A forward-looking energy storage frequency modulation power distribution method and system

Abstract

The present application relates to the technical field of energy storage battery, in particular to a forward-looking energy storage frequency modulation power distribution method and system; the method has a double-time scale structure of minute-level strategy adjustment + second-level real-time control. On the minute-level time scale, an AGC signal statistical feature extraction and prediction framework is constructed to predict the AGC signal characteristics and dynamically adjust and optimize the weight; on the second-level time scale, multi-objective optimization is performed every 2 seconds to distribute the charging and discharging power among the battery units, minimizing the AGC tracking error, battery degradation and SOC deviation. The scheme can significantly reduce the battery degradation cost while maintaining good frequency modulation performance and SOC consistency; by capturing the time statistical characteristics of the AGC signal, the potential impact of the AGC signal on the battery degradation is quantified, that is, the influence of the change trend of the AGC signal on the battery energy storage system is considered, providing forward-looking support for the control strategy of power distribution.

Description

Technical Field

[0001] This invention relates to the field of energy storage battery technology, and specifically to a forward-looking energy storage frequency regulation power distribution method and system. Background Technology

[0002] Automatic Generation Control (AGC) is the core mechanism of secondary frequency regulation in power systems. Based on signals such as Area Control Error (ACE), it issues active power regulation commands to grid-connected frequency regulation resources at fixed intervals. Battery energy storage systems, due to their excellent response speed and control capabilities, have become an important means of participating in frequency regulation ancillary services. A typical battery energy storage system consists of multiple battery cells. Upon receiving an AGC signal, the battery energy storage system needs to distribute the power regulation task among the various battery cells; this process is called power allocation.

[0003] With rising electricity demand and the widespread adoption of smart grids, AGC signal regulation of battery cells plays a crucial role in improving grid frequency stability. The second-level response capability of battery cells can instantly adjust minute frequency deviations caused by sudden load changes or fluctuations in renewable energy output. Simultaneously, AGC signal regulation can enhance renewable energy absorption capacity, effectively smoothing power fluctuations from wind and solar power, and providing reliable backup capacity for the smart grid. Compared to the frequent start-up and shutdown of thermal power units, battery energy storage offers higher regulation efficiency, lower cost, and zero greenhouse gas emissions, making it an important tool for achieving a low-carbon power grid.

[0004] Because different battery cells vary in terms of health status, usable capacity, and aging characteristics, an unreasonable power allocation strategy may lead to significant differences in State of Charge (SOC), excessive cycling of some cells, accelerated degradation, or even premature failure. Therefore, designing an efficient power allocation strategy is of significant research value, but existing power allocation technologies for battery energy storage systems lack foresight. Summary of the Invention

[0005] The main objective of this invention is to provide a forward-looking energy storage frequency regulation power allocation method and system, which aims to solve the problem that existing technical solutions for power allocation in battery energy storage systems lack foresight.

[0006] The technical solution proposed in this invention is as follows:

[0007] A forward-looking energy storage frequency regulation power allocation method is applied to a forward-looking energy storage frequency regulation power allocation system, the system including a computing module and an energy storage module; the energy storage module includes battery cells; the method includes:

[0008] The calculation module acquires historical AGC signals and performs statistical analysis on the historical AGC signals to extract features for each control period of the historical AGC signals, thereby obtaining AGC signal features;

[0009] The calculation module uses a sliding window algorithm to perform rolling predictions on a minute-level time scale to obtain the feature prediction results of future AGC signals;

[0010] The calculation module combines the feature prediction results of future AGC signals with the current state parameters of the battery cells, and evaluates the degradation cost of different battery cells based on the battery degradation neural network model.

[0011] The calculation module generates the weights of the optimized objective function for the battery cell based on the feature prediction results of the future AGC signal and the degradation cost of the battery cell.

[0012] The calculation module constructs a multi-objective optimization model on a second-level time scale, combining the current state of the battery cells with the weights of the optimization objective function, in order to allocate the charging and discharging power of each battery cell in real time.

[0013] Preferably, the AGC signal characteristics include the AGC signal in the first... The average value and frequency modulation signal within the time period in the 1st period The average absolute value within the time period; the calculation module acquires historical AGC signals, performs statistical analysis on the historical AGC signals, and extracts features for each control period of the historical AGC signals to obtain AGC signal features, including:

[0014] The computing module is used This represents the various time periods within an hour, and each time period is further divided into... The real-time transmission time of each AGC signal is used This represents the real-time transmission time of the AGC signal.

[0015] The calculation module records the acquired historical AGC signal dataset as follows:

[0016] ;

[0017] The calculation module calculates the AGC signal according to formula (1) at the 1st minute. Average value over the period Among them, the AGC signal is in the first... Average value over the period Capable of reflecting the directional shift of future AGC signals:

[0018] (1),

[0019] The calculation module calculates the AGC signal according to formula (2) at the 1st minute. The average of absolute values ​​over a period of time Among them, the AGC signal is in the first... The average of absolute values ​​over a period of time Able to represent energy intensity per unit time:

[0020] (2).

[0021] Preferably, the AGC signal characteristics further include the number of zero-crossing points of historical AGC signals and the absolute value of the segment integral of historical AGC signals; the calculation module calculates the AGC signal in the th... The average of absolute values ​​over a period of time And then it includes:

[0022] The calculation module is defined according to formula (3) in Number of zero-crossing points of historical AGC signals within the time period Among them, the number of times it crosses zero. Used to infer the number of loops:

[0023] (3),

[0024] In the formula, The sign function is used to determine the sign of the AGC signal at time t; when the value of the AGC signal at time t is positive, The value is +1; when the value of the AGC signal at time t is negative, The value is -1. When the value of the AGC signal at time t is 0, The value of is 0.

[0025] The calculation module is defined according to formula (4) in The absolute value of the segment integral of the historical AGC signal within the time period The absolute value of the segment integral. Cycle depth used to quantify battery response:

[0026] (4),

[0027] In the formula, Indicates in Within the time period The start time of each segment Indicates in Within the time period The end time of each segment, This represents the instantaneous value of the AGC signal at time t.

[0028] Preferably, the calculation module uses a sliding window algorithm for rolling prediction on a minute-level timescale to obtain feature prediction results for future AGC signals, including:

[0029] The computing module constructs and trains the first neural network model, the second neural network model, and the third neural network model, respectively.

[0030] The calculation module predicts the average value of future AGC signals over a first preset time period based on historical AGC signals and a trained first neural network model.

[0031] The calculation module predicts the average absolute value of future AGC signals over a first preset time period based on historical AGC signals and a trained first neural network model.

[0032] The calculation module predicts the number of zero crossings of future AGC signals within a first preset time period based on historical AGC signals and a trained first neural network model.

[0033] The calculation module uses the average value, the average absolute value, and the number of zero crossings within the first preset time period as the feature prediction results of the future AGC signal.

[0034] Preferably, the calculation module uses the average value, the average absolute value, and the number of zero crossings within a first preset time period as the feature prediction result of the future AGC signal, and then further includes:

[0035] The calculation module acquires historical segmented sample information and constructs an integral absolute value prediction model based on a neural network.

[0036] The calculation module trains the neural network-based integral absolute value prediction model based on historical segmented sample information.

[0037] The calculation module inputs the feature prediction results of the future AGC signal into the trained neural network-based integral absolute value prediction model to obtain the integral absolute value of each segment within the target time period.

[0038] Preferably, the calculation module combines the feature prediction results of future AGC signals with the current state parameters of the battery cells, and evaluates the degradation cost of different battery cells based on a battery degradation neural network model, including:

[0039] The computing module constructs and trains a neural network model for battery degradation.

[0040] The calculation module obtains the current state parameters of each battery cell;

[0041] The calculation module inputs the absolute values ​​of the integrals of each segment within the target time period, as well as the current state parameters of each battery cell, into the trained battery degradation neural network model to obtain the expected degradation cost of each battery cell.

[0042] Preferably, the system further includes a power divider for controlling the battery cells; the calculation module generates the optimization objective function weights for the battery cells based on the feature prediction results of future AGC signals and the degradation cost of the battery cells, including:

[0043] The calculation module generates two parameter weights for controlling the battery unit for the next second preset duration based on the feature prediction results of the future AGC signal, including:

[0044] The calculation module calculates the SOC balance weight of the battery cell based on the feature prediction results of the future AGC signal and formula (5). :

[0045] (5),

[0046] In the formula, The basic balancing weights; k is the proportionality coefficient;

[0047] The calculation module inputs the absolute values ​​of the integrals of each segment within the predicted target time period into the trained battery degradation neural network model and calculates the degradation surrogate cost weight of the battery cell using formula (6). :

[0048] (6),

[0049] In the formula, This is a proportionality coefficient used to adjust the weight of degradation costs in the objective function. Indicates the first There are 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 ... Let be the degradation function, representing the degradation amount corresponding to the absolute value of the integral in the k-th segment of the time period t for the i-th battery cell. This represents the absolute value of the integral of the k-th segment of the AGC signal within the t-th time period;

[0050] The calculation module uses the parameter weights of the battery unit used for the next second preset duration control as the updated parameter weights.

[0051] The calculation module directly transmits the updated parameter weights to the power divider;

[0052] The power divider performs real-time weighted scheduling of battery cells based on updated parameter weights.

[0053] Preferably, the calculation module constructs a multi-objective optimization model on a second-level timescale, combining the current state of the battery cells with the weights of the objective function, to allocate the charging and discharging power of each battery cell in real time, including:

[0054] The computation module constructs a multi-objective optimization model that is executed once every third preset time interval, including:

[0055] The calculation module uses the charging and discharging power commands of each battery cell in the current cycle as decision variables for a multi-objective optimization problem, and determines the decision variables based on formula (7). :

[0056] (7),

[0057] In the formula, This represents the charging power of the i-th battery cell. This represents the discharge power of the i-th battery cell. Indicates the total number of battery cells;

[0058] The computation module comprehensively considers three types of objectives to determine the objective function of the multi-objective optimization problem, as shown in formula (8):

[0059] (8),

[0060] In the formula, This represents the target power of AGC within the current optimization cycle. The weighting coefficients represent the tracking error of the AGC signal. The weighting coefficients represent the SOC consistency constraints. This represents the state of charge of the i-th battery cell. This represents the average SOC of all battery cells; This represents the degradation proxy cost weight of the i-th battery cell;

[0061] The calculation module determines the constraints of the multi-objective optimization problem, as shown in formulas (9)-(13):

[0062] (9),

[0063] (10)

[0064] (11),

[0065] (12),

[0066] (13)

[0067] In the formula, This represents the maximum charging power of the i-th battery cell. This represents the maximum discharge power of the i-th battery cell. This represents the minimum permissible state of charge of the i-th battery cell. This represents the state of charge of the i-th battery cell after the current optimization cycle. This represents the maximum permissible state of charge of the i-th battery cell. This indicates the charging efficiency of the battery cell. This represents the rated capacity of the i-th battery cell. This indicates the charge / discharge state of the i-th battery cell;

[0068] In each optimization cycle, the calculation module determines the optimal charging and discharging power of each battery cell based on the current AGC signal, the real-time status of the battery cell, and the preset weighting factors, according to the multi-objective optimization model.

[0069] The present invention also proposes a forward-looking energy storage frequency regulation power allocation system, which applies a forward-looking energy storage frequency regulation power allocation method; the system includes a computing module and an energy storage module; the energy storage module includes battery cells.

[0070] The above technical solution can achieve the following beneficial effects:

[0071] This invention proposes a forward-looking energy storage frequency regulation power allocation method with a dual-timescale structure of "minute-level strategy adjustment + second-level real-time control". At the minute-level timescale, a framework for extracting and predicting the statistical features of AGC signals is constructed to predict AGC signal characteristics and dynamically adjust optimization weights. At the second-level timescale, multi-objective optimization is performed every 2 seconds to allocate charging and discharging power among battery cells, minimizing AGC tracking error, battery degradation, and SOC deviation. The proposed technical solution can significantly reduce battery degradation costs while maintaining good frequency regulation performance and SOC consistency. Furthermore, this solution quantifies the potential impact of AGC signals on battery degradation by capturing the temporal statistical characteristics of AGC signals, thus considering the impact of AGC signal variation trends on the battery energy storage system and providing forward-looking information support for the power allocation control strategy. Attached Figure Description

[0072] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0073] Figure 1 This is a flowchart illustrating the first embodiment of a forward-looking energy storage frequency regulation power allocation method proposed in this invention. Detailed Implementation

[0074] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0075] This invention proposes a forward-looking energy storage frequency regulation power allocation method and system.

[0076] As attached Figure 1 As shown, in the first embodiment of the forward-looking energy storage frequency regulation power allocation method proposed in this invention, the method is applied to a forward-looking energy storage frequency regulation power allocation system, the system including a computing module and an energy storage module; the energy storage module includes battery cells; this embodiment includes the following steps:

[0077] Step S110: The calculation module acquires historical AGC signals and performs statistical analysis on the historical AGC signals to extract features for each control period of the historical AGC signals, thereby obtaining AGC signal features.

[0078] Step S120: The calculation module uses a sliding window algorithm to perform rolling prediction on a minute-level time scale to obtain the feature prediction results of future AGC signals.

[0079] Step S130: The calculation module combines the feature prediction results of the future AGC signal with the current state parameters of the battery cell, and evaluates the degradation cost of different battery cells based on the battery degradation neural network model.

[0080] Step S140: The calculation module generates the weights of the optimized objective function for the battery cell based on the feature prediction results of the future AGC signal and the degradation cost of the battery cell.

[0081] Step S150: The calculation module constructs a multi-objective optimization model on a second-level time scale, combining the current state of the battery cell with the weights of the optimization objective function, in order to allocate the charging and discharging power of each battery cell in real time.

[0082] This invention proposes a forward-looking frequency regulation power allocation method for energy storage, featuring a dual-timescale structure of "minute-level strategy adjustment + second-level real-time control." At the minute-level timescale, a framework for extracting and predicting the statistical features of AGC signals is constructed to predict AGC signal characteristics and dynamically adjust optimization weights. At the second-level timescale, multi-objective optimization is performed every 2 seconds to allocate charging and discharging power among battery cells, minimizing AGC tracking error, battery degradation, and SOC deviation. The proposed technical solution can significantly reduce battery degradation costs while maintaining good frequency regulation performance and SOC consistency. Furthermore, by capturing the temporal statistical characteristics of AGC signals, this solution quantifies the potential impact of AGC signals on battery degradation, considering the influence of AGC signal variation trends on the battery energy storage system. This provides forward-looking information support for the power allocation control strategy, thereby addressing the lack of foresight in existing power allocation techniques for battery energy storage systems.

[0083] In a second embodiment of a forward-looking energy storage frequency regulation power allocation method proposed in this invention, based on the first embodiment, the AGC signal characteristics include the AGC signal in the first... The average value and frequency modulation signal within the time period in the 1st period The average of absolute values ​​over the period; step S110 includes the following steps:

[0084] Step S210: The calculation module uses This represents the various time periods within an hour, and each time period is further divided into... The real-time transmission time of each AGC signal is used This represents the moment when the AGC signal is sent in real time.

[0085] Step S220: The calculation module records the acquired historical AGC signal dataset as follows:

[0086] ;

[0087] Step S230: The calculation module calculates the AGC signal according to formula (1) at the first... Average value over the period Among them, the AGC signal is in the first... Average value over the period Capable of reflecting the directional shift of future AGC signals:

[0088] (1),

[0089] Specifically, the AGC signal here is in the... Average value over the period It can reflect the directional shift of future AGC signals, therefore It has a significant impact on SOC (State of Charge, remaining battery power).

[0090] Step S240: The calculation module calculates the AGC signal according to formula (2) at the first... The average of absolute values ​​over a period of time Among them, the AGC signal is in the first... The average of absolute values ​​over a period of time Able to represent energy intensity per unit time:

[0091] (2).

[0092] Specifically, the AGC signal here is in the... The average of absolute values ​​over a period of time It can represent the energy intensity per unit time, therefore It indirectly affects the cycle depth.

[0093] In a third embodiment of the forward-looking energy storage frequency regulation power allocation method proposed in this invention, based on the second embodiment, the AGC signal characteristics further include the number of zero-crossing points of the historical AGC signal and the absolute value of the segment integral of the historical AGC signal; step S240, followed by the following steps:

[0094] Step S310: The calculation module is defined according to formula (3) in Number of zero-crossing points of historical AGC signals within the time period Among them, the number of times it crosses zero. Used to infer the number of loops:

[0095] (3),

[0096] In the formula, The sign function is used to determine the sign of the AGC signal at time t; when the value of the AGC signal at time t is positive, The value is +1; when the value of the AGC signal at time t is negative, The value is -1. When the value of the AGC signal at time t is 0, The value of is 0.

[0097] Step S320: The calculation module is defined according to formula (4) in The absolute value of the segment integral of the historical AGC signal within the time period The absolute value of the segment integral. Cycle depth used to quantify battery response:

[0098] (4),

[0099] In the formula, Indicates in Within the time period The start time of each segment Indicates in Within the time period The end time of each segment, This represents the instantaneous value of the AGC signal at time t.

[0100] In the fourth embodiment of the forward-looking energy storage frequency regulation power allocation method proposed in this invention, based on the third embodiment, step S120 includes the following steps:

[0101] Step S410: The computing module constructs and trains the first neural network model, the second neural network model, and the third neural network model respectively.

[0102] Step S420: The calculation module predicts the average value of future AGC signals over a first preset time period (e.g., 5 minutes) based on historical AGC signals and the first neural network model that has been trained.

[0103] Step S430: The calculation module predicts the average absolute value of the future AGC signal within a first preset time period based on the historical AGC signal and the first neural network model that has been trained.

[0104] Step S440: The calculation module predicts the number of zero-crossing points of the future AGC signal within a first preset time period based on the historical AGC signal and the first neural network model that has been trained.

[0105] Step S450: The calculation module uses the average value, the average absolute value, and the number of zero crossings within the first preset time period as the feature prediction result of the future AGC signal.

[0106] In the fifth embodiment of the forward-looking energy storage frequency regulation power allocation method proposed in this invention, based on the fourth embodiment, after step S450, the following steps are further included:

[0107] Step S510: The calculation module obtains historical segmented sample information and constructs an integral absolute value prediction model based on a neural network.

[0108] Step S520: The calculation module trains the neural network-based integral absolute value prediction model based on historical segmented sample information.

[0109] Step S530: The calculation module inputs the feature prediction results of the future AGC signal into the trained neural network-based integral absolute value prediction model to obtain the integral absolute value (i.e., the cycle depth) of each segment within the target time period.

[0110] In the sixth embodiment of the forward-looking energy storage frequency regulation power allocation method proposed in this invention, based on the fifth embodiment, step S130 includes the following steps:

[0111] Step S610: The computing module constructs and trains a battery degradation neural network model.

[0112] Step S620: The calculation module obtains the current state parameters of each battery cell, including the current SOC, SOH (State of Health, which reflects the difference between the current performance and the nominal performance of the battery, usually expressed as a percentage, reflecting the degree of battery aging), and temperature of each battery cell at the current moment.

[0113] Step S630: The calculation module inputs the absolute value of the integral of each segment within the target time period and the current state parameters of each battery cell into the trained battery degradation neural network model to obtain the expected degradation cost of each battery cell.

[0114] In a seventh embodiment of a forward-looking energy storage frequency regulation power distribution method proposed in this invention, based on the sixth embodiment, the system further includes a power divider for controlling the battery cells; step S140 includes the following steps:

[0115] Step S710: Based on the feature prediction results of future AGC signals, the calculation module generates two parameter weights for controlling the battery cell for the next second preset duration (e.g., 1 minute), specifically including the following steps:

[0116] Step S711: The calculation module calculates the SOC balance weight of the battery cell based on the feature prediction results of the future AGC signal and formula (5). :

[0117] (5),

[0118] In the formula, The basic balancing weights are defined by k, which is the proportionality coefficient. When the average value in the feature prediction results deviates significantly from zero, it indicates that the AGC signal has a clear directional trend, which may lead to greater differences in the SOC of each battery cell. To address this situation, This will be increased to enhance SOC consistency.

[0119] Step S712: The calculation module inputs the absolute values ​​of the integrals of each segment within the predicted target time period into the trained battery degradation neural network model and calculates the degradation proxy cost weight of the battery cell using formula (6). :

[0120] (6),

[0121] In the formula, This is a proportionality coefficient used to adjust the weight of degradation costs in the objective function. Indicates the first There are 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 ... Let be the degradation function, representing the degradation amount corresponding to the absolute value of the integral in the k-th segment of the time period t for the i-th battery cell. This represents the absolute value of the integral of the k-th segment of the AGC signal within the t-th time period;

[0122] Specifically, when the predicted degradation cost is high, the corresponding The value will also increase, thus imposing a stronger penalty weight on battery cells that are likely to experience significant degradation.

[0123] Step S713: The calculation module uses the parameter weights of the battery cell used for the next second preset duration control as the updated parameter weights.

[0124] Step S720: The calculation module directly passes the updated parameter weights to the power divider of the second layer.

[0125] Step S730: The power divider performs real-time weighted scheduling of the battery cells based on the updated parameter weights.

[0126] Specifically, the weighted scheduling process comprehensively considers AGC signal tracking error, SOC balance weight, and degradation agent cost weight, thereby achieving coordinated multi-objective optimization under forward-looking guidance.

[0127] In the eighth embodiment of the forward-looking energy storage frequency regulation power allocation method proposed in this invention, based on the seventh embodiment, step S150 includes the following steps:

[0128] Step S810: The calculation module constructs a multi-objective optimization model that is executed once every third preset time interval (e.g., 2 seconds), specifically including the following steps:

[0129] Step S811: The calculation module uses the charging and discharging power commands of each battery cell in the current cycle as decision variables for a multi-objective optimization problem, and determines the decision variables based on formula (7). :

[0130] (7),

[0131] In the formula, This represents the charging power of the i-th battery cell. This represents the discharge power of the i-th battery cell. Indicates the total number of battery cells;

[0132] Step S812: The calculation module comprehensively considers three types of objectives to determine the objective function of the multi-objective optimization problem, as shown in formula (8):

[0133] (8),

[0134] In the formula, This represents the AGC target power (determined by the power grid) within the current optimization cycle, and the AGC target power issued by the system (power grid). The weighting coefficients represent the tracking error of the AGC signal. The weighting coefficients represent the SOC consistency constraints. This represents the state of charge of the i-th battery cell. This represents the average SOC of all battery cells; This represents the degradation proxy cost weight of the i-th battery cell.

[0135] Step S813: The calculation module determines the constraints of the multi-objective optimization problem, as shown in formulas (9)-(13):

[0136] (9),

[0137] (10)

[0138] (11),

[0139] (12),

[0140] (13)

[0141] In the formula, This represents the maximum charging power of the i-th battery cell. This represents the maximum discharge power of the i-th battery cell. This represents the minimum permissible state of charge of the i-th battery cell. This represents the state of charge of the i-th battery cell after the current optimization cycle. This represents the maximum permissible state of charge of the i-th battery cell. This indicates the charging efficiency of the battery cell. This represents the rated capacity of the i-th battery cell. This indicates the charge / discharge state of the i-th battery cell;

[0142] Specifically, the aforementioned constraints include battery power limits, SOC boundaries, and energy balance, ensuring the feasibility and operational safety of the allocated strategies.

[0143] Step S820: In each optimization cycle, the calculation module determines the optimal charging and discharging power of each battery cell based on the current AGC signal, the real-time status of the battery cell, and the preset weighting factor, according to the multi-objective optimization model.

[0144] Specifically, the aforementioned weighting factors include the weighting coefficient for AGC signal tracking error. Weighting coefficients of SOC consistency constraints Degradation agency cost weight .

[0145] The present invention also proposes a forward-looking energy storage frequency regulation power allocation system, which applies a forward-looking energy storage frequency regulation power allocation method; the system includes a computing module and an energy storage module; the energy storage module includes battery cells.

[0146] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0147] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A forward looking energy storage frequency modulation power distribution method, characterized by, The application is applied to a forward-looking energy storage frequency modulation power distribution system, the system comprises a calculation module and an energy storage module; the energy storage module comprises a battery unit; the method comprises: The calculation module obtains historical AGC signals, statistically analyzes the historical AGC signals, extracts features of each control period of the historical AGC signals, and obtains AGC signal features; The calculation module uses a sliding window algorithm to perform rolling prediction on a minute-level time scale to obtain a feature prediction result of a future AGC signal; The calculation module combines the feature prediction result of the future AGC signal with current state parameters of the battery unit, evaluates the degradation cost of different battery units based on a battery degradation neural network model, and evaluates the degradation cost of different battery units based on the battery degradation neural network model; The calculation module generates an optimization objective function weight of the battery unit based on the feature prediction result of the future AGC signal and the degradation cost of the battery unit; The calculation module combines the current state of the battery unit with the optimization objective function weight on a second-level time scale to construct a multi-objective optimization model to real-time allocate the charge and discharge power of each battery unit; The calculation module takes the charge and discharge power instructions of each battery unit in the current period as decision variables of a multi-objective optimization problem, and determines the decision variables based on formula (7) : （7）， wherein represents the charging power of the i-th battery cell, represents the discharging power of the i-th battery cell, represents the total number of battery cells; The calculation module comprehensively considers three types of targets to determine the objective function of the multi-objective optimization problem, as shown in formula (8): （8）， wherein, denotes the AGC target power in the current optimization period, denotes the weight coefficient of the AGC signal tracking error, denotes the weight coefficient of the SOC consistency constraint, denotes the state of charge of the i-th battery cell, denotes the average value of all battery cell SOC; denotes the degradation proxy cost weight of the i-th battery cell.

2. The method of claim 1, wherein the method is a forward looking energy storage frequency modulation power distribution method. The AGC signal features include an average value of the AGC signal in the first time period and an average value of absolute values of the frequency modulation signal in the first time period. The AGC signal features include an average value of the AGC signal in the first time period and an average value of absolute values of the frequency modulation signal in the first time period. The calculation module obtains historical AGC signals, statistically analyzes the historical AGC signals, extracts features of each control time period of the historical AGC signals, and obtains AGC signal features, including: The computing module uses represent a respective time period in an hour, while each time period is further divided into represent the time when the real-time AGC signal is issued; and represent the time when the real-time AGC signal is sent. The calculation module records the obtained data set of the historical AGC signals as: ； The calculation module calculates the average value of the AGC signal in the first time period according to formula (1) AGC signal in the first time period AGC signal in the first time period AGC signal in the first time period The average value of the AGC signal in the first time period （1）， The calculation module calculates the average value of the absolute value of the AGC signal in the first time period according to formula (2) wherein the average value of the absolute value of the AGC signal in the first time period is wherein the average value of the absolute value of the AGC signal in the first time period is wherein the average value of the absolute value of the AGC signal in the first time period is which can represent the energy intensity in a unit of time: （2）。 3. A look-ahead energy-stored frequency-modulated power distribution method according to claim 2, wherein, The AGC signal features further include the number of zero-crossing points of the historical AGC signal and the absolute value of the segment integral of the historical AGC signal; the calculation module calculates the average value of the absolute value of the AGC signal in the period of time according to formula (2) , and then further includes: , and then further includes: The calculation module defines the number of zero-crossings of the historical AGC signal within the time period according to formula (3) wherein the number of zero-crossings wherein the number of zero-crossings for inferring the number of cycles: （3）， In the formula, represents a symbol function, which is used to determine the positive or negative of the value of the AGC signal at time t; when the value of the AGC signal at time t is positive, the value of the AGC signal at time t is negative, the value of the AGC signal at time t is negative, 1. when the value of the AGC signal at time t is 0, the value of the AGC signal at time t is negative, The calculation module defines the absolute value of the segment integral of the historical AGC signal in the time period according to formula (4) The absolute value of the segment integral of the historical AGC signal in the time period The absolute value of the segment integral of the historical AGC signal in the time period The cycle depth for quantifying the battery response: （4）， wherein denotes the start time of the segment within the time period, denotes the start time of the segment within the time period, denotes the end time of the segment within the time period, denotes the instantaneous value of the AGC signal at time t.

4. The method of claim 3, wherein the method further comprises: The calculation module uses a sliding window algorithm to perform rolling prediction on a minute-level time scale to obtain a feature prediction result of a future AGC signal, including: The calculation module respectively constructs and trains a first neural network model, a second neural network model and a third neural network model; The calculation module predicts the average value of the future AGC signal within a future first preset time length based on the historical AGC signal and the trained first neural network model; The calculation module predicts the average value of the absolute value of the future AGC signal within a future first preset time length based on the historical AGC signal and the trained first neural network model; The calculation module predicts the number of zero-crossing points of the future AGC signal within a future first preset time length based on the historical AGC signal and the trained first neural network model; The calculation module takes the average value, the average value of the absolute value, and the number of zero-crossing points within the future first preset time length as the feature prediction result of the future AGC signal.

5. A look-ahead energy-stored frequency-modulated power distribution method according to claim 4, wherein, The calculation module takes the average value, the average value of the absolute value, and the number of zero-crossing points within the future first preset time length as the feature prediction result of the future AGC signal, and then further comprises: The calculation module obtains historical segmented sample information and constructs a neural network-based integral absolute value prediction model; The calculation module trains the neural network-based integral absolute value prediction model based on the historical segmented sample information; The calculation module inputs the feature prediction result of the future AGC signal into the trained neural network-based integral absolute value prediction model to obtain the output of the integral absolute value of each segment within the target period.

6. A look-ahead energy-stored frequency-modulated power distribution method according to claim 5, wherein, The calculation module combines the feature prediction result of the future AGC signal with current state parameters of the battery unit, evaluates the degradation cost of different battery units based on a battery degradation neural network model, and evaluates the degradation cost of different battery units based on the battery degradation neural network model; The calculation module constructs and trains a battery degradation neural network model; The calculation module obtains the current state parameters of each battery unit; The calculation module inputs the integral absolute values of each segment in the target period and the current state parameters of each battery unit into the trained battery degradation neural network model to obtain the output expected degradation cost of each battery unit.

7. A look-ahead energy-stored frequency-modulated power distribution method according to claim 6, wherein, The system further comprises a power distributor for controlling the battery units; the calculation module generates the optimization objective function weight of the battery units based on the feature prediction results of the future AGC signals and the degradation cost of the battery units, including: The calculation module generates two parameter weights for controlling the battery units in the next second preset time length based on the feature prediction results of the future AGC signals, including: The calculation module calculates the SOC balancing weight of the battery unit based on the feature prediction result of the future AGC signal and formula (5) : （5）， In the formula, is the base balance weight; k is the proportional coefficient; The calculation module inputs the integral absolute values of each segment in the predicted target period into the trained battery degradation neural network model and calculates the degradation proxy cost weight of the battery unit through formula (6) : （6）， In the formula, This is a proportionality coefficient used to adjust the weight of degradation costs in the objective function. Indicates the first There are 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 ... Let be the degradation function, representing the degradation amount corresponding to the absolute value of the integral in the k-th segment of the time period t for the i-th battery cell. This represents the absolute value of the integral of the k-th segment of the AGC signal within the t-th time period; The calculation module takes the parameter weights for controlling the battery units in the next second preset time length as the updated parameter weights; The calculation module directly transmits the updated parameter weights to the power distributor; The power distributor performs real-time weighted scheduling on the battery units based on the updated parameter weights.

8. The method of claim 7, wherein the method is a forward looking energy storage frequency modulation power distribution method. The calculation module combines the current state of the battery units and the optimization objective function weight on a second time scale to construct a multi-objective optimization model for real-time allocation of the charge and discharge power of each battery unit, including: The calculation module constructs a multi-objective optimization model that is executed once every third preset time length, including: The calculation module determines the constraint conditions of the multi-objective optimization problem, as shown in formulas (9)-(13): （9）， （10）， （11）， （12）， （13）， wherein denotes the maximum charge power of the i-th battery cell, denotes the maximum discharge power of the i-th battery cell, denotes the minimum allowed state of charge of the i-th battery cell, denotes the state of charge of the i-th battery cell after the update of the current optimization period, denotes the maximum allowed state of charge of the i-th battery cell, denotes the charge efficiency of the battery cell, denotes the rated capacity of the i-th battery cell, denotes the state of charge and discharge of the i-th battery cell; In each optimization period, the calculation module determines the optimal charge and discharge power of each battery unit based on the current AGC signal, the real-time state of the battery units, and the preset weight factor according to the multi-objective optimization model.

9. A forward looking energy storage frequency modulation power distribution system, characterized by, The application of the prospective energy storage frequency modulation power distribution method according to any one of claims 1-8; the system comprises a calculation module and an energy storage module; the energy storage module comprises battery units.

Images