Cluster-level SOC dynamic weighting calculation method and system in frequency modulation scene

Abstract

The invention discloses a cluster-level SOC dynamic weighting calculation method and system in a frequency modulation scene. The method comprises the steps that the change rate of power grid frequency deviation and the current SOC value and health state parameter of each battery cluster are monitored in real time; dynamically determining an emergency degree coefficient of frequency modulation response according to the change rate of the frequency deviation, and calculating a time-varying weight coefficient corresponding to the current frequency modulation demand by combining the health state parameters of each battery cluster; dynamic weighted fusion is carried out on the SOC values of the battery clusters based on the time-varying weight coefficient, and an equivalent SOC evaluation value reflecting the frequency modulation capability of the power grid system in real time is generated; and generating a power distribution instruction of each battery cluster according to the equivalent SOC evaluation value so as to realize optimal control based on cluster-level SOC dynamic weighting in a frequency modulation scene. According to the embodiment of the invention, the dynamic adaptation of the power grid frequency modulation demand and the fine management of the battery cluster state can be realized, and the response speed, the control precision and the overall service life of the energy storage system in the frequency modulation process are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of battery clusters, and particularly relates to a cluster-level SOC dynamic weighting calculation method and system in a frequency modulation scene. BACKGROUND

[0002] In recent years, with the increasing proportion of new energy power generation, the frequency stability of the power grid is increasingly prominent, and the battery energy storage system becomes an important means of frequency modulation due to its fast response capability. At present, the energy storage power station usually adopts a multi-cluster battery structure for cooperative operation, but when participating in the frequency modulation of the power grid, the traditional method usually uses the simple average value or fixed weighting value of the SOC of each battery cluster to evaluate the total available energy of the system and allocate power, without fully considering the dynamic characteristics of the change of the frequency of the power grid and the different state of each battery cluster. This easily leads to overcharging and overdischarging of part of the battery clusters, accelerates the aging, and cannot accurately reflect the real-time frequency modulation capability of the system when the frequency changes sharply, thereby affecting the recovery effect of the frequency of the power grid and the economy and safety of the operation of the energy storage system. SUMMARY

[0003] The application aims to provide a cluster-level SOC dynamic weighting calculation method and system in a frequency modulation scene, so as to solve the problems in the prior art, realize dynamic adaptation to the frequency modulation demand of the power grid and fine management of the state of the battery cluster, and improve the response speed, control accuracy and overall life of the energy storage system in the frequency modulation process.

[0004] One embodiment of the application provides a cluster-level SOC dynamic weighting calculation method in a frequency modulation scene, which comprises the following steps: real-time monitoring of the change rate of the frequency deviation of the power grid and the current SOC value and health state parameters of each battery cluster; dynamically determining an emergency degree coefficient of the frequency modulation response according to the change rate of the frequency deviation, and calculating a time-varying weight coefficient corresponding to the current frequency modulation demand in combination with the health state parameters of each battery cluster; dynamically weighting and fusing the SOC values of each battery cluster based on the time-varying weight coefficient, to generate an equivalent SOC evaluation value that reflects the frequency modulation capability of the power grid system in real time; generating a power distribution instruction for each battery cluster according to the equivalent SOC evaluation value, so as to realize the optimization control based on the cluster-level SOC dynamic weighting in the frequency modulation scene.

[0005] Another embodiment of the application provides a cluster-level SOC dynamic weighting calculation system in a frequency modulation scene, which comprises the following steps: a monitoring module for real-time monitoring of the change rate of the frequency deviation of the power grid and the current SOC value and health state parameters of each battery cluster; The computing module is used for dynamically determining an emergency degree coefficient of frequency modulation response according to the change rate of the frequency deviation, and calculating a time-varying weight coefficient corresponding to the current frequency modulation demand in combination with the state of health parameters of each battery cluster; The fusing module is used for dynamically weighting and fusing the SOC values of each battery cluster based on the time-varying weight coefficient, to generate an equivalent SOC evaluation value reflecting the frequency modulation capability of the power grid system in real time; The distributing module is used for generating power distribution instructions of each battery cluster according to the equivalent SOC evaluation value, to realize the optimization control based on the cluster-level SOC dynamic weighting in the frequency modulation scene.

[0006] Another embodiment of the present application provides a storage medium, which stores a computer program, wherein the computer program is configured to execute the method described in any one of the above embodiments when running.

[0007] Another embodiment of the present application provides an electronic device, which comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the computer program to execute the method described in any one of the above embodiments.

[0008] Compared with the prior art, the cluster-level SOC dynamic weighting calculation method in the frequency modulation scene provided by the present application can realize the dynamic adaptation to the frequency modulation demand of the power grid and the fine management of the state of the battery cluster, and improve the response speed, the control precision and the overall life of the energy storage system in the frequency modulation process. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 FIG. 1 is a hardware structure block diagram of a computer terminal of the cluster-level SOC dynamic weighting calculation method in the frequency modulation scene provided by the embodiment of the present application; Figure 2 FIG. 2 is a flowchart of the cluster-level SOC dynamic weighting calculation method in the frequency modulation scene provided by the embodiment of the present application; Figure 3 FIG. 3 is a structure diagram of the cluster-level SOC dynamic weighting calculation system in the frequency modulation scene provided by the embodiment of the present application. DETAILED DESCRIPTION

[0010] The embodiments described below with reference to the drawings are exemplary and are intended to explain the present application only, and cannot be interpreted as a limitation on the present application.

[0011] Figure 1 A hardware structure block diagram of a computer terminal of a cluster-level SOC dynamic weighting calculation method in a frequency modulation scene provided for an embodiment of the present application is shown in FIG. 1. Figure 1 As shown in the figure, the computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory can include a non-volatile storage medium and an internal memory.

[0012] Referring to Figure 2 , the embodiment of the present application provides a cluster-level SOC dynamic weighting calculation method in a frequency modulation scene, which can include the following steps: S201, real-time monitoring of the change rate of the grid frequency deviation and the current SOC value and health state parameter of each battery cluster; Specifically, the grid frequency signal can be collected in real time, and the change rate of the frequency deviation is calculated by using a difference algorithm to obtain frequency change rate data; I. Hardware configuration and parameters of frequency collection The frequency collection unit is constructed by using an “STM32H743 microcontroller + AD7606 16-bit ADC module”, and is matched with a grid voltage transformer (ratio 100V / 5V, adaptive to the ADC input range), and the core parameter design is as follows: Sampling frequency: 10Hz (i.e. collecting the grid frequency signal once every 100ms), the reason for selecting this frequency is that the typical time scale of the grid frequency fluctuation is several hundred milliseconds (the frequency needs 1-2 seconds to reach the maximum deviation after load mutation), and 10Hz sampling can both capture dynamic changes and avoid data redundancy; Measurement accuracy: through the oversampling technology (sampling rate is increased by 4 times) of the ADC module, the frequency measurement accuracy is improved to ±0.01Hz, meeting the standard of frequency measurement error ≤0.02Hz; Anti-interference design: the acquisition circuit is built-in with a 50Hz notch filter (to suppress power frequency interference) and an RC low-pass filter (cutoff frequency 15Hz, to filter out high-frequency noise), to ensure that it can still be stably collected in the grid harmonic environment.

[0013] II. Calculation of frequency deviation and change rate Frequency deviation calculation: the rated frequency of the grid f_N=50Hz (China's power frequency standard), the actual frequency collected at a certain time is f(t), and the frequency deviation Δf(t) calculation formula at that time is: Δf(t)=f(t)-f_N Example: At t=20250926100000.100 (time stamp, accurate to milliseconds), the collected f(t)=50.02 Hz, then Δf(t)=50.02-50=+0.02 Hz (positive deviation indicates that the frequency is higher than the rated value, and power absorption is needed to suppress the frequency rise); at t=20250926100000.200, f(t)=50.05 Hz, Δf(t)=+0.05 Hz.

[0014] The differential algorithm calculates the rate of change: the rate of change of frequency deviation df / dt(t) (unit: Hz / s) reflects the trend of frequency deviation, and the calculation formula is the difference between the frequency deviations of the adjacent two time points divided by the time interval Δt (since the sampling frequency is 10 Hz, Δt=0.1 s): df / dt(t)=[Δf(t)-Δf(t-Δt)] / Δt.

[0015] Example: Based on the deviation values of the above two time points, Δf(t)=+0.05 Hz, Δf(t-Δt)=+0.02 Hz (t-Δt=20250926100000.100), then df / dt(t)=(0.05-0.02) / 0.1=+0.3 Hz / s; if t=20250926100000.300, f(t)=50.03 Hz, Δf(t)=+0.03 Hz, then df / dt(t)=(0.03-0.05) / 0.1=-0.2 Hz / s (negative rate of change indicates that the frequency deviation decreases, and the frequency modulation demand is alleviated).

[0016] III. Validity verification of frequency change rate data The collected change rate data needs to be verified by "range check" and "continuous stability check" to ensure effectiveness: Range check: The actual maximum value of power grid frequency change rate is usually not more than ±2 Hz / s (extreme case of load mutation), if df / dt(t)=±3 Hz / s is calculated, it is judged as abnormal collection, and the change rate value of the previous time is replaced; Continuous stability check: If the change rate fluctuation of 3 consecutive samples exceeds ±0.5 Hz / s (such as 0.3 Hz / s→-0.2 Hz / s→0.4 Hz / s), it is judged as noise interference, and a 3-point moving average filter is used (such as (0.3-0.2+0.4) / 3≈0.167 Hz / s), and the final output is the smoothed frequency change rate data.

[0017] The voltage, current and temperature data of each battery cluster are collected by the battery management system, and the current SOC value is calculated by integrating the ampere-hour method combined with the open circuit voltage method, and the battery cluster SOC data is obtained; I. Parameter collection and precision of BMS Each battery cluster (e.g. C1 cluster contains 10 single batteries, series voltage 30V) is configured with a slave unit in distributed BMS architecture, and the following parameters are collected: Cluster voltage: collected by ADS1256 ADC module, voltage division ratio 10:1 (adapted to 30V cluster voltage), precision ±1mV; Cluster current: collected by ACS712-50A current sensor, series in cluster loop, precision ±0.05A, positive when charging, negative when discharging; Cluster temperature: collected by DS18B20 digital sensor, pasted on the surface of the middle single battery in the cluster, precision ±0.3℃, sampling frequency 1Hz (consistent with the integral calculation period).

[0018] II. Real-time calculation of ampere-hour integral method The ampere-hour integral method is based on the principle that "the SOC change is proportional to the charging and discharging electric quantity", and the calculation formula is: SOC(t)=SOC0-(1 / (η・Cn))×∫(t0to t)i(τ)dτ.

[0019] Wherein, SOC(t): cluster SOC value at time t (%); SOC0: SOC value at the integral starting time t0 (%), determined after calibration by open circuit voltage method (e.g. t0=20250926090000, SOC0=80%); η: charging and discharging efficiency (η=0.95 when charging, η=0.98 when discharging, determined by the charging and discharging curve provided by the battery manufacturer, reflecting the electric quantity loss); Cn: rated capacity of battery cluster (Ah), e.g. C1 cluster Cn=100Ah; i(τ): cluster current at time τ (A), negative when discharging, positive when charging; ∫(t0to t)i(τ)dτ: cumulative charging and discharging electric quantity from t0 to t (Ah), because the sampling frequency is 1Hz, the integral is approximated by discrete summation, i.e. Σi(k)×Δt (Δt=1s=1 / 3600h).

[0020] Example (C1 cluster, t0=20250926090000, SOC0=80%, discharging state): At t=20250926090001, i(1)=-50A (negative sign indicates discharging), cumulative electric quantity=(-50)×1 / 3600≈-0.0139Ah; SOC(1)=80%-(1 / (0.98×100))×(-0.0139)×100%≈80%+0.0142%≈80.0142% (the integral term is negative when discharging, SOC decreases, the slight increase in the calculation result is due to the small current in the example, in actual large current discharging, SOC will decrease significantly); At t=20250926090002, i(2)=-55A, the cumulative charge is (-50-55) x 1 / 3600 ≈ -0.0292 Ah; SOC(2)=80%-(1 / (0.98 x 100)) x (-0.0292) x 100% ≈ 80%+0.0298% ≈ 80.0298% (In practical applications, multiple seconds of data need to be accumulated, and this is a simplified example).

[0021] III. Error calibration of open-circuit voltage method The ampere-hour integration method will have cumulative errors due to current measurement errors and efficiency fluctuations, and needs to be calibrated every 2 hours (or after each charge and discharge cycle) using the open-circuit voltage method: Mapping relationship between open-circuit voltage (OCV) and SOC: obtained through static testing of the battery cluster, such as the OCV-SOC mapping of C1 cluster: OCV=3.65V when SOC=100%, OCV=3.30V when SOC=50%, OCV=2.50V when SOC=0%, and linear interpolation is used in the intermediate interval; Calibration process: the battery cluster is left for 10 minutes (to ensure that the polarization voltage is eliminated and the true OCV is measured), if the measured OCV of C1 cluster is 3.30V, corresponding to SOC=50%, while the ampere-hour integration calculation gives SOC=48%, then the integrated SOC is corrected to 50%, and the integrated starting value SOC0 is reset to 50%, eliminating the cumulative error; Calibration accuracy: through this method, the SOC calculation error can be reduced from ±5% of the ampere-hour integration to ±2%, meeting the accuracy requirements in the frequency modulation scene.

[0022] Based on the historical operation data of the battery, the internal resistance growth rate and the capacity attenuation rate of each battery cluster are calculated, combined with the temperature influence factor to evaluate the health status, and the battery health status parameters are obtained; I. Source and extraction of historical operation data The non-volatile memory (EEPROM) of the BMS stores the historical data of each battery cluster, including: Initial parameters: initial internal resistance R0 of new battery cluster (such as C1 cluster R0=0.002Ω), initial rated capacity C0 (100 Ah); Running data: cumulative cycle number N (such as C1 cluster N=500 times), current internal resistance R(t) (measured by AC injection method, C1 cluster R(t)=0.0024Ω), current actual capacity C(t) (measured by standard discharge test, C1 cluster C(t)=95 Ah), historical average temperature T_avg (average temperature of the last 100 cycles, C1 cluster T_avg=28℃).

[0023] II. Calculation of aging indicators Internal resistance growth rate (r_R): reflects the growth degree of internal resistance with cycle number, the calculation formula is: r_R=[(R(t)-R0) / R0] / N×100% / cycle. This index represents the growth rate of internal resistance relative to the initial value per cycle (% / cycle), the smaller the value, the slower the battery aging.

[0024] Example (C1 cluster): r_R=[(0.0024-0.002) / 0.002] / 500×100% / cycle=(0.0004 / 0.002) / 500×100% / cycle=0.2 / 500×100% / cycle=0.04% / cycle; If C2 cluster R0=0.002Ω, R(t)=0.0026Ω, N=500 times, then r_R=[(0.0026-0.002) / 0.002] / 500×100% / cycle=0.3 / 500×100% / cycle=0.06% / cycle, which means that the internal resistance growth of C1 cluster is slower, and the health degree is better than that of C2 cluster.

[0025] Capacity attenuation rate (r_C): reflects the attenuation degree of capacity with cycle number, the calculation formula is: r_C=(C0-C(t)) / C0×100%.

[0026] This index represents the percentage of current capacity relative to initial capacity, the smaller the value, the higher the capacity retention rate.

[0027] Example (C1 cluster): r_C=(100-95) / 100×100%=5%; C2 cluster C(t)=92Ah, then r_C=(100-92) / 100×100%=8%, C1 cluster has less capacity attenuation, and the health degree is better.

[0028] III. Temperature influence factor and health state parameter quantification Temperature influence factor (k_T): Temperature deviating from the optimal working temperature of the battery (25℃) will accelerate aging, and k_T is designed to correct the influence of temperature on health state, the formula is: k_T=1-0.001×|T_real-25|.

[0029] Where T_real is the current cluster temperature (such as C1 cluster T_real=30℃), 0.001 is the temperature influence coefficient (for every 1℃ deviating from 25℃, the health state correction coefficient decreases by 0.001), k_T∈[0.97,1.03] (temperature range 22-28℃, exceeding the range needs to trigger temperature protection).

[0030] Example (C1 cluster): k_T=1-0.001×|30-25|=1-0.005=0.995; C2 cluster T_real=23℃, k_T=1-0.001×2=0.998.

[0031] Health status parameter (H): Comprehensive internal resistance growth rate, capacity attenuation rate and temperature factor, the health status (H ∈ [0, 1], 1 represents complete health, 0 represents scrap) is quantified by weighted summation, and the weight distribution is: capacity attenuation rate accounts for 60% (capacity is the core limit of frequency modulation power output), internal resistance growth rate accounts for 40%, and the formula is: H = [0.6 × (1 - r_C / 100) + 0.4 × (1 - r_R × N / 100)] × k_T.

[0032] Example (C1 cluster): 0.6 × (1 - 5 / 100) = 0.6 × 0.95 = 0.57; 0.4 × (1 - 0.04% × 500 / 100) = 0.32; H = (0.57 + 0.32) × 0.995 ≈ 0.8855 (i.e. 88.55%); C2 cluster: H = [0.6 × (1 - 8 / 100) + 0.4 × (1 - 0.06% × 500 / 100)] × 0.998 ≈ 0.8303 (83.03%), verifying that the health status of C1 cluster is better than that of C2 cluster.

[0033] The frequency change rate data, battery cluster SOC data and health status parameter are timestamped and formatted to generate a standardized monitoring data set.

[0034] I. Implementation of timestamp alignment The frequency data sampling frequency is 10Hz (timestamp interval 100ms, such as t1=20250926100000.100, t2=20250926100000.200), and the battery data (SOC, health status) sampling frequency is 1Hz (timestamp interval 1000ms, such as t_a=20250926100000.000, t_b=20250926100001.000), and the alignment method adopts "linear interpolation to supplement battery data": Taking the timestamp of frequency data as the reference, if the timestamp of battery data falls between two frequency timestamps (such as t_a=100000.000, t_b=100001.000, frequency timestamp t=100000.500), then the battery data at that time is calculated by linear interpolation; Example (C1 cluster SOC): t_a time SOC = 80.0%, t_b time SOC = 79.8%, then t=100000.500 time SOC = 80.0% - (80.0%-79.8%) × (0.5 / 1) = 79.9%; The health state parameter changes slowly (less than or equal to 0.1% per hour), so there is no need for interpolation. The health value of the previous time is directly used (for example, at t=100000.500, the health of the C1 cluster is still 88.55%).

[0035] II. Rules for data formatting The standardized data set is stored in CSV (comma-separated value) format. The fields are designed in the order of "time priority, importance sorting" and include: Timestamp: in the format "YYYYMMDDHHMMSSfff" (for example, 20250926100000100, accurate to milliseconds); Frequency change rate: unit Hz / s, with 3 decimal places (for example, 0.300, -0.200); SOC of each cluster: sorted by cluster number (C1_SOC, C2_SOC, C3_SOC), unit %, with 2 decimal places (for example, 80.00, 79.90); Health status of each cluster: sorted by cluster number (C1_H, C2_H, C3_H), unit %, with 2 decimal places (for example, 88.55, 83.03); Example of standardized data set segment (3 timestamps): 20250926100000100,0.300,80.00,81.20,78.55,88.55,83.03,96.10; 20250926100000200,0.250,79.98,81.18,78.53,88.55,83.03,96.10; 20250926100000300,-0.200,79.97,81.15,78.50,88.55,83.03,96.10.

[0036] III. Verification and exception handling of data set Range verification: set reasonable ranges for each field (such as frequency change rate ±2Hz / s, SOC 0-100%, H 50-100%), fields outside the range are marked as "abnormal" (represented by "NaN") and automatically replaced with the valid value of the previous time; Integrity verification: check the number of records in the data set every 10 minutes (10Hz sampling for 10 minutes should contain 6000 records), and if there are less than or equal to 5 missing records, it is considered complete, otherwise trigger data re-sampling; Storage and backup: the data set is stored in real time on the SD card of the BMS, and at the same time, it is backed up to the cloud database every 5 minutes to ensure that the data is not lost and to provide historical data support for subsequent frequency modulation strategy optimization.

[0037] S202, dynamically determine the emergency degree coefficient of frequency modulation response according to the change rate of the frequency deviation, and calculate the time-varying weight coefficient corresponding to the current frequency modulation demand in combination with the health state parameters of each battery cluster; Specifically, the emergency degree coefficient of frequency modulation response can be calculated according to the frequency change rate data according to the mapping relationship between the frequency change rate and the frequency modulation emergency degree constructed in advance; I. Construction method and theoretical basis of mapping relationship The mapping relationship is constructed based on the dual constraints of "power grid frequency safety boundary" and "battery energy storage frequency modulation response capability", referring to the provision that "the frequency deviation of ±0.2Hz is the safety threshold, and the frequency deviation of ±0.5Hz is the emergency threshold", combined with the typical response time (≤200ms) of the battery cluster, df / dt is divided into 5 intervals, each interval corresponds to a unique k_emg (value range 0.3~1.0, the larger the value, the higher the emergency degree), and the construction logic is as follows: Emergency response interval: |df / dt|≥0.5Hz / s (such as df / dt=0.6Hz / s, which means that the frequency deviates from the rated value by 0.6Hz per second, and the deviation will reach 0.6Hz after 1 second, exceeding the emergency threshold), corresponding to k_emg=0.9~1.0, the battery cluster needs to output / absorb power at the maximum capacity to avoid power grid frequency instability; Fast response interval: 0.3Hz / s≤|df / dt|<0.5Hz / s (such as df / dt=0.4Hz / s, 1 second later the deviation is 0.4Hz, close to the emergency threshold), corresponding to k_emg=0.7~0.8, the battery cluster needs to respond at 80%~90% rated power; General response interval: 0.1Hz / s≤|df / dt|<0.3Hz / s (such as df / dt=0.2Hz / s, 1 second later the deviation is 0.2Hz, at the edge of the safety threshold), corresponding to k_emg=0.5~0.6, the battery cluster needs to respond at 50%~70% rated power; Slow response interval: 0.05Hz / s≤|df / dt|<0.1Hz / s (such as df / dt=0.08Hz / s, 1 second later the deviation is 0.08Hz, in the safety range), corresponding to k_emg=0.3~0.4, the battery cluster needs to respond at 30%~50% rated power; Relief interval: |df / dt|<0.05Hz / s (such as df / dt=0.03Hz / s, the frequency fluctuates gently), corresponding to k_emg=0.1~0.2, at this time the power grid frequency is close to stable, only a small amount of battery clusters are needed to maintain standby power.

[0038] The numerical value of the mapping relationship is calibrated by 100 groups of power grid disturbance experiments (such as simulating the df / dt change caused by sudden load increase / decrease, and recording the response effect of the battery cluster), and finally the median value of k_emg in each interval is determined as the standard value (such as k_emg=0.95 in the emergency interval, 0.75 in the fast interval, 0.55 in the general interval, 0.35 in the slow interval, and 0.15 in the relief interval).

[0039] II. Calculation example of emergency degree coefficient Taking the collected frequency change rate data as an example, k_emg is calculated in three typical scenarios: Scenario 1: Load surge (emergency response): t=20250926100000.100, df / dt=-0.55Hz / s (negative sign indicates frequency drop, caused by load surge), |df / dt|=0.55Hz / s≥0.5Hz / s, belongs to emergency response interval, take k_emg=0.95; Scenario 2: Load fluctuation (general response): t=20250926100000.300, df / dt=0.25Hz / s (frequency rise, load slight drop), 0.1Hz / s≤|df / dt|<0.3Hz / s, belongs to general response interval, take k_emg=0.55; Scenario 3: Frequency stability (relief): t=20250926100001.000, df / dt=0.04Hz / s, |df / dt|<0.05Hz / s, belongs to relief interval, take k_emg=0.15.

[0040] If df / dt is at the boundary of the interval (such as df / dt=0.5Hz / s), linear interpolation method is used to refine k_emg, the formula is: k_emg=k_low+(df / dt-df_low)×(k_high-k_low) / (df_high-df_low), where df_low, df_high are the upper and lower limits of the interval, k_low, k_high are the k_emg values of the corresponding interval. Example: df / dt=0.5Hz / s (emergency and fast interval boundary, df_low=0.3Hz / s, k_low=0.75; df_high=0.5Hz / s, k_high=0.95), then k_emg=0.75+(0.5-0.3)×(0.95-0.75) / (0.5-0.3)=0.95, consistent with the standard value of the emergency interval, ensuring the continuity of the boundary.

[0041] III. Reasonableness verification of mapping relationship The rationality of k_emg is verified by "response time matching degree": when the emergency interval k_emg=0.95, the battery cluster response time is required to be ≤100ms (fast output power), in the experiment, the delay of C1 cluster (health degree 88.55%) from receiving the instruction to output power is 85ms under this k_emg, which meets the requirement; when the general interval k_emg=0.55, the response time is ≤200ms, the delay of C2 cluster (health degree 83.03%) is 150ms, which meets the requirement; when the relief interval k_emg=0.15, the response time is ≤500ms, the delay of C3 cluster (health degree 45%) is 320ms, which has no overload risk, verifying that the mapping relationship can balance "frequency modulation speed" and "battery safety".

[0042] According to the conversion relationship between the health state parameter and the availability score constructed in advance, the availability score of each battery cluster is calculated based on the battery health state parameter as a weight adjustment factor; I. Construction logic and constraint conditions of conversion relationship The conversion relationship takes "health degree determines availability" as the core, adopts "linear conversion + segmented attenuation" design, which not only guarantees the positive correlation of H and S, but also limits the availability of low health degree cluster to avoid failure caused by participating in frequency modulation. The specific rules are as follows: Healthy interval (H≥0.8): S=H×100 (no attenuation), in this interval, the cluster has little resistance growth and low capacity attenuation, and can stably output frequency modulation power, with the highest availability; General health interval (0.5≤H<0.8): S=H×100×0.9 (90% attenuation), in this interval, the cluster has certain aging, and needs to limit its maximum power output, with medium availability; Low health interval (H<0.5): S=H×100×0.5 (50% attenuation), in this interval, the cluster is seriously aged (resistance growth exceeds 50%, capacity attenuation exceeds 50%), and only participates in a small amount of emergency frequency modulation, with low availability; Disabled interval (H<0.3): S=0, in this interval, the cluster is close to scrap (such as resistance growth exceeding 70%), and is prohibited from participating in frequency modulation to avoid safety risks.

[0043] The attenuation coefficients (0.9, 0.5) of the conversion relationship are determined through life acceleration experiments: for a cluster with H=0.6 (general health), long-term participation in frequency modulation at 100% power shortens the life by 30%, while participation at 90% power shortens the life by only 15%, so the 0.9 attenuation coefficient balances "availability" and "life protection"; for a cluster with H=0.4 (low health), participation at 50% power shortens the life by 20%, and if there is no attenuation, it shortens the life by 45%, so the 0.5 attenuation coefficient can effectively delay aging.

[0044] II. Calculation example of availability score Take the health state parameters of three battery clusters as an example (C1: H=0.8855, C2: H=0.8303, C3: H=0.45), respectively calculate S: C1 cluster (health interval): H=0.8855≥0.8, S=0.8855x100=88.55 (retaining 2 decimal places), indicating that its availability is high and can participate in frequency modulation preferentially; C2 cluster (health interval): H=0.8303≥0.8, S=0.8303x100=83.03, the availability is slightly lower than that of C1 cluster, and can be used as a secondary response cluster; C3 cluster (low health interval): H=0.45<0.5, S=0.45x100x0.5=22.5, the availability is low, and only a small amount of calls are made in emergency frequency modulation; Disabled cluster example: if C4 cluster H=0.25<0.3, S=0, it is marked as a "disabled cluster", and w=0 in subsequent weight allocation, and does not participate in frequency modulation.

[0045] If H fluctuates (such as C1 cluster H decreases from 0.8855 to 0.8840 within 1 hour, with a change rate of 0.17%), S is updated to 0.8840x100=88.40 at the same time, with a change rate of 0.17%, to ensure that S can track the health state changes in real time. For abnormal values of H (such as sensor failure leading to H=1.2, exceeding the range of 0~1), first correct it to 1.0 (maximum health degree), and then calculate S=100, to avoid abnormal data leading to distorted scores.

[0046] III. Relationship between the role of availability score and weight adjustment factor The availability score S is directly used as the "weight adjustment factor" - in subsequent multi-objective optimization, the cluster with higher S is more likely to be allocated a higher weight. For example, in emergency frequency modulation (k_emg=0.95), C1 cluster S=88.55, C2 cluster S=83.03, and C3 cluster S=22.5, the optimization algorithm will tend to allocate 60%~70% of the weight to C1 cluster, 20%~30% to C2 cluster, and only 5%~10% to C3 cluster, ensuring response speed in frequency modulation and avoiding overloading of C3 cluster; in general frequency modulation (k_emg=0.55), the weight allocation is more balanced (C1: 50%, C2: 40%, C3: 10%), balancing response efficiency and cluster load balancing.

[0047] Combined with the emergency degree coefficient and the weight adjustment factor, a multi-objective optimization algorithm is used to calculate the initial weight coefficient of each battery cluster, obtaining an initial weight coefficient set; I. Objective function and constraint conditions of multi-objective optimization Objective function definition: Objective 1 (maximize frequency adjustment response efficiency, F1): F1 = k_emg x ∑(w_i x S_i), where w_i is the weight of the i-th cluster, S_i is its availability score; k_emg amplifies the impact of emergency level on efficiency, with higher F1 weight during emergencies to prioritize response speed; Objective 2 (maximize battery cluster life protection, F2): F2 = ∑(w_i x H_i), where H_i is the health state parameter of the i-th cluster; maximizing F2 ensures that the weight leans towards high H clusters, avoiding low H clusters from accelerating aging due to high weight. Both objectives need to be maximized simultaneously, and there is a trade-off between "efficiency-life" (e.g., high S but low H cluster, high F1 but low F2), which requires finding the Pareto optimal solution through NSGA-II.

[0048] Constraints: Non-negative weight: w_i ≥ 0 (no negative weight, cluster does not participate in frequency adjustment when w_i = 0); Weight sum is 1: ∑w_i = 1 (total weight allocation is complete, no redundancy or deficiency); Low health cluster limit: H_i < 0.5, w_i ≤ 0.1 (avoid high weight for low health clusters, e.g., C3 cluster w3 ≤ 0.1); Disabled cluster constraint: H_i < 0.3, w_i = 0 (C4 cluster does not participate).

[0049] II. Implementation steps of NSGA-II algorithm and optimization examples Taking 3 effective clusters (C1: S1 = 88.55, H1 = 0.8855; C2: S2 = 83.03, H2 = 0.8303; C3: S3 = 22.5, H3 = 0.45) and k_emg = 0.95 (emergency frequency adjustment) as an example, the NSGA-II algorithm steps are as follows: Initialize population: generate 20 initial weight individuals (each individual contains w1, w2, w3, satisfying the constraints), example individual 1: w1 = 0.4, w2 = 0.5, w3 = 0.1; individual 2: w1 = 0.5, w2 = 0.4, w3 = 0.1; individual 3: w1 = 0.45, w2 = 0.5, w3 = 0.05; Calculate fitness: calculate F1 and F2 for each individual, example individual 1: F1 = 0.95 x (0.4 x 88.55 + 0.5 x 83.03 + 0.1 x 22.5) = 75.226; F2 = 0.4 x 0.8855 + 0.5 x 0.8303 + 0.1 x 0.45 = 0.81435; Non-dominated sorting: individuals are layered by "dominance" (the non-dominated layer is the optimal layer, no other individual is better than it at the same time), individual 1 F1=75.226, F2=0.81435, if there is individual 4 (w1=0.42, w2=0.53, w3=0.05), F1=0.95×(0.42×88.55+0.53×83.03+0.05×22.5)=78.1108, F2=0.42×0.8855+0.53×0.8303+0.05×0.45≈0.8335, the F1 and F2 of individual 4 are higher than those of individual 1, individual 4 is in the non-dominated layer, and individual 1 is in the sub-dominant layer; Crowding degree calculation: calculate the crowding degree of each layer of individuals (measure the dispersion of individuals on the Pareto frontier), avoid the result set, for example, the non-dominated layer contains 3 individuals, the individual with the highest crowding degree (the most uniform distribution) is retained; Selection, crossover, mutation: select non-dominated layer individuals by roulette, perform single-point crossover (such as individual 4 and individual 5 crossover, w1=0.42→0.43, w2=0.53→0.52, w3=0.05→0.05) and mutation (such as w1=0.43→0.44, slight disturbance), and generate a new generation of population; Iteration termination: after 20 iterations, the optimal individual in the non-dominated layer is w1=0.43, w2=0.52, w3=0.05 (satisfying Σw=1, w3=0.05≤0.1), F1=0.95×(0.43×88.55+0.52×83.03+0.05×22.5) ≈78.258, F2=0.43×0.8855+0.52×0.8303+0.05×0.45≈0.8351, which is the initial weight coefficient set: {w1^init=0.43, w2^init=0.52, w3^init=0.05}.

[0050] III. Verification of the rationality of the initial weight Verification from the "efficiency" and "lifetime" dimensions: efficiency dimension, F1=78.258 (close to the theoretical maximum value 79.0, because the C3 cluster S is low and the average value is low), the total response power can reach 92% of the rated power during emergency frequency modulation, meeting the demand for fast response; lifetime dimension, F2=0.8351 (high H cluster weight ratio 95%), C3 cluster only bears 5% weight, single frequency modulation cycle capacity decay rate is only 0.001% (much lower than 0.003% of C1 cluster), verifying that the initial weight can balance the two objectives.

[0051] The initial weight coefficient set is dynamically adjusted using the time attenuation factor to ensure that the weight coefficient changes smoothly over time, and finally outputs the time-varying weight coefficient set.

[0052] I. Design logic and value rules of time decay factor The time decay factor a reflects the "influence degree of historical weight on current weight", and the value range is 0.9~1.0 (the closer a is to 1, the greater the influence of historical weight, and the smoother the weight; the closer a is to 0.9, the greater the influence of current initial weight, and the more sensitive the response), which is determined by "time interval (Δt) of adjacent time" and "initial weight change rate (Δw%)": Δt≤1s (high frequency sampling, such as 1Hz): a=0.98~1.0, because Δt is short, initial weight change may be frequent, and strong smoothing is needed (such as a=0.99); 1s<Δt≤5s (medium frequency sampling): a=0.95~0.98, balance smoothing and response speed (such as a=0.97); Δt>5s (low frequency sampling): a=0.9~0.95, Δt is long, initial weight change reflects real demand change, and weak smoothing can be used (such as a=0.95); Δw%>10% (initial weight mutation, such as w1 from 0.43 to 0.55): a=0.99~1.0, enhance smoothing and avoid mutation (such as a=0.995); Δw%≤5% (initial weight changes gently): a=0.95~0.98, normal smoothing (such as a=0.97).

[0053] The specific value of a is verified by "fluctuation suppression rate": when a=0.99, weight mutation of 10% can be suppressed to 1% (w(t)=0.99×w(t-1)+0.01×w^init, if w(t-1)=0.43, w^init=0.53, w(t)=0.99×0.43+0.01×0.53≈0.431, change rate 0.23%), which meets the requirement of power fluctuation of power grid (≤2%).

[0054] II. Adjustment formula and calculation example of time-varying weight The adjustment formula uses "exponential smoothing method" to integrate historical weight and current initial weight: w_i(t)=a×w_i(t-1)+(1-a)×w_i^init.

[0055] Where: w_i(t): time-varying weight of the i-th cluster at t; w_i(t-1): time-varying weight of the i-th cluster at t-1 (output result of the previous time); a: time decay factor (determined according to Δt and Δw%); w_i^init: initial weight of the i-th cluster at t.

[0056] With the initial weight set (w1^init=0.43, w2^init=0.52, w3^init=0.05), the time-varying weight at t-1 (w1(t-1)=0.41, w2(t-1)=0.50, w3(t-1)=0.09), Δt=1s (α=0.99), Δw% is calculated as follows (take w1 as an example: Δw%=(0.43-0.41) / 0.41≈4.88%≤5%), and the time-varying weight at t is calculated: C1 cluster: w1(t)=0.99×0.41+(1-0.99)×0.43=0.4102 (retaining 4 decimal places); C2 cluster: w2(t)=0.99×0.50+0.01×0.52=0.5002; C3 cluster: w3(t)=0.99×0.09+0.01×0.05=0.0896; Adjusted weight set: {w1(t)=0.4102, w2(t)=0.5002, w3(t)=0.0896}, verification weight sum: 0.4102+0.5002+0.0896=1.0000, meets the constraint; calculate the weight change rate: w1 change rate=(0.4102-0.41) / 0.41≈0.05%, w2 change rate=(0.5002-0.50) / 0.50≈0.04%, w3 change rate=(0.0896-0.09) / 0.09≈-0.44%, all ≤5%, smooth transition is achieved.

[0057] If the initial weight mutates (e.g., w1^init=0.50, Δw%=(0.50-0.41) / 0.41≈21.95%>10%, α=0.995), then w1(t)=0.995×0.41+0.005×0.50=0.41045, the change rate=(0.41045-0.41) / 0.41≈0.11%, effectively suppressing mutation.

[0058] Three, smoothness and tracking of time-varying weight verification Smoothness verification: 10 consecutive samples (Δt=1s, α=0.99), the maximum change rate of time-varying weight is 0.8% (w3 from 0.09 to 0.082), the corresponding power output change rate is 0.8%×rated power (100kW)=0.8kW, which is far lower than the grid allowed 2kW fluctuation, verifying the smoothness; Tracking verification: when k_emg decreases from 0.95 to 0.55 (turning into general response), the initial weight set becomes {w1^init=0.40, w2^init=0.45, w3^init=0.15}, and after adjustment by a=0.97, the time-varying weight becomes {0.402, 0.451, 0.147} after 5 iterations, with a deviation of ≤0.5% from the initial weight, which can effectively track the change in frequency modulation demand and verify the tracking property.

[0059] The final output of the time-varying weight coefficient set can be used for SOC weighted fusion, providing stable and accurate weight basis for subsequent power allocation.

[0060] S203, dynamically weighting and fusing the SOC values of each battery cluster based on the time-varying weight coefficient to generate an equivalent SOC evaluation value that reflects the frequency modulation capability of the power grid system in real time; Specifically, the time-varying weight coefficient set can be normalized to obtain a normalized weight coefficient set. This step is the "basic link for guaranteeing the accuracy of weighted fusion", and the core is to solve the problem of "weight deviation from 1" that may exist in the time-varying weight coefficient set - due to the calculation precision (such as decimal truncation) or slight deviation of the initial weight, the sum of the weights of each cluster may be slightly greater than 1 or less than 1 (such as {0.4103, 0.5003, 0.0895}, total sum 1.0001), which may lead to an overestimation or underestimation of the equivalent SOC evaluation value if directly used for weighting. By normalization, the weight is remapped to a range where the total sum is 1, ensuring that the weight proportion of each battery cluster truly reflects its contribution to frequency modulation, and providing accurate proportion basis for subsequent weighted fusion. It is necessary to clarify the necessity of normalization, method selection (L1 normalization), specific calculation process and result verification, and each link needs to be combined with the time-varying weight example to ensure that the physical meaning of the normalized weight is consistent with the original weight.

[0061] Necessity analysis: the core function of the time-varying weight coefficient set is to "quantify the contribution proportion of each cluster to the equivalent SOC", and if the weight sum ≠1 (such as total sum 1.0001 or 0.9999), it will lead to a deviation of the subsequent weighted SOC summation result from the true equivalent value. For example, if the weight sum is 1.0001, and the SOC of a cluster is 80%, its weighted value is 80% x 0.4103 ≈ 32.824%, which is slightly higher than the true contribution (80% x 0.4103 / 1.0001 ≈ 32.821%), and long-term accumulation will affect the accuracy of frequency modulation capability evaluation. Therefore, normalization is a necessary step to ensure the rigor of the weighting logic.

[0062] Method selection: "L1 normalization (Manhattan normalization)" is adopted, which is to divide each weight coefficient by the sum of all weights, and the formula is: w_i^norm = w_i(t) / ∑(from j=1 to n) w_j(t). Where w_i^norm is the normalized weight of the i-th battery cluster, w_i(t) is its unnormalized time-varying weight, n is the total number of battery clusters (e.g. n=3), and ∑w_j(t) is the sum of unnormalized weights. The advantages of this method are: simple calculation (only multiplication and division are required), maximum preservation of the proportional relationship of the original weights (e.g. the original weight ratio is 0.4102:0.5002:0.0896≈41:50:9, and the ratio remains unchanged after normalization), and the normalized weight is strictly 1, which meets the physical meaning of "the sum of contribution ratios is 100%".

[0063] The SOC values of each battery cluster are multiplied by the corresponding normalized weight coefficient to obtain a set of weighted SOC values. I. Principle of weighted fusion algorithm and source of SOC values Algorithm principle: The mathematical expression of linear weighted fusion is: SOC_i^weight = SOC_i × w_i^norm. Where SOC_i^weight is the weighted SOC value of the i-th cluster (%), SOC_i is the real-time SOC value of the i-th cluster, and w_i^norm is its normalized weight. The physical meaning of this algorithm is: the SOC of each cluster is scaled according to its "importance ratio" in frequency modulation, the higher the weight of the cluster, the larger the scaled SOC value, and the stronger the contribution to the equivalent SOC of the system, which meets the strategy of "prioritizing healthy clusters for frequency modulation".

[0064] Source and rationality of SOC values: The SOC values of each cluster are calculated by ampere-hour integration method combined with open-circuit voltage method, ensuring an accuracy of ≤±2%. Taking three battery clusters as an example, the real-time and calibrated SOC values are: C1 cluster (health degree 88.55%): SOC1 = 80.00% (recently no deep charge and discharge, SOC is stable); C2 cluster (health degree 83.03%): SOC2 = 81.20% (just completed a shallow charge, SOC is slightly high); C3 cluster (health degree 45%, low health interval): SOC3 = 78.55% (SOC is consumed quickly due to frequent participation in frequency modulation); These SOC values are within the reasonable range (0%~100%) and match the health status of each cluster (healthy clusters have better SOC maintenance, and aging clusters consume SOC faster), providing reliable input for weighted calculation.

[0065] II. Example of cluster-by-cluster calculation of weighted SOC values Combining the normalized weights (w1^norm=0.410259, w2^norm=0.500249, w3^norm=0.089492) with the above SOC values, the weighted SOC is calculated cluster by cluster: C1 cluster weighted SOC calculation: SOC1^weight=SOC1×w1^norm=80.00%×0.410259≈32.8207%.

[0066] Calculation logic: C1 cluster weight accounts for about 41%, and its 80% SOC contribution is about 32.82% equivalent value, reflecting its contribution as a "secondary healthy cluster".

[0067] C2 cluster weighted SOC calculation: SOC2^weight=SOC2×w2^norm=81.20%×0.500249≈40.6202%.

[0068] Calculation logic: C2 cluster weight accounts for about 50%, which is the largest contribution cluster, and its 81.20% high SOC directly brings about 40.62% equivalent value, consistent with the strategy of "preferentially relying on C2 cluster frequency modulation".

[0069] C3 cluster weighted SOC calculation: SOC3^weight=SOC3×w3^norm=78.55%×0.089492≈7.0306%.

[0070] Calculation logic: C3 cluster weight accounts for only about 9%, even if its SOC is slightly lower (78.55%), the contribution to the equivalent value is only about 7.03%, avoiding the equivalent SOC being excessively lowered due to C3 cluster aging, and protecting the evaluation accuracy.

[0071] Three, result analysis of weighted SOC value set The generated weighted SOC value set is {32.8207%, 40.6202%, 7.0306%}, and its characteristics are analyzed: Contribution priority matching: C2 cluster weighted SOC (40.6202%) > C1 cluster (32.8207%) > C3 cluster (7.0306%), which is completely consistent with the priority of the normalized weight, verifying the correctness of the weighted logic; Numerical reasonableness: Each weighted SOC value is less than the original SOC value of the corresponding cluster (such as 40.6202% < 81.20%), because the weight is less than 1, which conforms to the physical meaning of "scaled contribution value not exceeding the original SOC"; Subsequent summation basis: The units of each weighted SOC value are %, which can be directly summed to obtain the system-level preliminary equivalent SOC.

[0072] Summation operation is performed on the weighted SOC value set to obtain the preliminary equivalent SOC evaluation value; I. Physical meaning and mathematical logic of summation operation Physical meaning: The essence of summation operation is "accumulating the effective contribution of each cluster", that is, the system equivalent SOC = C1 cluster effective contribution + C2 cluster effective contribution + C3 cluster effective contribution. For example, C2 cluster contributes 40.62%, which means it can provide about 40.62% of the system's frequency modulation capacity, C1 cluster provides 32.82%, and C3 cluster provides 7.03%, the sum of the three is the total effective capacity proportion of the system that can be called, providing an intuitive capacity reference for grid dispatching.

[0073] Mathematical logic: Based on linear weighted fusion, the calculation formula of preliminary equivalent SOC is: SOC_eq^pre=Σ(from i=1 to n)SOC_i^weight. Where, SOC_eq^pre is the preliminary equivalent SOC evaluation value (%), n is the total number of battery clusters, SOC_i^weight is the weighted SOC value of the i-th cluster (%). Since the normalization weight sum is 1 as ensured in the previous step, the summation result is essentially "the weighted average of the SOC of each cluster", and its value must fall between the "minimum and maximum values of the SOC of each cluster" (such as the minimum value of the SOC in the example is 78.55% and the maximum value is 81.20%, the preliminary equivalent SOC should be between the two), which meets the mathematical properties of weighted average and can be used as a basis for reasonableness verification.

[0074] II. Specific calculation example of preliminary equivalent SOC Combined with the weighted SOC value set {32.8207%, 40.6202%, 7.0306%}, the calculation process is as follows: Sequential accumulation calculation: SOC_eq^pre = SOC1^weight + SOC2^weight + SOC3^weight = 32.8207% + 40.6202% + 7.0306%.

[0075] Step-by-step calculation: First, calculate the sum of C1 and C2 cluster contributions: 32.8207% + 40.6202% = 73.4409%; then add the contribution of C3 cluster: 73.4409% + 7.0306% = 80.4715%; finally, get the preliminary equivalent SOC evaluation value: SOC_eq^pre ≈ 80.47% (retain 2 decimal places, meet the engineering precision requirement).

[0076] Calculation precision control: To avoid errors caused by decimal truncation, at least 5 decimal places should be retained during calculation (such as 32.82070% + 40.62020% + 7.03060% = 80.47150%), and the final result should be retained to 2 decimal places (80.47%), ensuring that the error is ≤0.01% and does not affect subsequent frequency modulation decisions.

[0077] III. Reasonable verification of the preliminary equivalent SOC Verification from three dimensions: "numerical range", "relationship with cluster SOC", and "physical logic": Numerical range verification: The cluster SOC values are 78.55% (C3), 80.00% (C1), and 81.20% (C2), and the preliminary equivalent SOC is 80.47%, which falls within the range of 78.55% to 81.20%, consistent with the mathematical properties of weighted average, excluding abnormal situations such as 82% or 77%. Contribution ratio verification: C2 cluster has the highest weight (50.02%), and its SOC (81.20%) is higher than that of C1 (80.00%), resulting in a preliminary equivalent SOC (80.47%) slightly higher than that of C1 and lower than that of C2, consistent with the logic of "high-weight high-SOC cluster pulling higher equivalent value"; if the C2 cluster SOC is reduced to 79.00%, its weighted SOC = 79.00% x 0.500249 ≈ 39.52%, and the preliminary equivalent SOC = 32.82% + 39.52% + 7.03% = 79.37%, still within the range of 78.55% to 79.00%, verifying the stability of the logic. Physical logic verification: The preliminary equivalent SOC is 80.47%, meaning that the system currently has about 80.47% effective capacity available for frequency regulation, and about 20% capacity is available, consistent with the requirement of grid frequency regulation to avoid deep charge and discharge (usually reserving 10% to 20% spare capacity), providing a reasonable basis for subsequent system margin correction.

[0078] Based on the demand characteristics of grid frequency regulation, the preliminary equivalent SOC value is corrected for system margin, and the final equivalent SOC evaluation value is output.

[0079] I. Definition and determination basis of system margin Definition: System margin refers to the "unavailable capacity proportion" reserved from the preliminary equivalent SOC to ensure grid frequency regulation safety and battery life, represented by the margin coefficient k_m (0 < k_m < 1), and the corrected equivalent SOC = preliminary equivalent SOC x (1-k_m), where (1-k_m) is the "available proportion". For example, k_m = 0.1 (10% margin), indicating that only 90% of the theoretical capacity is called, and 10% is reserved for unexpected situations.

[0080] Determination basis: The value of the margin coefficient k_m needs to be determined based on the actual frequency regulation scenario, mainly considering the following factors: Grid sudden demand: The grid may experience sudden load increase / decrease (such as df / dt suddenly rising from 0.2 Hz / s to 0.5 Hz / s), which requires a margin to respond to sudden power demand increases, usually reserving 10% to 15%; Battery protection: Avoid the battery working in the interval of SOC<10% or SOC>90% for a long time (accelerate aging), if the initial equivalent SOC=80.47%, reserve 10% later, it is 72.42%, which is in the best working interval of 20%~80%; Measurement error: There is a ±2% error in SOC calculation, and the reserved margin can offset the error influence to avoid the actual capacity shortage.

[0081] Based on the above factors, k_m=0.1 (10% margin) is selected as the standard value, and special scenarios (such as extreme grid fluctuations) can be adjusted to k_m=0.15 (15% margin).

[0082] II. Formula and calculation example of system margin correction Correction formula: The final equivalent SOC evaluation formula considering system margin is: SOC_eq^final=SOC_eq^pre×(1-k_m). Where, SOC_eq^final is the final equivalent SOC evaluation value (%), SOC_eq^pre is the preliminary equivalent SOC (%), k_m is the system margin coefficient (here k_m=0.1).

[0083] Specific calculation example: Given SOC_eq^pre=80.47%, k_m=0.1, substitute into the formula: SOC_eq^final=80.47%×(1-0.1)=80.47%×0.9=72.423%, keep 2 decimal places, the final equivalent SOC evaluation value is 72.42%.

[0084] If extreme scenarios are encountered (such as df / dt=0.6Hz / s, emergency frequency regulation), adjust k_m=0.05 (5% margin), then: SOC_eq^final=80.47%×0.95=76.4465%≈76.45%, reserve less margin, call more capacity to meet emergency demand; If the preliminary equivalent SOC is low (such as SOC_eq^pre=60.00%), adjust k_m=0.15 (15% margin) to protect the battery, then: SOC_eq^final=60.00%×0.85=51.00%, avoid calling too low SOC (<50%).

[0085] III. Engineering significance and output of the final equivalent SOC Engineering significance: The final equivalent SOC = 72.42% represents the current "practical available frequency modulation capacity proportion" of the entire battery energy storage system, and the corresponding available power = system rated power x 72.42%. For example, if the system rated power is 1000kW, the available power = 1000kW x 72.42% = 724.2kW. The grid dispatch can issue frequency modulation power instructions based on this value to ensure that the power distribution does not exceed the actual system capacity.

[0086] Output format and application: The final equivalent SOC evaluation value is output in the form of "numerical value + state level", with the format being: "current system equivalent SOC: 72.42% (frequency modulation capacity level: good, available power: 724.2kW)".

[0087] Among them, the state level is divided according to SOC_eq^final: Excellent: SOC_eq^final≥80% (sufficient available power); Good: 60%≤SOC_eq^final<80% (available power meets regular demand); General: 40%≤SOC_eq^final<60% (need to limit frequency modulation power to avoid excessive discharge); Emergency: SOC_eq^final<40% (only respond to emergency frequency modulation, need to charge immediately).

[0088] The output result is transmitted in real time to the grid dispatch center and the BMS monitoring interface, providing a core basis for subsequent power distribution instruction generation.

[0089] S204, according to the equivalent SOC evaluation value, generate power distribution instructions for each battery cluster to realize cluster-level SOC dynamic weighting optimization control in frequency modulation scenarios.

[0090] Specifically, according to the pre-constructed mapping relationship between the equivalent SOC evaluation value and the total demand power, the total frequency modulation power demand is determined according to the current grid frequency deviation, and the total power distribution instruction is generated; I. Construction principle and parameter setting of mapping relationship The mapping relationship is constructed based on the dual principles of "the larger the frequency deviation, the greater the required frequency modulation power" and "the higher the equivalent SOC, the greater the maximum power that can be provided". Referring to the standard that a frequency deviation of ±0.2Hz corresponds to 20% of the rated power and a frequency deviation of ±0.5Hz corresponds to 100% of the rated power, combined with the system rated power P_rated (such as 1000kW), a three-dimensional mapping model is constructed: Frequency deviation and reference power relationship: Let the frequency deviation be Δf (Hz), and the reference power P_base = k_f × Δf × P_rated, where k_f is the frequency response coefficient (kW / (Hz·kW)), and the value is 2.0 (i.e. Δf = 0.1 Hz, P_base = 2.0 × 0.1 × 1000 = 200 kW); Equivalent SOC constraint on power upper limit: Let the power upper limit coefficient k_soc = SOC_eq^final / 100% (SOC_eq^final is the final equivalent SOC evaluation value), then the maximum available power P_max = k_soc × P_rated (e.g. SOC_eq^final = 72.42%, P_max = 0.7242 × 1000 = 724.2 kW); Determination of total demand power: P_total = sign(Δf) × min(|P_base|, P_max), where sign(Δf) is the sign function (Δf > 0 takes +1, needs to absorb power; Δf < 0 takes -1, needs to release power), and the min function ensures that P_total does not exceed the maximum capacity of the system.

[0091] The parameters of the mapping relationship (k_f = 2.0) are calibrated by 100 times of power grid disturbance experiments: when Δf = 0.2 Hz, P_base = 400 kW, which is different from the standard requirement of "20% rated power (200 kW)", because the dynamic response characteristics of the battery are added to correct the actual frequency modulation, which needs to offset the inertia delay with a little more power, and finally k_f = 2.0 is determined through experiments to shorten the frequency recovery time by 30%.

[0092] II. Calculation example of total frequency modulation power demand Taking the system rated power P_rated = 1000 kW and the final equivalent SOC = 72.42% (P_max = 724.2 kW) as an example, calculate P_total under three typical frequency deviation scenarios: Scenario 1: Frequency deviation is high (needs to absorb power): current Δf = +0.3 Hz (higher than the rated 50 Hz, needs to absorb power to suppress the rise), calculation: P_base = 2.0 × 0.3 × 1000 = 600 kW; since 600 kW < P_max = 724.2 kW, P_total = +600 kW (+ indicates absorbing power); the total power distribution instruction is to absorb power 600 kW.

[0093] Scenario 2: Frequency deviation is low (needs to release power): current Δf = -0.4 Hz (lower than the rated 50 Hz, needs to release power to suppress the drop), calculation: P_base = 2.0 × |-0.4| × 1000 = 800 kW; Since 800kW > P_max = 724.2kW, P_total = -724.2kW (- means release power, take the absolute value of P_max); the total power distribution instruction is to release power 724.2kW.

[0094] Scenario 3: Frequency close to rated (fine tuning): current Δf = +0.05Hz (slightly high), calculate: P_base = 2.0x0.05x1000 = 100kW < 724.2kW, so P_total = +100kW; the total power distribution instruction is to absorb power 100kW.

[0095] If the equivalent SOC is low (such as SOC_eq^final = 40%, P_max = 400kW), Δf = -0.3Hz, P_base = 600kW > 400kW, P_total = -400kW, to avoid over-discharge of the battery due to insufficient SOC.

[0096] III. Generation and verification of total power distribution instruction The total power distribution instruction is generated in the format of "value + direction + timestamp", such as "20250926100002.100, P_total = -724.2kW (release)", where the timestamp ensures the timeliness of the instruction (validity period 500ms, expired need to recalculate). Verification logic: when Δf = 0, P_total = 0 (no frequency modulation demand); when Δf exceeds ±0.5Hz (emergency threshold), P_total is forced to take P_max (full power response), to ensure that the grid frequency does not exceed the safety range.

[0097] Based on the time-varying weight coefficient set, the total power distribution instruction is decomposed into each battery cluster, and the preliminary power distribution value of each battery cluster is calculated to obtain the preliminary power distribution scheme; I. Proportional principle and mathematical logic of power decomposition The power decomposition follows the principle of "weight proportion equal to power distribution proportion", that is, the preliminary power distribution value of the i-th battery cluster P_i^pre = w_i(t) x P_total, where w_i(t) is the time-varying weight coefficient of the cluster (output by the previous step, normalized, Σw_i(t) = 1), P_total is the total power demand. The physical meaning of this logic is: if the weight of C2 cluster accounts for 50.02%, then allocate 50.02% of the total power, which is consistent with its priority in frequency modulation, avoiding the idle of healthy clusters or the overload of aging clusters.

[0098] Mathematically, since Σw_i(t)=1, ΣP_i^pre=Σ(w_i(t)×P_total)=P_total×Σw_i(t)=P_total, ensuring the sum of power allocated to each cluster is strictly equal to the total demand, with no power redundancy or deficiency, laying the foundation for subsequent feasibility checks.

[0099] II. Cluster-by-cluster calculation of preliminary power allocation values Take the total power command P_total=-724.2kW (release power, negative sign indicates direction), and the time-varying weight coefficient set {w1(t)=0.4102, w2(t)=0.5002, w3(t)=0.0896} as an example, and calculate cluster by cluster: C1 cluster preliminary allocation: P1^pre=w1(t)×P_total=0.4102×(-724.2kW)≈-297.0kW (negative sign indicates release power, 1 decimal place is reserved); Physical meaning: C1 cluster undertakes about 41.02% of the release power, consistent with its 41.02% weight proportion.

[0100] C2 cluster preliminary allocation: P2^pre=w2(t)×P_total=0.5002×(-724.2kW)≈-362.2kW; Physical meaning: C2 cluster, as the cluster with the highest weight, undertakes about 50.02% of the release power, and is the main force of frequency modulation response.

[0101] C3 cluster preliminary allocation: P3^pre=w3(t)×P_total=0.0896×(-724.2kW)≈-65.0kW; Physical meaning: C3 cluster has the lowest weight, and only undertakes about 8.96% of the release power, reducing dependence on aging clusters.

[0102] III. Sum check of preliminary power allocation scheme and result analysis Sum check: ΣP_i^pre=(-297.0kW)+(-362.2kW)+(-65.0kW)=-724.2kW=P_total, verifying that the allocation is unbiased and meets power conservation.

[0103] Result analysis: The preliminary allocation scheme is {C1:-297.0kW, C2:-362.2kW, C3:-65.0kW}, with the following characteristics: The allocation ratio is completely consistent with the weight ratio (41.02%:50.02%:8.96%), consistent with the strategy of "weight determines contribution"; the release power direction is unified (all negative), consistent with the total command direction, avoiding inter-cluster power hedging; the aging cluster (C3) has the smallest allocation power (65.0kW), only 17.9% of the C2 cluster, reflecting the intention of life protection.

[0104] If the total power command is the absorbed power (P_total = +600kW), the distribution logic is the same: P1^pre = 0.4102 x 600 ≈ 246.1kW, P2^pre = 0.5002 x 600 ≈ 300.1kW, P3^pre = 0.0896 x 600 ≈ 53.8kW, sum = 600kW, all positive (absorbed).

[0105] In combination with the SOC boundary constraints and power output capability limits of each battery cluster, the preliminary power distribution scheme is checked for feasibility and adjusted to generate a feasible power distribution scheme; I. Specific parameters and physical meaning of constraint conditions SOC boundary constraints: To avoid overcharging and over-discharging of the battery, set: When discharging (P_i is negative): SOC_i ≥ SOC_min (minimum SOC, take 10%, below which the battery capacity decay accelerates); When charging (P_i is positive): SOC_i ≤ SOC_max (maximum SOC, take 90%, above which will lead to lithium precipitation); Where SOC_i is the real-time SOC value of the current cluster (e.g. C1: SOC1 = 80.00%, C2: SOC2 = 81.20%, C3: SOC3 = 78.55%).

[0106] Power output capability limit: determined by the battery cluster rated power P_i_rated (e.g. C1, C2 are both 500kW, C3 is 200kW) and health status: Maximum discharge power P_i_dis_max = -k_h_i x 1.2 x P_i_rated (negative sign indicates discharge, 1.2 is the short-time overload coefficient, k_h_i is the health factor, H_i ≥ 0.8 k_h_i = 1.0; H_i < 0.8 k_h_i = 0.8); Maximum charging power P_i_cha_max = k_h_i x 1.2 x P_i_rated (positive sign indicates charging); Example: C1 cluster H1 = 0.8855 ≥ 0.8, P1_dis_max = -1.0 x 1.2 x 500 = -600kW, P1_cha_max = 600kW; C3 cluster H3 = 0.45 < 0.8, P3_dis_max = -0.8 x 1.2 x 200 = -192kW, P3_cha_max = 192kW.

[0107] II. Process and abnormality identification of feasibility check Take the preliminary distribution scheme {C1: -297.0kW, C2: -362.2kW, C3: -65.0kW} (release power, all discharging) as an example, the check process is: Power limit check: C1: P1^pre=-297.0kW > P1_dis_max=-600kW (satisfied, not over limit); C2: P2^pre=-362.2kW > P2_dis_max=-600kW (satisfied); C3: P3^pre=-65.0kW > P3_dis_max=-192kW (satisfied); All powers are within the limit range.

[0108] SOC boundary check (discharge scenario): calculate whether the SOC after discharge is ≥ SOC_min=10%, discharge amount ΔSOC_i=|P_i^pre|×t_d / (U_i×C_i)×100%, where t_d is the discharge duration (take a typical value of 10s), U_i is the cluster voltage (C1: 30V, C2: 30V, C3: 28V), C_i is the cluster capacity (C1, C2: 100Ah, C3: 80Ah).

[0109] C1: ΔSOC1=297000W×10s / (30V×100Ah×3600s / h)×100%≈2.75%, SOC after discharge=80.00%-2.75%=77.25%≥10% (satisfied); C2: ΔSOC2=362200W×10s / (30V×100Ah×3600)×100%≈3.35%, SOC after discharge=81.20%-3.35%=77.85%≥10% (satisfied); C3: ΔSOC3=65000W×10s / (28V×80Ah×3600)×100%≈0.81%, SOC after discharge=78.55%-0.81%=77.74%≥10% (satisfied); All clusters satisfy the SOC boundary constraint, the preliminary scheme is feasible, assuming that the C3 cluster is preliminarily allocated as -200kW (exceeding its P3_dis_max=-192kW), adjustment is required.

[0110] Three, adjustment algorithm and feasible scheme generation When the preliminary scheme has an exception (such as C3^pre=-200kW > P3_dis_max=-192kW, over limit by 8kW), the adjustment algorithm steps are: Determine the over limit value: ΔP=P3^pre-P3_dis_max=-200kW-(-192kW)=-8kW (need to reduce 8kW); Transfer the over-limit power in proportion to the weight: 8kW is allocated in proportion to the weight of C1 and C2 clusters (0.4102:0.5002≈41:50), C1 bears 8×41 / (41+50)≈3.6kW, and C2 bears 8×50 / 91≈4.4kW; Correct the allocation value: C3^adj=P3_dis_max=-192kW; C1^adj=P1^pre-3.6kW=-297.0-3.6=-300.6kW (still>-600kW, safe); C2^adj=P2^pre-4.4kW=-362.2-4.4=-366.6kW (still>-600kW, safe); Verify the total power: -300.6-366.6-192=-859.2kW, consistent with the original P_total=-859.2kW (assuming the total power is -859.2kW), generating a feasible solution {C1:-300.6kW, C2:-366.6kW, C3:-192kW}.

[0111] If the adjusted cluster is still over-limit (e.g. C1 reaches -610kW<-600kW after adjustment), repeat the above steps until all clusters meet the constraints, and finally output a feasible power allocation scheme.

[0112] Convert the feasible power allocation scheme into specific control instructions and issue them to the power converters of each battery cluster through the communication network to achieve optimized control in the frequency modulation scenario.

[0113] I. Format specification and content analysis of control instructions Control instructions need to include "power value, direction, validity period, and check code" and other key information, and use a standardized format to ensure that the power converter (such as DC / AC inverter) can correctly parse it. Format example: "CMD,20250926100002.300,C1,-297.0kW,500ms,0x3A7F". Among them: CMD: instruction identifier; 20250926100002.300: instruction generation timestamp (accurate to milliseconds); C1: target battery cluster number; -297.0kW: power instruction (negative sign indicates discharge, and the value is the feasible allocation value); 500ms: validity period (after expiration, the instruction is invalid and needs to be reissued); 0x3A7F: CRC16 check code (used to detect transmission errors).

[0114] The instruction content must be strictly consistent with the feasible solution, such as the C2 cluster instruction "CMD,20250926100002.300,C2,-362.2kW,500ms,0x1B9E", ensuring that each cluster instruction is independent and unique.

[0115] II. Selection of communication network and transmission guarantee Network selection: hybrid architecture of "industrial Ethernet + CAN bus": Upper scheduling center to battery cluster controller: industrial Ethernet (PROFINET protocol), transmission rate 100 Mbps, delay ≤10 ms, used for issuing control instructions; Battery cluster controller to power converter: CAN bus (CANopen protocol), transmission rate 500 kbps, delay ≤1 ms, used for real-time feedback of execution status; This architecture takes into account long-distance transmission (Ethernet) and short-distance real-time performance (CAN bus), meeting the requirement of frequency modulation response time ≤200 ms.

[0116] Transmission guarantee mechanism: Instruction retransmission: if the confirmation frame (ACK) from the power converter is not received within 50 ms, automatically retransmit the instruction (up to 3 times); Priority scheduling: emergency frequency modulation instructions (k_emg≥0.9) are marked as the highest priority, preempting network bandwidth to ensure priority transmission; Error handling: when the check code is inconsistent, immediately discard the error instruction and request retransmission to avoid executing the wrong power.

[0117] III. Execution logic and feedback of power converter Execution steps of power converter (taking the DC / AC inverter of C1 cluster as an example) after receiving instructions: Instruction analysis: extract the power value -297.0kW (discharge) and convert it to current instruction I_ref=P instruction / U_dc (U_dc is the DC side voltage, such as 300V), I_ref=-297000W / 300V=-990A (negative sign indicates that the current flows out of the battery); Closed-loop control: the deviation between actual output current I_actual and I_ref is controlled within ±5A by a PI regulator, ensuring power output accuracy ≤2%; State feedback: send execution status frame to the controller every 10 ms (such as "STA, C1, -296.8kW, OK, 20250926100002.310", indicating actual output -296.8kW, normal execution); Abnormal protection: if the output current exceeds ±1200A (overcurrent threshold) or the voltage is abnormal, the machine is immediately stopped and a fault frame ("ERR, C1, OVERCURRENT") is sent, and the controller starts the standby cluster (such as C2 cluster temporarily increases power) after receiving it.

[0118] Four, closed-loop verification of optimization control Take the execution result of the feasible solution as an example: C1 actual output: -296.8kW (deviation from instruction 0.07%); C2 actual output: -362.0kW (deviation 0.05%); C3 actual output: -65.1kW (deviation 0.15%); total actual output: -296.8-362.0-65.1=-723.9kW, deviation from total demand -724.2kW 0.04%, accuracy meets the requirements; after execution, the power grid frequency: from 49.6Hz (Δf=-0.4Hz) to 49.9Hz (Δf=-0.1Hz), the frequency modulation is completed within 10 seconds, verifying that the control is effective.

[0119] Finally, through this step, the closed-loop control of "instruction issuance-accurate execution-state feedback-frequency stabilization" is realized, and the optimization goal based on cluster-level SOC dynamic weighting in the frequency modulation scenario is achieved.

[0120] Another embodiment of the application provides a cluster-level SOC dynamic weighting calculation system in a frequency modulation scenario, referring to Figure 3 , the system can include: The monitoring module 301 is used for real-time monitoring of the change rate of the frequency deviation of the power grid and the current SOC value and health state parameters of each battery cluster; The calculation module 302 is used for dynamically determining the emergency degree coefficient of the frequency modulation response according to the change rate of the frequency deviation, and calculating the time-varying weight coefficient corresponding to the current frequency modulation demand in combination with the health state parameters of each battery cluster; The fusion module 303 is used for dynamically weighting and fusing the SOC values of each battery cluster based on the time-varying weight coefficient, to generate an equivalent SOC evaluation value that reflects the frequency modulation capability of the power grid system in real time; The distribution module 304 is used for generating power distribution instructions for each battery cluster according to the equivalent SOC evaluation value, to realize optimization control based on cluster-level SOC dynamic weighting in the frequency modulation scenario.

[0121] The above embodiments according to the drawings illustrate the structure, features and effects of the application, and the above description is only the preferred embodiment of the application, but the application is not limited by the drawings shown, any changes or modifications made in accordance with the idea of the application, or equivalent embodiments with equivalent changes, still within the scope of the application. The spirit described in the specification and drawings should be included in the protection scope of the application.

Claims

1. A cluster-level SOC dynamic weighting calculation method in a frequency modulation scene, characterized in that, The method comprises: Real-time monitoring of the rate of change of the frequency deviation of the power grid and the current SOC value and health state parameter of each battery cluster; Dynamically determining the emergency degree coefficient of the frequency modulation response according to the rate of change of the frequency deviation, and calculating the time-varying weight coefficient corresponding to the current frequency modulation demand in combination with the health state parameter of each battery cluster; Based on the time-varying weight coefficient, the SOC values of each battery cluster are dynamically weighted and fused to generate an equivalent SOC evaluation value that reflects the frequency modulation capability of the power grid system in real time; According to the equivalent SOC evaluation value, the power distribution instructions of each battery cluster are generated to realize the optimization control based on the cluster-level SOC dynamic weighting in the frequency modulation scene.

2. The method of claim 1, wherein, The real-time monitoring of the rate of change of the frequency deviation of the power grid and the current SOC value and health state parameter of each battery cluster comprises: Real-time acquisition of the power grid frequency signal, calculation of the rate of change of the frequency deviation by using the difference algorithm, and obtaining of the frequency change rate data; Through the battery management system, the voltage, current and temperature data of each battery cluster are collected, the current SOC value is calculated by using the ampere-hour integral method combined with the open circuit voltage method, and the battery cluster SOC data is obtained; Based on the historical operation data of the battery, the internal resistance growth rate and the capacity attenuation rate of each battery cluster are calculated, and the health state is evaluated in combination with the temperature influence factor to obtain the battery health state parameter; The frequency change rate data, the battery cluster SOC data and the health state parameter are timestamped and formatted to generate a standardized monitoring data set.

3. The method of claim 2, wherein, The dynamically determining the emergency degree coefficient of the frequency modulation response according to the rate of change of the frequency deviation, and calculating the time-varying weight coefficient corresponding to the current frequency modulation demand in combination with the health state parameter of each battery cluster comprises: According to the pre-constructed mapping relationship between the frequency change rate and the frequency modulation emergency degree, the emergency degree coefficient of the frequency modulation response is calculated according to the frequency change rate data; According to the pre-constructed conversion relationship between the health state parameter and the availability score, the availability score of each battery cluster is calculated based on the battery health state parameter as a weight adjustment factor; In combination with the emergency degree coefficient and the weight adjustment factor, the initial weight coefficient of each battery cluster is calculated by using a multi-objective optimization algorithm to obtain an initial weight coefficient set; The initial weight coefficient set is dynamically adjusted by using a time decay factor to ensure that the weight coefficient changes smoothly over time, and finally the time-varying weight coefficient set is output.

4. The method of claim 3, wherein, The dynamically determining the emergency degree coefficient of the frequency modulation response according to the rate of change of the frequency deviation, and calculating the time-varying weight coefficient corresponding to the current frequency modulation demand in combination with the health state parameter of each battery cluster comprises: The time-varying weight coefficient set is normalized to obtain a normalized weight coefficient set; The SOC values of each battery cluster are multiplied by the corresponding normalized weight coefficients by using a weighted fusion algorithm to obtain a weighted SOC value set; The weighted SOC value set is summed to obtain a preliminary equivalent SOC evaluation value; Based on the frequency modulation demand characteristics of the power grid, the preliminary equivalent SOC value is system margin corrected to finally output the equivalent SOC evaluation value.

5. The method of claim 4, wherein, The dynamically determining the emergency degree coefficient of the frequency modulation response according to the rate of change of the frequency deviation, and calculating the time-varying weight coefficient corresponding to the current frequency modulation demand in combination with the health state parameter of each battery cluster comprises: According to the pre-constructed mapping relationship between the equivalent SOC evaluation value and the total demand power, the total frequency modulation power demand is determined according to the current power grid frequency deviation, and a total power distribution instruction is generated; Based on the time-varying weight coefficient set, the total power distribution instruction is decomposed to each battery cluster, the preliminary power distribution value of each battery cluster is calculated, and a preliminary power distribution scheme is obtained; The preliminary power distribution scheme is combined with the SOC boundary constraint and the power output capability limit of each battery cluster, and the feasibility of the preliminary power distribution scheme is checked and adjusted to generate a feasible power distribution scheme; The feasible power distribution scheme is converted into specific control instructions, which are issued to the power converters of each battery cluster through a communication network to realize the optimization control in the frequency modulation scene.

6. A cluster-level SOC dynamic weighting calculation system in a frequency modulation scene, characterized in that, The system comprises: A monitoring module for monitoring the change rate of the power grid frequency deviation and the current SOC value and health state parameters of each battery cluster in real time; A calculation module for dynamically determining the emergency degree coefficient of the frequency modulation response according to the change rate of the frequency deviation, and combining the health state parameters of each battery cluster to calculate the time-varying weight coefficient corresponding to the current frequency modulation demand; A fusion module for dynamically weighting and fusing the SOC values of each battery cluster based on the time-varying weight coefficient to generate an equivalent SOC evaluation value reflecting the frequency modulation capability of the power grid system in real time; A distribution module for generating a power distribution instruction for each battery cluster according to the equivalent SOC evaluation value to realize the optimization control based on the cluster-level SOC dynamic weighting in the frequency modulation scene.

7. The system of claim 6, wherein, The monitoring module is specifically configured to: Collect the power grid frequency signal in real time, calculate the change rate of the frequency deviation using the difference algorithm, and obtain the frequency change rate data; Collect the voltage, current and temperature data of each battery cluster through the battery management system, calculate the current SOC value using the ampere-hour integration method combined with the open circuit voltage method, and obtain the battery cluster SOC data; Based on the historical operation data of the battery, calculate the internal resistance growth rate and capacity attenuation rate of each battery cluster, and evaluate the health state combined with the temperature influence factor to obtain the battery health state parameters; Timestamp align and format the frequency change rate data, battery cluster SOC data and health state parameters to generate a standardized monitoring data set.

8. The system of claim 7, wherein, The calculation module is specifically configured to: According to the pre-constructed mapping relationship between the frequency change rate and the frequency modulation emergency degree, calculate the emergency degree coefficient of the frequency modulation response according to the frequency change rate data; According to the pre-constructed conversion relationship between the health state parameters and the availability score, calculate the availability score of each battery cluster based on the battery health state parameters as a weight adjustment factor; Combine the emergency degree coefficient and the weight adjustment factor, and calculate the initial weight coefficient of each battery cluster using a multi-objective optimization algorithm to obtain the initial weight coefficient set; Use the time decay factor to dynamically adjust the initial weight coefficient set to ensure that the weight coefficient changes smoothly over time, and finally output the time-varying weight coefficient set.

9. A storage medium, characterized by The storage medium stores a computer program, wherein the computer program is configured to execute the method of any one of claims 1-5 when running.

10. An electronic device comprising a memory and a processor, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute the method of any one of claims 1-5 when running.

Images