffered dynamic Imax threshold by combining the fluctuation envelope.

[0067] S300 compares the cumulative high-power time of the standardized current waveform data with the dynamic Imax threshold, calls the temperature rise model to output suggested Tmax and Trest parameters, triggers intermittent commands and generates the target Trest value.

[0068] S400: Parse the trigger intermittent instruction to generate a MOSFET cut-off control signal, synchronize the target Trest value to initialize the timer, and generate a power execution control package.

[0069] S500 receives the power execution control package, executes the charging circuit cutoff and intermittent timing, collects temperature and current recovery data, and generates an execution status feedback report.

[0070] S600 analyzes the temperature drop data in the execution status feedback report, corrects the temperature rise model parameters, updates the Tmax and Trest calculation rules and sends them back to the collaborative decision engine to complete closed-loop learning.

[0071] Step S100 includes at least steps S110-S130:

[0072] S110: Acquire the lithium battery charging current signal and battery SOC status data, perform signal filtering and AD conversion processing to obtain digital current signal and SOC value.

[0073] Specifically, the lithium battery charging current signal is acquired through a high-precision closed-loop Hall effect sensor installed on the positive electrode wire of the charging circuit. This sensor employs a magnetic balance measurement principle, covering a DC current range of 0-200A, with a nonlinearity error of less than 0.1%FS. The sensor output is a 0-5V analog voltage signal, transmitted to the signal conditioning circuit via a twisted-pair shielded cable. The battery SOC status data is acquired via the vehicle's CAN bus, using the BMS message (PGN65472) in the SAE J1939 protocol to parse the SOC percentage value, with the data update cycle strictly controlled within 100±5ms.

[0074] The signal filtering process employs a three-stage cascaded filtering architecture: the first stage is a hardware RC low-pass filter with a cutoff frequency set to 1kHz to suppress high-frequency switching noise; the second stage uses a digital FIR filter with a Hanning window as the window function type, and the passband ripple is controlled within 0.01dB; the third stage is a moving average filter with a window width set to 10 sampling points. Understandably, this composite filtering scheme can effectively eliminate high-frequency interference above 100kHz caused by PWM charging, while preserving the dynamic characteristics of the current signal.

[0075] The AD conversion process employs a 24-bit Σ-Δ analog-to-digital converter (ADS124S08) with a built-in programmable gain amplifier (PGA). When the gain is set to 8x, the effective resolution reaches 20 bits. The converter operates in continuous conversion mode with a sampling rate configured at 2kSPS and a 2.5V precision voltage source as the reference voltage. The conversion result is transmitted to the main control unit via SPI interface using DMA, and CRC-16 checksum is used during data transmission to ensure data integrity.

[0076] Furthermore, the digital current signal processing includes a real-time calibration step: zero-point calibration is automatically performed when the system is powered on, using the average output value under 30 seconds of no-load condition as the bias compensation value; the full-scale calibration coefficient is written to non-volatile memory during factory calibration, achieving a calibration accuracy of ±0.05%. The SOC value processing includes a validity check: when the jump between two adjacent sampling values ​​exceeds 5%, data validity verification is initiated by cross-validating the remaining capacity (Ah) reported by the BMS with the voltage-SOC lookup table value.

[0077] The digital current signal and SOC value storage employ a dual-buffering mechanism: the front-end buffer stores the original sampled values ​​with a data depth of 1 second; the back-end buffer stores the processed valid data using a timestamp index structure. The data storage format is defined as follows: the current value is of type int32, with a unit of 0.1mA; the SOC value is of type uint16, with a unit of 0.01%; accompanied by a 64-bit timestamp (μs-level precision) and an 8-bit status flag.

[0078] S120. Extract fluctuation characteristics from the digital current signal, combine them with generator temperature sensor data, and perform signal normalization processing to obtain standardized current waveform data.

[0079] Specifically, the wave feature extraction employs a multi-scale analysis method: in the time domain, peak-to-peak value, root mean square value, and waveform factor are calculated within a 100ms window; in the frequency domain, the energy distribution in the 0-1kHz frequency band is analyzed using a 1024-point FFT, with a focus on harmonic components in the 100-400Hz range. The feature extraction algorithm runs on a dedicated DSP coprocessor, with computation latency controlled within 5ms.

[0080] The generator temperature sensor uses a three-wire PT100 platinum resistance thermometer for data acquisition. The measurement circuit employs a constant current source drive (1mA) and a 24-bit ADC (ADS1220) to form a four-wire measurement system, achieving a temperature resolution of 0.01℃. The sensor is installed at the hottest point of the generator stator winding, and thermal coupling is ensured through thermal grease. Temperature data is sampled once per second and then digitally filtered (first-order low-pass, time constant 10s) before subsequent processing.

[0081] Understandably, the signal normalization processing includes a dynamic reference adjustment algorithm: when the generator temperature is below 70°C, a fixed reference value (100% of the rated current) is used; when the temperature rises to the 70-90°C range, the reference value decreases linearly by 0.5% for every 1°C increase; when the temperature exceeds 90°C, an emergency derating mode is activated, and the reference value is forcibly set to 80% of the rated value. This algorithm is implemented through a lookup table, and the table data can be updated online.

[0082] Furthermore, the standardization process includes: first, temperature compensation is applied to the current data, with the compensation coefficient calculated according to the Arrhenius equation and the activation energy set to 0.65 eV; second, dimensional normalization is performed, converting the actual current value into a percentage relative to the current reference value; finally, sliding standardization is executed, using the Z-score method to convert the data into a distribution with a mean of 0 and a standard deviation of 1. Data quality is monitored in real time during the processing, and an anomaly alarm is triggered when three consecutive sampling points exceed the ±3σ range.

[0083] The standardized current waveform data output format includes: a 32-bit floating-point normalized value (range -5.0 to +5.0), a 16-bit integer raw value, an 8-bit quality flag (bits 0-3 indicate sensor status, bits 4-7 indicate processing status), and a 64-bit precise timestamp. Data is packaged into fixed-length frames (32 bytes / sample point) and transmitted via a Gigabit Ethernet interface using the UDP protocol at a transmission rate of 1 Mpps.

[0084] S130. The standardized current waveform data and SOC values ​​are timestamped to generate a sensing dataset with time-series labels.

[0085] Specifically, the timestamp alignment adopts a high-precision time synchronization protocol (IEEE 1588v2), and the master clock source is a GPS-disciplined atomic clock, with a clock synchronization error of less than 1μs at each node within the system. The alignment algorithm is based on the least squares method: first, a time-value sequence of SOC data is established; second, for each sampling point of the standardized current waveform data, the nearest SOC data point is found within a 10ms window before and after it; finally, the accurate alignment value is calculated through linear interpolation.

[0086] The timing tag generation incorporates a dual system of relative and absolute time stamps: the relative time is based on the system startup time and uses a 64-bit counter to record nanosecond-level offsets; the absolute time uses UTC time format and includes year, month, day, hour, minute, second, and millisecond fields. The tag also contains a 16-bit sequence number for detecting data packet loss.

[0087] Furthermore, the sensing dataset is organized into a circular buffer structure, with a buffer capacity configured for 30 minutes of data (approximately 5.4GB), employing memory-mapped file technology for high-speed access. The data index uses a B+ tree structure, supporting fast retrieval by time range (minimum granularity 1ms) or by numerical characteristics (e.g., current greater than 80%). The data storage format uses Apache Parquet columnar storage, supporting efficient compression (Snappy algorithm) and partitioned storage.

[0088] Strict data integrity protection measures are implemented during the dataset generation process: each frame of data is appended with a CRC-32 checksum; critical fields use ECC memory automatic error correction; and data consistency checks are performed periodically (a full check is performed every 15 minutes). The dataset output interface supports multiple protocols, including CAN FD (8Mbps), Ethernet / IP, and TSN (Time-Sensitive Networking), ensuring seamless integration with subsequent processing modules.

[0089] Understandably, the time-labeled sensing dataset also contains rich metadata, such as sensor calibration date, filter parameter version, and temperature compensation curve ID. This metadata is associated with the master data through a dedicated description file (XML format) and is automatically loaded and verified when the data is used. The dataset also supports data playback functionality, allowing the system state to be reproduced by replaying data from a specific time period.

[0090] Step S200 includes at least steps S210-S230:

[0091] S210. Based on the generator rated current parameters and standardized current waveform data, perform dynamic threshold calculation to obtain the initial Imax threshold.

[0092] Specifically, the generator rated current parameters are obtained from the generator nameplate data and stored in non-volatile memory, including two parameters: rated continuous operating current and peak current. The standardized current waveform data comes from the output of the S120 module and is a real-time current signal that has undergone temperature compensation and normalization processing. Understandably, the dynamic threshold calculation is implemented using a sliding window statistical method, with a window width set to 10 seconds and a window sliding step size of 1 second.

[0093] Furthermore, the initial Imax threshold calculation process includes the following steps: First, the 95th percentile value of the current signal is calculated within the sliding window as a basic reference value; second, the basic reference value is corrected according to the current generator temperature, and the temperature correction coefficient is obtained by looking up a table, with the table data sourced from the thermal derating curve provided by the generator manufacturer; finally, the smaller value between the corrected value and 80% of the rated current is taken as the initial Imax threshold. The temperature correction coefficient table contains 20 temperature intervals, each spanning 5°C, with a coefficient range of 0.7-1.0.

[0094] The dynamic threshold calculation process implements multiple protection mechanisms: when abnormal standardized current waveform data is detected (such as three consecutive sampling points exceeding the reasonable range), it automatically switches to the backup calculation mode and directly takes 75% of the rated current as the threshold using a fixed ratio method. The initial Imax threshold output is accompanied by a confidence index, which is comprehensively evaluated based on data integrity and calculation process stability, and is divided into three levels: high, medium, and low.

[0095] The initial Imax threshold is updated synchronously with the standardized current waveform data, with a minimum interval of 100ms. The threshold data is stored in double-precision floating-point format, with units consistent with the standardized current waveform data, facilitating subsequent comparison calculations. The threshold calculation module has a built-in self-checking function, automatically performing an integrity check every 60 seconds to ensure the correctness of the calculation logic.

[0096] S220. Analyze the pulse fluctuation characteristics from the standardized current waveform data, perform fluctuation envelope fitting, and obtain the current fluctuation compensation coefficient.

[0097] Specifically, the pulse fluctuation feature analysis is implemented using a multi-resolution analysis method: in the time domain, the peak-to-peak value, rise time, and fall time within a 1-second window are calculated; in the frequency domain, the energy distribution in the 50-500Hz frequency band is analyzed using a 512-point FFT. The fluctuation feature extraction algorithm runs on a dedicated digital signal processor, with a computation delay controlled within 2ms.

[0098] Understandably, the wave envelope fitting employs a modified Hilbert transform method: first, a window function (Kaiser window, β = 6) is applied to the standardized current waveform data; second, an analytical signal is obtained through a 90-degree phase-shift network; finally, the amplitude of the analytical signal is smoothed (Gaussian filter, σ = 0.5) to obtain the envelope trajectory. The envelope sampling rate is consistent with the original data to ensure time alignment.

[0099] Furthermore, the calculation process for the current fluctuation compensation coefficient includes: first, statistically analyzing the maximum and minimum values ​​of the envelope within one minute; second, calculating the ratio of the fluctuation amplitude (the difference between the maximum and minimum values) to the reference value; and finally, converting the ratio into a compensation coefficient using a nonlinear mapping function, the parameters of which are obtained through experimental data calibration. The compensation coefficient ranges from 0.9 to 1.1, with a step size of 0.01.

[0100] The fluctuation envelope fitting module implements a dynamic adjustment mechanism: when a significant change in the current waveform shape is detected (such as the proportion of harmonic components exceeding 30%), it automatically switches to the backup fitting algorithm and uses the moving extreme value method to directly track the peaks and troughs. The compensation coefficient output includes a validity flag; when the envelope fitting quality index is below a threshold, the flag is set to invalid.

[0101] S230. Adjust the initial Imax threshold according to the current fluctuation compensation coefficient to generate a dynamic Imax threshold with a buffer.

[0102] Specifically, the threshold adjustment algorithm employs a product correction model: the initial Imax threshold is multiplied by the current fluctuation compensation coefficient to obtain a preliminary adjustment value; then, a safety margin of ±5% is applied to form the final threshold. This safety margin is dynamically adjusted according to the system operating status: when the generator temperature exceeds 85°C, the margin increases to ±7%; when the lithium battery SOC is below 20%, the margin decreases to ±3%.

[0103] Understandably, the buffer settings employ dynamic windowing technology: the window width adaptively matches the current fluctuation period, with a minimum of 100ms and a maximum of 1 second. The buffer boundary values ​​are determined through historical data statistics, including an upper buffer and a lower buffer, set to 105% and 95% of the dynamic Imax threshold, respectively. The buffer parameters are updated every 10 seconds to ensure matching with the current operating conditions.

[0104] Furthermore, the dynamic Imax threshold generation process implements multiple verifications: first, it checks whether the threshold exceeds the safe range of the rated current (50%-90%); second, it verifies whether the buffer width is reasonable (not less than 20ms); finally, it performs a logical consistency check to ensure that the upper buffer value is always greater than the lower buffer value. If the verification fails, the system automatically reverts to the previous valid value and triggers an alarm.

[0105] The dynamic Imax threshold output format includes: a threshold baseline value, an upper buffer value, a lower buffer value, a valid timestamp, and a status flag. Data is transmitted to subsequent comparison modules via a dedicated communication interface using Time-Triggered Ethernet (TTEthernet) to ensure real-time performance. Threshold update events are notified to relevant modules via a hardware interrupt mechanism, with a response latency of less than 50μs.

[0106] Step S300 includes at least steps S310-S330:

[0107] S310. Compare the standardized current waveform data with the dynamic Imax threshold, mark the high power state, and obtain the cumulative value of the high power duration.

[0108] Specifically, the standardized current waveform data comes from the output of module S120, and is a real-time current signal that has undergone temperature compensation and normalization processing, with a sampling rate of 1kHz and a data precision of 0.1%. The dynamic Imax threshold comes from the output of module S230, and includes three parameters: a reference value, an upper buffer, and a lower buffer, with an update period of 100ms. Understandably, the comparison operation is implemented using a hardware comparator, with a response time of less than 10μs.

[0109] Furthermore, the high-power status marking process includes the following steps: first, determining whether the current value exceeds the lower buffer of the dynamic Imax threshold; if it does, starting a high-power timer; and pausing the timer when the current value falls back below the lower buffer. The timer uses a 32-bit accumulator, with a minimum timing unit of 1ms and a maximum accumulation time of 24 hours. The timing value is stored in non-volatile memory to prevent loss due to power failure.

[0110] The high-power state marking implements a multi-verification mechanism: when three consecutive sampling points exceed the upper buffer, it is confirmed as a valid high-power state; when a single sampling point exceeds the upper buffer but subsequent points fall back, it is considered an interference signal and is not recorded. The marking process also includes anti-jitter processing: a 50ms delay confirmation window is set to avoid false triggering caused by instantaneous fluctuations.

[0111] The high-power duration cumulative value output is accompanied by status flags: normal state (cumulative value continues to increase), paused state (timekeeping pauses when current falls back), and abnormal state (cumulative value is frozen when data is abnormal). The cumulative value is transmitted to subsequent modules via the CAN bus with a transmission period of 100ms, and the data format is a 32-bit floating-point number according to the IEEE 754 standard.

[0112] In another embodiment, high-power state marking is implemented based on a modified Navier-Stokes equation. Specifically, the normalized current waveform data is defined as I. n(t)(from S120 output), the dynamic Imax threshold is I th (t)(from S230 output), establish the high-power state determination equation:

[0113]

[0114] in:

[0115] I n (t): Standardized current waveform data;

[0116] t: time variable;

[0117] v c : Current change velocity field vector, describing the propagation characteristics of the current waveform;

[0118] v d Current diffusion coefficient, which reflects the rate at which current diffuses in a conductor;

[0119] S p (t): Power state function, representing the degree of deviation of the current power state from the threshold;

[0120] α b Dynamic sensitivity coefficient;

[0121] I th (t): Dynamic current threshold;

[0122] To represent gradient operations in three-dimensional space;

[0123] This is the second spatial derivative of the current diffusion.

[0124] Furthermore, when S p (t)>β d High-power timer triggered at time:

[0125]

[0126] in:

[0127] T a : Cumulative value for high power duration;

[0128] H(·): Heaviside step function, when S p (t)>β d Output 1 if the condition is met, otherwise output 0.

[0129] β d The dynamic determination threshold is set to 5% of the generator's rated current.

[0130] t sHigh-power state start time;

[0131] t e High-power state end time;

[0132] This equation introduces fluid dynamics principles to model current fluctuations as viscous fluid motion, thus improving transient response speed compared to the traditional threshold method.

[0133] S320: Call the generator temperature rise model to process the cumulative value of high power duration, and calculate the recommended Tmax and Trest parameters in combination with the SOC value.

[0134] Specifically, the generator temperature rise model is a digital twin model based on the thermal network method, containing 20 hot nodes and 15 thermal resistance parameters, with a simulation step size of 1 second. Model inputs include: cumulative high-power duration, current winding temperature (from S100), ambient temperature, and cooling airflow rate. Model outputs are the predicted winding hot spot temperatures and the rate of temperature rise.

[0135] Understandably, the SOC value comes from the output of the S130 module, with an update cycle of 1 second and an accuracy of 0.1%. The processing method for incorporating the SOC value is as follows: when the SOC is below 30%, a temperature weighting coefficient is added; when the SOC is above 80%, a time weighting coefficient is added. The weighting coefficients are obtained through a two-dimensional interpolation table, and the table data comes from experimental calibration.

[0136] Furthermore, the proposed Tmax parameter calculation process is as follows: First, based on the temperature rise rate output by the temperature rise model, the remaining time to reach the maximum allowable temperature (155℃) is calculated; second, the SOC factor is considered to correct the remaining time; finally, 80% of the corrected value is taken as the proposed Tmax. The Trest parameter calculation is based on the thermal time constant: by fitting the temperature drop curve through the model, the time required for the temperature to fall back to the safe threshold (105℃) is calculated, and then multiplied by the SOC correction factor.

[0137] The parameter calculation module performs real-time self-checks: it verifies the model calculation time every 60 seconds, automatically switching to a simplified model if it exceeds 100ms; it performs model accuracy verification every 24 hours, triggering a calibration reminder if the error exceeds 5%. The recommended parameter output format includes: baseline value, upper and lower limits, confidence level, and effective time, which is transmitted to the decision module through a dedicated data channel.

[0138] In another embodiment, a temperature rise model is constructed based on the improved Hamiltonian equation. Specifically, the thermodynamic Hamiltonian of the generator is established:

[0139]

[0140] in:

[0141] Thermodynamic Hamiltonian of generator system;

[0142] p h Thermal momentum describes the inertia of temperature changes;

[0143] m e Equivalent thermal mass reflects the thermal capacity characteristics of generator components;

[0144] V t (q): Potential energy function, which includes the thermodynamic potential of winding copper loss and iron loss;

[0145] λ c Electromechanical coupling coefficient; (range 0.8-1.2)

[0146] γ s SOC Correction Index;

[0147] SOC: State of Charge of the battery;

[0148] Furthermore, the optimal parameters are solved using canonical equations:

[0149]

[0150] in:

[0151] q: Generalized coordinates, representing the thermodynamic state variables of the system;

[0152] Partial derivative of Hamiltonian with respect to thermal momentum;

[0153] The partial derivative of the Hamiltonian with respect to generalized coordinates;

[0154] Output, T m Recommended maximum continuous working time; T r Suggested interval time.

[0155] The model will use electrical parameter T a The SOC is converted into a thermodynamic quantity, and the temperature prediction error is less than ±3℃.

[0156] S330. When the cumulative value of high power duration exceeds the recommended Tmax parameter, trigger the intermittent command and generate the target Trest value.

[0157] Specifically, the trigger condition determination adopts a window comparison method: a 10ms judgment window is set, and when the cumulative value of the high power duration continuously exceeds the recommended Tmax parameter within the window, it is confirmed as a valid trigger. The intermittent command includes three fields: command type (immediate cutoff / delayed cutoff), cutoff level (complete cutoff / partial derating), and expected recovery time, encoded as an 8-byte data packet.

[0158] The target Trest value is generated using a tiered strategy: the base value is taken from the recommended Trest parameter; a first-level correction is performed based on the SOC value (shortening by 10% when SOC is below 20% and extending by 15% when SOC is above 90%); and a second-level correction is performed based on the historical intermittent frequency (extending by an additional 5% when used at high frequencies). The target Trest value is a minimum of 30 seconds and a maximum of 15 minutes, adjusted in 10-second increments.

[0159] Furthermore, the triggering process incorporates safety interlocks: when an abnormal generator speed is detected (fluctuation exceeding ±5%), triggering commands are prohibited; when the battery temperature exceeds 45°C, a complete shutdown mode is forcibly adopted. The command transmission employs a redundant design: the main channel is a CAN FD bus, and the backup channel is a hard-wired signal, ensuring reliability.

[0160] The target Trist value is output along with its execution priority: normal priority (asynchronous execution), high priority (synchronous immediate execution), and urgent priority (direct hardware triggering). The value is transmitted via a dual mechanism of shared memory and message queues to ensure real-time performance and consistency. Each triggered event generates a log record containing complete information such as timestamp, triggering conditions, and execution result.

[0161] In another embodiment, the target Trest value is generated based on an improved Monte Carlo method.

[0162] Define the trigger probability density function:

[0163]

[0164] in:

[0165] P(T a ): Probability density of high-power state triggering;

[0166] Z n : Normalization constant, ensuring the probability integral is 1;

[0167] T m Recommended maximum continuous working time;

[0168] σ t Uncertainty parameters of the temperature rise model (range 0.1-0.3);

[0169] ηs SOC weighting coefficient;

[0170] exp (natural exponential function): the standard exponential function;

[0171] Furthermore, using Markov chain Monte Carlo sampling:

[0172]

[0173] in:

[0174] Optimal interval time decision value;

[0175] Expectation value operator;

[0176] argmax: The parameter corresponding to the maximum probability;

[0177] f c (T a Cooling efficiency function (range 0-1);

[0178] Among them, f c (·) represents the cooling efficiency function: K is the heat dissipation coefficient, reflecting the efficiency of the generator cooling system; e is the natural constant.

[0179]

[0180] The algorithm completes 1000 iterations within 10ms and outputs... Used for S420 timer initialization.

[0181] Step S400 includes at least steps S410-S430:

[0182] S410: Analyze the trigger intermittent instruction, encode the MOSFET drive signal of the charging circuit, and obtain the cut-off control signal.

[0183] Specifically, the trigger intermittent command originates from the output of the S330 module and is a data packet containing the command type, cutoff level, and expected recovery time. It is transmitted using the CAN FD protocol at a transmission rate of 2Mbps. The parsing process is implemented through a dedicated protocol parsing chip, with a parsing delay of less than 50μs. Understandably, the MOSFET drive signal encoding in the charging circuit uses PWM modulation, with a carrier frequency set to 20kHz and a duty cycle adjustment accuracy of 0.1%.

[0184] Furthermore, the generation of the cut-off control signal includes the following steps: first, selecting the cut-off mode (hard cut-off / soft cut-off) according to the instruction type; second, determining the PWM duty cycle according to the cut-off level (0% for hard cut-off, 30%-70% adjustable for soft cut-off); and finally, superimposing fault protection signals (overcurrent, overtemperature, and other protection signals). The signal generation module uses an isolated gate driver with a drive current capability of up to 4A and a rise / fall time of less than 100ns.

[0185] The drive signal encoding implements multiple protection mechanisms: at the hardware level, dead time control (minimum 1μs) is set to prevent shoot-through between the upper and lower transistors; at the software level, duty cycle gradual control is implemented, with each adjustment not exceeding 5% steps. The cut-off control signal output is accompanied by a status feedback signal to monitor the MOSFET's Vgs voltage and drain current in real time to ensure accurate switching status.

[0186] The cutoff control signal is transmitted through dual channels: the main channel uses fiber optic isolation for strong anti-interference capabilities; the backup channel uses differential signal transmission for redundancy. Signal transmission delay is controlled within 200μs to ensure real-time system response. The output interface uses industry-standard connectors with an IP67 protection rating.

[0187] S420: Set the intermittent timer parameters according to the target Trest value, perform countdown initialization, and generate the timer start command.

[0188] Specifically, the target Trest value comes from the output of the S330 module and is an interval duration parameter corrected by the SOC, ranging from 30 seconds to 15 minutes with a resolution of 1 second. The interval timer is a 32-bit hardware timer, and the reference clock source is a temperature-compensated crystal oscillator with a frequency stability of ±5ppm.

[0189] Understandably, the parameter setting process includes: first, converting the target Trest value into the number of timer count pulses (clock frequency 1MHz); second, configuring the timer operating mode (single / cycle); and finally, setting the interrupt trigger condition (timeout / mid-cycle alarm). The initialization process implements a verification mechanism: after writing the parameters, a readback and comparison are performed immediately, and if there is a discrepancy, the process is automatically retried 3 times.

[0190] Furthermore, the countdown initialization employs a tiered loading strategy: a base value loading phase (directly writing the target value); a fine-tuning phase (dynamically adjusting according to system load); and a final confirmation phase (double-checking to lock parameters). The timer start command includes a start command, a pause command, and a reset command, encoded as a 3-bit binary code.

[0191] The timer parameter settings implement temperature compensation: a built-in temperature sensor monitors the ambient temperature in real time and dynamically adjusts the timing parameters according to the temperature-crystal frequency characteristic curve. The start command is transmitted via a dedicated control bus, using Time Triggered Ethernet (TTEthernet) as the transmission protocol to ensure timing determinism.

[0192] S430, synchronously cut off control signal and timer start command, generate power execution control package.

[0193] Specifically, the synchronization operation is implemented through a hardware synchronization controller, achieving a synchronization accuracy of 100ns. The power execution control packet is a fixed-format data frame, containing: a cutoff control signal (16-bit), timer parameters (32-bit), a synchronization timestamp (64-bit), and a checksum (16-bit).

[0194] Understandably, the generation process is implemented in a pipeline: the first stage assembles control signal data; the second stage embeds timing parameters; the third stage adds synchronization markers; and the fourth stage calculates checksums. The control packet generation cycle is 100ms, and a double buffering mechanism is used to ensure data continuity.

[0195] Furthermore, the synchronization mechanism includes dual guarantees of clock synchronization and event synchronization: clock synchronization is based on the IEEE 1588 Precision Time Protocol (PTP), with a master-slave clock deviation of less than 1 μs; event synchronization uses hardware trigger signals, with a response delay of less than 50 ns. The control packet output interface is a Gigabit Ethernet, with a transmission rate of 1 Gbps, and supports CRC32 checksum and retransmission mechanism.

[0196] The power execution control packets are stored in non-volatile memory, retaining records of the most recent 100 operations, including complete timestamps and execution result feedback. The data packet format is compatible with industry standard protocols, supporting parsing and monitoring by third-party devices. Each generated operation produces an operation log, recording key parameters and system status.

[0197] Step 500 includes at least steps S510-S530:

[0198] S510 receives the power execution control package, drives the MOSFET switch to cut off the charging circuit, and enters intermittent mode.

[0199] Specifically, the power execution control packet originates from the output of the S430 module and is a data structure containing a cutoff control signal, timer parameters, and a synchronization timestamp. It is transmitted using the Gigabit Ethernet protocol at a transmission rate of 1Gbps. The receiving process is implemented through a dedicated network protocol stack, with a receiving delay controlled within 200μs. Understandably, the MOSFET switch drive circuit employs an isolated gate driver, with a peak drive current of 10A and a switching speed of less than 100ns.

[0200] Furthermore, the charging circuit cutoff operation implements a graded control strategy: the first stage is pre-cutoff (PWM duty cycle drops to 30% for 50ms); the second stage is soft cutoff (duty cycle linearly drops to 0% for 100ms); and the third stage is hard cutoff (completely shut off and activates the body diode freewheeling). The cutoff process monitors the drain-source voltage and gate current in real time, and automatically triggers the protection mechanism when abnormal conduction is detected.

[0201] The criteria for entering intermittent mode include: the charging current dropping below 5% of the rated value for 10ms; the MOSFET die temperature being below the safety threshold; and the generator speed fluctuation range being less than ±2%. The mode switching signal is output through an opto-isolated relay with a response time of less than 1ms, simultaneously triggering the status indicator light and system log recording.

[0202] The drive circuit implements a triple protection design: hardware overcurrent protection (fast-acting fuse), software overtemperature protection (temperature sensor feedback), and mechanical interlock (mechanical indication of cut-off status). All protection signals are aggregated to the safety monitoring unit to form a forced interlock logic, ensuring the reliability of the cut-off operation.

[0203] S520: Start the intermittent timer to monitor the intermittent duration. When the target Trest value is reached, generate a resume charging command.

[0204] Specifically, the intermittent timer is derived from the initialization configuration of the S420 module. It is a 32-bit high-precision timer, and the reference clock uses a temperature-compensated crystal oscillator with a frequency stability of ±1ppm. The startup process includes clock calibration (synchronizing with the system master clock), initial value loading (writing the target Trest value), and working mode setting (single countdown).

[0205] Understandably, the monitoring process implements a dynamic compensation mechanism: adjusting timing parameters based on changes in ambient temperature (temperature coefficient 0.1ppm / ℃); and correcting the clock reference based on power quality fluctuations (voltage compensation coefficient 0.05% / V). The monitoring data is recorded every 100ms, including remaining time, clock drift, and temperature influence factors.

[0206] Furthermore, the recovery charging command generation adopts a multi-condition triggering strategy: the primary condition is the arrival of the target Trest value; auxiliary conditions include the generator temperature dropping to a safe threshold, lithium battery SOC demand triggering, etc. The command is encoded as an 8-bit control word, containing the recovery mode (gradual / direct), target current, and slope control parameters.

[0207] The timer implements self-diagnostic operation: it performs an integrity check (counting pulse verification) every 10 seconds and clock drift calibration every minute (comparison with GPS clock). When an anomaly is detected, it automatically switches to a backup clock source to ensure timing continuity. Recovery command transmission uses redundant channels (CAN bus and hardwired parallel), with a transmission delay of less than 50μs.

[0208] S530: Collects generator temperature change data and current recovery response data during intermittent mode, and generates an execution status feedback report.

[0209] Specifically, the temperature change data is collected by 12 PT100 sensors arranged on the generator stator windings, bearings, and radiators, with a sampling period of 100ms and a resolution of 0.1℃. The current recovery response data is measured using a high-precision Hall sensor with a bandwidth of 100kHz and a linearity of ±0.2%.

[0210] Furthermore, the data acquisition employs a synchronous sampling strategy: all sensor channels use the same sample-and-hold signal; the time alignment accuracy of current and temperature data reaches 10 μs. Real-time filtering (IIR low-pass filtering, cutoff frequency 1 kHz) and outlier removal (3σ criterion) are implemented during the acquisition process.

[0211] The execution status feedback report generation includes the following data processing stages: raw data preprocessing (outlier removal, baseline correction); feature parameter extraction (temperature drop slope, current recovery time constant); and comprehensive evaluation index calculation (thermal stress coefficient, electrical stress index). The report uses a standardized XML format and includes data blocks, quality flags, and metadata.

[0212] The feedback report transmission employs a circular storage mechanism: the most recent 100 intermittent periodic data are stored locally (using non-volatile memory); and uploaded in real-time to a cloud database (using 4G wireless transmission). Report generation triggers system log events, recording key parameters and execution context for subsequent analysis. All data is digitally signed to ensure integrity and immutability.

[0213] Step S600 includes at least steps S610-S630:

[0214] S610. Analyze the temperature drop data in the execution status feedback report and calculate the generator model error coefficient.

[0215] Specifically, the execution status feedback report comes from the output of the S530 module and is a structured dataset containing generator temperature change curves and current recovery characteristics during intermittent mode. The data sampling interval is 100 milliseconds, and the storage format uses both binary and JSON encoding. The temperature drop data analysis uses a piecewise linear regression method to divide the entire intermittent cycle into three characteristic intervals: a heating segment (30 seconds after cutoff), a steady-state segment, and a cooling segment.

[0216] Furthermore, the error coefficient calculation employs a multi-dimensional evaluation: in the time domain, the root mean square error between the measured temperature curve and the model-predicted curve is calculated; in the frequency domain, the difference in temperature fluctuation spectrum is analyzed using Fast Fourier Transform; and in the spatial domain, the temperature gradient distribution differences at different locations of the generator (winding ends, bearing housings, radiators) are compared. The calculation process uses a sliding time window mechanism, with the window width set to 10 intermittent data periods.

[0217] The generator model error coefficient comprises four components: steady-state error coefficient (reflecting long-term prediction deviation), dynamic error coefficient (characterizing transient response differences), spatial error coefficient (describing temperature field distribution deviation), and aging correction coefficient (reflecting performance degradation trends). Each coefficient ranges from 0.8 to 1.2, with an accuracy of 0.001. The coefficient calculation module has a built-in self-test function, performing a benchmark test every 24 hours to ensure the stability of the core calculation algorithm.

[0218] An anomaly handling mechanism is implemented during data analysis: when a temperature sensor failure (such as open circuit or short circuit) is detected, interpolation compensation of adjacent sensor data is automatically activated; when abnormal fluctuations in current data exceed a threshold, a data quality flag is triggered and the anomaly event is recorded. Detailed logs are generated for all processing steps, including raw data, processing methods, and correction results.

[0219] S620. Correct the parameters of the temperature rise model based on the generator model error coefficient, and update the recommended calculation rules for Tmax and Trest.

[0220] Specifically, the temperature rise model parameter correction is achieved using the recursive least squares method, with each correction adjusting only the three parameters with the greatest impact (thermal resistance, heat capacity, and time constant). The correction process includes three stages: parameter sensitivity analysis (calculating the Jacobian matrix), correction amount calculation (based on error coefficient weighting), and stability verification (Lyapunov exponent detection).

[0221] Understandably, the Tmax calculation rule update implements a hierarchical strategy: the basic rule layer maintains the physical principles unchanged; the intermediate rule layer adjusts the temperature-time mapping coefficient; and the application rule layer optimizes the safety margin setting. The Trest calculation rule update focuses on thermal time constant compensation, dynamically adjusting the cooling curve fitting parameters based on the error coefficient.

[0222] Furthermore, the rule update process implements version control: each modification generates a new rule version number (V1.0.0 format); historical versions are retained for rollback; version difference records include modified content, effective time, and impact assessment. New rules must be verified through simulation testing before deployment, with test cases covering typical operating conditions, boundary conditions, and fault scenarios.

[0223] The calculation rules are represented using an object-oriented data structure, including parameter domains, constraints, and an incidence matrix. The rules are stored in non-volatile memory with ECC verification, supporting remote updates and partial refreshes. Each rule update triggers a system event, notifying relevant modules to make adaptation adjustments and ensuring system interoperability.

[0224] S630. The updated suggested Tmax and Trest calculation rules are sent back to the collaborative decision engine to complete the closed-loop learning.

[0225] Specifically, the feedback process adopts a publish-subscribe model, and the data channels include: a real-time data bus (for transmitting emergency updates), a configuration management channel (for transmitting complete rule sets), and a log synchronization link (for recording transmission status). The collaborative decision engine is located in the system control core and includes three functional units: a rule interpreter, a policy arbitrator, and an exception handler.

[0226] Furthermore, the closed-loop learning mechanism comprises three layers: a data layer establishing a two-way verification mechanism (cross-validation of uploaded rules and feedback results); an algorithm layer implementing incremental learning (retaining 10% of historical data weights with each update); and a system layer achieving multi-replica consistency (real-time synchronization of primary and backup decision engines). The learning process generates a knowledge graph, recording parameter relationships and adjustment history.

[0227] The transmitted rules employ multiple security safeguards: AES-256 encryption is used at the transport layer; digital signatures are added at the data layer; and firewall rules are set at the system layer. The transmission protocol defines a comprehensive state machine, including four states: ready, in transit, checksum, and active. State transitions require strict conditional checks.

[0228] After completing the closed-loop learning process, the system performs comprehensive functional verification: basic verification testing (checking the completeness of the rule syntax); integration testing (verifying compatibility with interfaces of various modules); and scenario testing (simulating the effectiveness of the rules under typical working conditions). All test results are compiled into a verification report, serving as an important output of the learning process.

[0229] Example 2: Figure 2 A structural block diagram of a generator-lithium battery cooperative control method and system according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include:

[0230] The multi-source sensing module 10 is used to collect lithium battery current, SOC status, and temperature data, and output standardized current waveform data and a time-labeled sensing dataset. The multi-source sensing module acquires the lithium battery charging current signal in real time through a high-precision Hall sensor array, simultaneously obtaining SOC status data provided by the battery management system (BMS) and temperature data from PT100 temperature sensors distributed at key parts of the generator. The sampling frequency is set to 10kHz to ensure transient characteristic capture. This module incorporates a three-stage signal processing pipeline: first, an adaptive Kalman filter is used to eliminate high-frequency noise introduced by PWM charging; second, an improved Z-score algorithm normalizes the current signal to a standard normal distribution; and finally, a GPS-disciplined clock source is used to timestamp all data with microsecond-level precision, generating a time-labeled sensing dataset. This preprocessed standardized data is transmitted to downstream modules via a CAN FD bus, with transmission delay controlled within 200μs.

[0231] The dynamic parameter quantization module 20 is used to calculate the dynamic Imax threshold and generate a buffered adjustment threshold. After receiving standardized current waveform data, the module first establishes an initial reference threshold based on the rated current value in the generator nameplate parameters. Then, it extracts transient fluctuation envelope features from the current signal using an improved Hilbert-Huang transform. The module's built-in adaptive algorithm dynamically adjusts two key parameters: first, it corrects the thermal derating factor in real time based on the generator winding temperature (derating by 5% for every 10°C increase in temperature); second, it calculates the 95% confidence interval of the current fluctuation using a sliding window statistical method as the buffer band width. The final output dynamic Imax threshold not only includes the reference value but also includes upper and lower buffer boundaries, forming a three-dimensional threshold vector. This design improves the system's tolerance to load mutations by more than 40%.

[0232] The collaborative decision engine 30 is used to invoke the temperature rise model to output suggested parameters and trigger intermittent commands. The collaborative decision engine is the intelligent core of the system, employing a hybrid architecture processing strategy. At the hardware level, a dedicated DSP chip runs the temperature rise model, which transforms the current thermal effect into a three-dimensional temperature field distribution based on the improved Navier-Stokes equations. At the software level, a two-layer decision tree is constructed: the first layer generates a primary warning signal by comparing the continuous deviation time of the standardized current from the dynamic threshold; the second layer performs multi-objective optimization based on the lithium battery's SOC state, outputting two key parameters—the suggested maximum continuous operating time Tmax (accuracy ±1 second) and the recommended intermittent duration Trest (adjustable step size 10 seconds). When the cumulative value of the high-power duration exceeds Tmax, the engine immediately triggers a graded protection command, with command priority divided into three levels: normal, emergency, and critical, based on the temperature rise rate.

[0233] The control strategy generation module 40 is used to generate MOSFET cutoff control signals and power execution control packages. This module converts decision commands into executable control signals. For the generation of the cutoff control signal, the module employs PWM soft-shutdown technology: first, the duty cycle is reduced from 100% to 30% in 5% steps over 100ms, and then a hard-shutdown operation is performed. This stepped load reduction strategy avoids voltage spikes caused by sudden current changes. The synchronously generated timer parameter package includes the target interval duration Trest and its allowable fluctuation range (±5%). All control commands are transmitted via fiber optic channel to ensure anti-interference capability, with a transmission error rate of less than 10%. -9 .

[0234] The power execution module 50 is used to perform charging circuit disconnection and intermittent timing operations. The power execution module adopts a redundant design architecture. The main execution unit contains a parallel SiC MOSFET array with an on-resistance of only 15mΩ, reducing switching losses by 60% compared to traditional IGBTs; the backup execution unit uses an electromechanical relay as a backup channel. Key parameters monitored in real time by the module include: MOSFET junction temperature (triggers protection when exceeding 125℃), DC bus voltage fluctuation (alarm when exceeding ±10% of rated value), and leakage current (threshold set at 5mA). Real-time status data collected during execution is converted into digital signals via an isolated ADC and compared with a preset safe operating area (SOA) for verification.

[0235] The dynamic learning feedback module 60 is used to correct the temperature rise model parameters and update the calculation rules. This module constitutes the self-evolutionary hub of the system. By analyzing the temperature drop curve and current recovery characteristics within each intermittent cycle, this module calculates three key correction parameters: the temperature rise model error coefficient (based on the maximum deviation between measured and predicted temperatures), the time constant compensation factor (reflecting changes in heat dissipation efficiency), and the SOC weight adjustment parameter (optimizing the proportion of battery state influence). The updated parameter set is transmitted back to the collaborative decision engine through a secure encrypted channel, completing the control loop. The learning process employs an incremental update strategy, adjusting only the three parameters with the greatest impact on system performance each time to ensure stability.

[0236] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A generator-lithium battery coordinated control method, characterized in that, include: Collect lithium battery current, SOC status and temperature data, and after filtering, normalization and timestamp alignment, obtain standardized current waveform data and a sensing dataset with time-series labels. Based on the generator rated current and standardized current waveform data, the dynamic Imax threshold is calculated, and a buffered dynamic Imax threshold is generated by combining the fluctuation envelope. Compare the cumulative high-power time of the standardized current waveform data with the dynamic Imax threshold, call the temperature rise model to output suggested Tmax and Trest parameters, trigger intermittent commands and generate the target Trest value; The expression for the cumulative high-power time of comparing the standardized current waveform data with the dynamic Imax threshold is as follows: Among them, T a S represents the cumulative value during high power duration; H(·) is the Heaviside step function; S p (t) is the power state function; β d For dynamic determination of threshold; t s t is the start time of high-power state. e This is the end time of the high-power state; The expression for triggering the intermittent command and generating the target Trest value is: in, The optimal interval time decision value; For expectation value operators; argmax is the parameter corresponding to the maximum probability; f c (T a P(T) is the cooling efficiency function; a ) represents the probability density of high-power state triggering; The system analyzes the trigger intermittent command to generate a MOSFET cutoff control signal, synchronizes the target Trest value to initialize the timer, and generates a power execution control package. Receive power execution control package, execute charging circuit cut-off and intermittent timing, collect temperature and current recovery data and generate execution status feedback report; Analyze the temperature drop data in the execution status feedback report, correct the temperature rise model parameters, update the Tmax and Trest calculation rules and send them back to the collaborative decision engine to complete the closed-loop learning.

2. The method according to claim 1, characterized in that, The collected lithium battery current, SOC status, and temperature data are processed through filtering, normalization, and timestamp alignment, including: The lithium battery charging current signal and battery SOC status data are acquired, and signal filtering and AD conversion are performed to obtain digital current signal and SOC value. Fluctuation characteristics are extracted from digital current signals, and combined with generator temperature sensor data, signal normalization processing is performed to obtain standardized current waveform data. The standardized current waveform data and SOC values ​​are timestamped to generate a sensing dataset with time-series labels.

3. The method according to claim 1, characterized in that, The calculation of the dynamic Imax threshold, combined with the fluctuation envelope to generate a buffered dynamic Imax threshold, includes: Based on the generator rated current parameters and standardized current waveform data, dynamic threshold calculation is performed to obtain the initial Imax threshold. The pulse fluctuation characteristics are analyzed from the standardized current waveform data, and the fluctuation envelope is fitted to obtain the current fluctuation compensation coefficient. The initial Imax threshold is adjusted based on the current fluctuation compensation coefficient to generate a dynamic Imax threshold with a buffer.

4. The method according to claim 1, characterized in that, The proposed Tmax and Trest parameters output by the temperature rise model include: By comparing the standardized current waveform data with the dynamic Imax threshold, high-power state marking is performed to obtain the cumulative value of high-power duration; The generator temperature rise model is called to process the cumulative value of high power duration, and combined with the SOC value, the suggested Tmax and Trest parameters are calculated. When the cumulative high-power duration exceeds the recommended Tmax parameter, an intermittent command is triggered and a target Trest value is generated.

5. The method according to claim 1, characterized in that, The generation of the MOSFET cutoff control signal includes: The trigger intermittent command is parsed, and the MOSFET drive signal of the charging circuit is encoded to obtain the cut-off control signal.

6. The method according to claim 1, characterized in that, The synchronization target Trest value initialization timer includes: Set the intermittent timer parameters according to the target Trest value, perform countdown initialization, and generate the timer start command.

7. The method according to claim 1, characterized in that, The generated power execution control package includes: Synchronously cut off the control signal and the timer start command to generate a power execution control package.

8. The method according to claim 1, characterized in that, The execution of the charging circuit cutoff and intermittent timing includes: Upon receiving the power execution control package, the MOSFET switch is driven to cut off the charging circuit and enter intermittent mode; Start an intermittent timer to monitor the intermittent duration. When the target Trest value is reached, generate a resume charging command.

9. The method according to claim 1, characterized in that, The generation of the execution status feedback report includes: Collect generator temperature change data and current recovery response data during intermittent mode, and generate an execution status feedback report.

10. A generator-lithium battery collaborative control system, used to implement the generator-lithium battery collaborative control method according to any one of claims 1-9, characterized in that, include: The multi-source sensing module is used to collect lithium battery current, SOC status and temperature data, and output standardized current waveform data and sensing dataset with time-series labels. The dynamic parameter quantization module is used to calculate the dynamic Imax threshold and generate an adjusted threshold with a buffer. The collaborative decision-making engine is used to call the temperature rise model to output suggested parameters and trigger intermittent commands. The control strategy generation module is used to generate MOSFET cutoff control signals and power execution control packages; The power execution module is used to perform charging circuit disconnection and intermittent timing operations. The dynamic learning feedback module is used to correct the temperature rise model parameters and update the calculation rules.

Images