Energy storage load shedding control method and device based on SOC estimation and mode self-adaption

Abstract

The invention provides an energy storage load shedding control method and device based on SOC estimation and mode self-adaption. The method comprises the steps of fault detection and demand evaluation, wherein frequency deviation is collected in real time, adjustment is triggered when the frequency deviation exceeds a threshold value, and power vacancy is calculated; sOC estimation and resource screening: predicting SOC based on a long short-term memory network, input voltage, current, temperature and the like, and screening an adjustable energy storage set; the control mode is self-adaptive, wherein tripping or power adjustment is selected according to the frequency deviation and the SOC; optimization and power distribution: taking power vacancy as a target, constructing a'minimum frequency deviation + minimum control cost 'multi-target optimization model, and solving an adjustment power or tripping instruction; and instruction execution and closed loop: issuing an instruction to execute adjustment or tripping to form a closed loop. According to the invention, the resource allocation accuracy is improved through SOC estimation; dual-mode collaborative dynamic optimization is carried out, and rigidity and flexibility are fused; the multi-objective optimization overall planning frequency stability and economical efficiency is better in response speed, resource utilization and reliability.

Description

Technical Field

[0001] This invention relates to the field of ultra-high voltage direct current transmission technology, specifically a method and device for energy storage load shedding control based on SOC estimation and mode adaptation. Background Technology

[0002] With the development of ultra-high voltage direct current (UHVDC) transmission technology, it has been widely used in the optimal allocation of energy resources between eastern and western my country. However, once a fault occurs in an UHVDC transmission system (such as DC channel blockage), it will cause a massive loss of power instantaneously, leading to a rapid drop in grid frequency, and in severe cases, even frequency collapse and system disconnection, posing a huge threat to grid security. To cope with such sudden large-scale power shortage events, traditional power grids mainly rely on low-frequency load shedding systems to rebalance system output and load by tripping some loads. Although this method has a fast response, the control measures are rigid, have low precision, and are difficult to balance the continuity of power supply to users with the goal of economical system operation.

[0003] In recent years, with the rapid development of electrochemical energy storage technology, energy storage power stations have become a new key resource for emergency frequency control of power grids due to their millisecond-level response capability, rapid charging and discharging capability, and flexible dispatch characteristics. In scenarios of sudden large power shortages, energy storage systems can rapidly release power to participate in frequency support, preventing further frequency decline and improving system stability and control capabilities. Meanwhile, some research and application practices have shown that incorporating energy storage into load shedding systems can effectively improve the flexibility and precision of control strategies, mitigating the impact of traditional rigid load shedding methods on users.

[0004] However, effective regulation of energy storage systems depends not only on their hardware capabilities but also on precise understanding of their operating status, especially the battery's state of charge (SOC), a key parameter for assessing its regulatory capacity. Accurate and real-time SOC estimation is crucial for developing reasonable control strategies. Furthermore, flexibly selecting appropriate control methods (such as tripping or power regulation) based on the degree of frequency drop and the energy storage status is also an important direction for improving control efficiency and resource utilization. Therefore, the effectiveness of energy storage in emergency control depends not only on its deployment and response speed but also on the scientific nature of the control strategy, the accuracy of state perception, and the rationality of mode selection.

[0005] Currently, some technical solutions have attempted to introduce energy storage systems to participate in grid load regulation, which can be mainly divided into the following two categories:

[0006] Centralized control scheme: The dispatch center or EMS monitors the system frequency uniformly. When the frequency drops and triggers a threshold, a fixed power command is sent to the energy storage station via the communication system to control its discharge and participate in regulation. This method is relatively simple in dispatch logic, but it has the following problems:

[0007] 1. The control mode is simple, usually only using power regulation, and lacks a combined strategy with tripping action;

[0008] 2. Failure to fully consider the differences in individual operating states of energy storage units (such as SOC, voltage, current, etc.) results in a "blindly calling upon" phenomenon, which can easily lead to resource waste or regulation failure;

[0009] 3. It lacks a dynamic and intelligent mode selection mechanism, making it unable to flexibly determine different response methods based on the actual adjustment needs of the system;

[0010] 4. Control decisions rely entirely on the master station, which poses risks of communication bottlenecks, scheduling delays, and reliability issues.

[0011] Preset threshold triggering scheme: Some distributed control schemes use preset frequency thresholds and power curves. When the frequency drops to a certain set point, the energy storage automatically adjusts the power according to the issued strategy. However, this method usually has a fixed response mode and cannot adaptively switch the tripping and adjustment modes according to the energy storage status and system requirements, thus limiting its control capability.

[0012] In summary, existing energy storage solutions for load shedding control suffer from problems such as fixed control modes, inaccurate resource scheduling, and a lack of SOC status recognition and control linkage mechanisms. These shortcomings fail to meet the dual requirements of "rapid response + precise resource allocation" at the UHV receiving end under extreme operating conditions. Therefore, there is an urgent need to develop an advanced control strategy that integrates status recognition, mode adaptation, and flexible scheduling capabilities to enable energy storage systems to participate efficiently and reliably in high-power shortage scenarios. Summary of the Invention

[0013] This invention provides a method and device for energy storage load shedding control based on SOC estimation and mode adaptation. It adopts a state perception and mode adaptation load shedding control strategy for UHV receiving-end scenarios, which can effectively cope with the large-scale power deficit problem caused by DC channel faults.

[0014] A method for energy storage load shedding control based on SOC estimation and mode adaptation includes the following steps:

[0015] (1) Fault detection and regulation requirement assessment: Real-time acquisition of power grid frequency and calculation of frequency deviation Δf(t). When Δf(t) exceeds the warning threshold Δf th At that time, the energy storage regulation mechanism is triggered; simultaneously, based on the grid inertia constant H and the system capacity S sys Calculate the regulation power deficit This serves as the target benchmark for subsequent control strategies.

[0016] (2) Energy storage state estimation and resource selection based on neural network: Construct a long short-term memory network SOC estimation model, using the voltage V of the energy storage unit as the basis for the estimation.i (t), Current I i (t), temperature T i (t) and historical time series data are used as inputs, and the real-time SOC prediction value of each energy storage unit is output; based on the adjustable minimum SOC threshold SOC min and adjustable maximum SOC threshold SOC max Select adjustable energy storage sets This serves as a boundary condition for resource scheduling.

[0017] (3) Adaptive selection of control mode: The control mode variable u is determined based on the frequency deviation Δf(t) in step (1) and the SOC prediction value in step (2). i When the frequency drops sharply and the response time requirement is extremely high, trip control is selected; when the frequency change is mild and the energy storage SOC is sufficient, power regulation mode is selected.

[0018] (4) Optimization model solution and power allocation: Adjusting the power deficit in step (1) For the objective, the adjustable energy storage set in step (2) is the resource constraint, and the control mode variable u in step (3) is... i Using these as decision variables, a multi-objective optimization model is constructed with the objectives of "minimum frequency deviation + minimum control cost" to solve for the regulation power ΔP of each energy storage unit. i Or trip command;

[0019] (5) Control command execution and response closed loop: Based on the optimization results of step (4), commands are issued to the units in the adjustable energy storage set: if u i If u = 1, then power adjustment ΔPi is executed. i If the value is 0, a tripping operation is performed, forming a closed loop of "state awareness-decision-execution".

[0020] Furthermore, in step (1), based on the grid inertia constant H and the system capacity S... sys Calculate the regulation power deficit ,include:

[0021] ;

[0022] Where H represents the power grid inertia constant, S sys This refers to the system capacity.

[0023] Furthermore, in step (2), the input vector of the Long Short-Term Memory Network SOC estimation model... It contains time-series data with at least three dimensions, namely voltage, current, and temperature, and is trained using the mean squared error (MSE) loss function to ensure that the state of charge (SOC) estimation error is ≤5%; the adjustable energy storage set only contains data that meets the SOC requirements.min ≤SOC_i(t)≤SOC max The energy storage unit, in which SOC min The value range is 20%-30%, SOC max The value range is 85%-95%.

[0024] Furthermore, the objective function of the multi-objective optimization model described in step (4) is as follows:

[0025] ;

[0026] in:

[0027] λ1, λ2: Weighting coefficients

[0028] Tripping cost of the i-th device

[0029] Unit power regulation cost

[0030] ΔP i : Adjusting power variables.

[0031] Furthermore, the multi-objective optimization model described in step (4) includes the following constraints:

[0032] (1) The total power meets the requirements:

[0033] (9);

[0034] Where a i For the controllability of energy storage unit i;

[0035] (2) Adjusting power boundary constraints:

[0036] (10);

[0037] (3) Trip power limit:

[0038] (11);

[0039] (4) Control mode constraints:

[0040] (12).

[0041] An energy storage load shedding control device based on SOC estimation and mode adaptation includes:

[0042] The fault detection and regulation demand assessment module is used to collect the power grid frequency in real time and calculate the frequency deviation Δf(t). When Δf(t) exceeds the warning threshold Δf thAt that time, the energy storage regulation mechanism is triggered; simultaneously, based on the grid inertia constant H and the system capacity S sys Calculate the regulation power deficit This serves as the target benchmark for subsequent control strategies.

[0043] A neural network-based energy storage state estimation and resource selection module is used to construct a long short-term memory network SOC estimation model, using the voltage V of the energy storage unit as the basis for the estimation. i (t), Current I i (t), temperature T i (t) and historical time series data are used as inputs, and the real-time SOC prediction value of each energy storage unit is output; based on the adjustable minimum SOC threshold SOC min and adjustable maximum SOC threshold SOC max Select adjustable energy storage sets This serves as a boundary condition for resource scheduling.

[0044] The adaptive control mode selection module is used to determine the control mode variable u based on the frequency deviation Δf(t) and the SOC prediction value. i When the frequency drops sharply and the response time requirement is extremely high, trip control is selected; when the frequency change is mild and the energy storage SOC is sufficient, power regulation mode is selected.

[0045] The optimization model solution and power allocation module is used to adjust power deficit. As a target, adjustable energy storage clusters For resource constraints, the control mode variable u i Using these as decision variables, a multi-objective optimization model is constructed with the objectives of "minimum frequency deviation + minimum control cost" to solve for the regulation power ΔP of each energy storage unit. i Or trip command;

[0046] The control command execution and response closed-loop module is used to issue commands to the units in the adjustable energy storage set based on the optimization model solution and the optimization results of the power allocation module: if u i If u = 1, then power adjustment ΔPi is executed. i If the value is 0, a tripping operation is performed, forming a closed loop of "state awareness-decision-execution".

[0047] Furthermore, the method based on the grid inertia constant H and system capacity S... sys Calculate the regulation power deficit ,include:

[0048] ;

[0049] Where H represents the power grid inertia constant, S sys This refers to the system capacity.

[0050] Furthermore, the input vector of the Long Short-Term Memory Network SOC estimation model It contains time-series data with at least three dimensions, namely voltage, current, and temperature, and is trained using the mean squared error (MSE) loss function to ensure that the state of charge (SOC) estimation error is ≤5%; the adjustable energy storage set only contains data that meets the SOC requirements. min ≤SOC_i(t)≤SOC max The energy storage unit, in which SOC min The value range is 20%-30%, SOC max The value range is 85%-95%.

[0051] Furthermore, the objective function of the multi-objective optimization model is as follows:

[0052] ;

[0053] in:

[0054] λ1, λ2: Weighting coefficients

[0055] Tripping cost of the i-th device

[0056] Unit power regulation cost

[0057] ΔP i : Adjusting power variables.

[0058] Furthermore, the multi-objective optimization model includes the following constraints:

[0059] (1) The total power meets the requirements:

[0060] (9);

[0061] Where a i For the controllability of energy storage unit i;

[0062] (2) Adjusting power boundary constraints:

[0063] (10);

[0064] (3) Trip power limit:

[0065] (11);

[0066] (4) Control mode constraints:

[0067] (12).

[0068] This invention constructs an energy storage SOC estimation model based on operational data to identify the dispatchable status of energy storage units in real time, thereby improving the accuracy of resource allocation. Simultaneously, it designs a dual-mode collaborative mechanism for tripping and regulation, allowing the system to dynamically select the optimal control mode based on frequency deviation and energy storage capacity, achieving an organic integration of rigid control and flexible regulation. The control strategy incorporates a multi-objective optimization principle, with frequency stability and control economy as core objectives, comprehensively considering control accuracy and resource efficiency. Compared to traditional solutions, it offers advantages in response speed, resource utilization, and control reliability. Attached Figure Description

[0069] Figure 1 This is a flowchart of the energy storage load shedding control method based on SOC estimation and mode adaptation of the present invention. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] To effectively address the large-scale power deficit problem caused by faults at the receiving end of ultra-high voltage direct current (UHVDC) transmission systems, this invention proposes a highly reliable load shedding control method that integrates energy storage state perception and adaptive control mode selection. This method achieves a balance between "rapid response" and "precise control" through dynamic identification and flexible scheduling of energy storage resources. The entire method consists of the following five core steps, as follows: Figure 1 As shown:

[0072] Step 1: Fault Detection and Adjustment Needs Assessment

[0073] During the operation of an ultra-high voltage (UHV) receiving-end system, if the DC transmission channel experiences a power interruption due to reasons such as blocking, converter failure, or protection operation, it will cause a sharp drop in the grid frequency within a very short time, which may lead to system instability in severe cases. To achieve rapid response, this invention first designs a fault identification and frequency deviation monitoring mechanism.

[0074] The system acquires the frequency f(t) in real time and compares it with the set rated frequency f. re When comparing, when the frequency deviation Δf(t) = f(t) − f ref Exceeding the warning threshold Δf th When the Hz level reaches -0.3 Hz, the energy storage mechanism is triggered to participate in regulation.

[0075] At the same time, the current required regulation power deficit is estimated based on the equivalent inertia and frequency change rate of the power grid:

[0076]

[0077] Where H represents the power grid inertia constant, S sys This represents the system capacity. This calculation quantifies the load gap, providing a target benchmark for subsequent adjustment strategies.

[0078] Step 2: Energy Storage State Estimation and Resource Selection Based on Neural Networks

[0079] To achieve accurate identification of the regulation capability of energy storage units, this invention introduces a data-driven neural network model into the original SOC estimation process, replacing the traditional current integration (coulomb measurement) method. Although the traditional SOC estimation method has a simple structure, it is sensitive to sensor accuracy, battery nonlinearity, and temperature, and has a large cumulative error over a long period of time, making it difficult to meet the accuracy and scalability requirements under complex operating conditions.

[0080] To this end, this invention utilizes the powerful time series modeling and nonlinear fitting capabilities of Long Short-Term Memory (LSTM) networks to construct a high-precision online SOC estimation model, providing a reliable state basis for subsequent control strategies.

[0081] (1) Model input and output design:

[0082] Constructing a neural network model for SOC estimation:

[0083] \

[0084] in:

[0085] : The predicted state of charge of the i-th energy storage unit at time t;

[0086] θ: The set of trainable parameters for the LSTM network

[0087] : An LSTM neural network model with parameter θ;

[0088] The multidimensional time series input vector of the i-th energy storage unit at time t is defined as follows:

[0089]

[0090] in:

[0091] V i(t): Voltage of energy storage unit i at time t

[0092] I i (t): Current of energy storage unit i at time t

[0093] T i (t): Temperature of energy storage unit i at time t

[0094] n: Length of the time window (number of steps in the input sequence, such as 10~20)

[0095] The model supports the integration of other optional features such as charge / discharge status indicators, historical SOC tags, battery aging levels, and manufacturer characteristics as auxiliary inputs to further improve estimation accuracy.

[0096] (2) Network training and accuracy assurance:

[0097] Training employs a supervised learning approach, using experimental data or labeled energy storage datasets from actual operation for fitting. Label values ​​are typically derived from high-precision experimental instruments, laboratory benches, or reliable offline SOC estimation tools.

[0098] The model training objective is to minimize the error between the predicted SOC and the true SOC, using the mean squared error (MSE) loss function:

[0099]

[0100] in:

[0101] N: Total number of training samples

[0102] : The true SOC label of sample j

[0103] Model predictions

[0104] After training, the model can be deployed to various energy storage management systems (BMS) or regional controllers to support online and edge-based SOC estimation.

[0105] (3) Network training and accuracy assurance:

[0106] After obtaining the predicted SOC value, a state threshold function is set to determine the energy storage adjustability:

[0107]

[0108] in:

[0109] a i (t): The controllability of energy storage unit i at time t

[0110] SOC min Adjustable minimum SOC threshold (e.g., 20%)

[0111] SOC max Adjustable maximum SOC threshold (e.g., 95%)

[0112] Based on this, an adjustable energy storage system can be constructed:

[0113]

[0114] This set serves as the resource boundary for subsequent optimization model scheduling. Power regulation or tripping commands are only issued to the energy storage units within it, thereby improving the executability of control commands and system security.

[0115] Step 3: Adaptive selection of control mode

[0116] Traditional energy storage regulation often employs a single mode (such as regulation only or tripping only), which suffers from problems such as untimely response and unreasonable resource utilization. This invention introduces an adaptive control mode mechanism, which autonomously selects either "tripping control" or "power regulation" mode based on the current frequency drop of the system and the energy storage status, or even uses a combination of the two modes in a coordinated manner, to achieve a balance between "rigid speed" and "flexible precision".

[0117] The control mode is represented by the following binary variables:

[0118]

[0119] Where u i =0 indicates trip control, u i =1 indicates that continuous power regulation is used.

[0120] For example:

[0121] If the frequency drops sharply and the response time requirement is extremely high, tripping should be the preferred option to ensure rapid power reduction.

[0122] If the frequency variation is mild and the energy storage SOC is sufficient, the regulation mode should be used first to improve economy and controllability.

[0123] If some devices have insufficient SOC, they can be set to trip control, while the rest participate in regulation, forming a "hybrid strategy".

[0124] This mechanism is highly flexible, making it easy to construct multi-scenario control strategies and providing clear variable definitions for subsequent mathematical optimization modeling.

[0125] Step 4: Optimize model solution and power allocation

[0126] After identifying the adjustable energy storage devices and their respective states, a multi-objective optimization model was constructed with the goal of "minimum frequency deviation + minimum control cost". This model takes into account response speed, control costs and equipment load capacity in a comprehensive manner, and scientifically configures the control tasks.

[0127] The objective function is as follows:

[0128]

[0129] in:

[0130] λ1, λ2: Weighting coefficients

[0131] Tripping cost of the i-th device

[0132] Unit power regulation cost

[0133] ΔP i Adjusting power variables

[0134] The main constraints include:

[0135] (1) The total power meets the requirements:

[0136] (9)

[0137] (2) Adjusting power boundary constraints:

[0138] (10)

[0139] (3) Trip power limit:

[0140] (11)

[0141] (4) Control mode constraints:

[0142] (12)

[0143] The above model can be solved using mixed integer programming (MIP) or intelligent optimization algorithms such as reinforcement learning and adaptive gradient descent, supporting parallel scheduling of large-scale energy storage devices and possessing practicality and scalability.

[0144] Step 5: Closed loop of control command execution and response

[0145] After the optimization results are generated, the system sends control commands to each energy storage terminal, in the following form:

[0146] If u i =1: Issue continuous power setpoint ΔP iThis requires the terminal to adjust its output power instantaneously;

[0147] If u i =0: Issues a trip command to quickly disconnect part of the load or equipment branch, thereby releasing the load.

[0148] To ensure that the response time meets the needs of emergency control, this invention suggests deploying a localized strategy execution module on the terminal side to reduce reliance on the main station's decision chain and improve response speed and anti-interference capabilities.

[0149] Ultimately, a complete closed-loop control chain of "state perception - optimized decision-making - rapid execution" is formed, significantly improving the ability of the UHV receiving-end system to cope with sudden power shortages.

[0150] This invention also provides an energy storage load shedding control device based on SOC estimation and mode adaptation, comprising:

[0151] The fault detection and regulation demand assessment module is used to collect the power grid frequency in real time and calculate the frequency deviation Δf(t). When Δf(t) exceeds the warning threshold Δf th At that time, the energy storage regulation mechanism is triggered; simultaneously, based on the grid inertia constant H and the system capacity S sys Calculate the regulation power deficit This serves as the target benchmark for subsequent control strategies.

[0152] A neural network-based energy storage state estimation and resource selection module is used to construct a long short-term memory network SOC estimation model, using the voltage V of the energy storage unit as the basis for the estimation. i (t), Current I i (t), temperature T i (t) and historical time series data are used as inputs, and the real-time SOC prediction value of each energy storage unit is output; based on the adjustable minimum SOC threshold SOC min and adjustable maximum SOC threshold SOC max Select adjustable energy storage sets This serves as a boundary condition for resource scheduling.

[0153] The adaptive control mode selection module is used to determine the control mode variable u based on the frequency deviation Δf(t) and the SOC prediction value. i When the frequency drops sharply and the response time requirement is extremely high, trip control is selected; when the frequency change is mild and the energy storage SOC is sufficient, power regulation mode is selected.

[0154] The optimization model solution and power allocation module is used to adjust power deficit. As a target, adjustable energy storage clusters For resource constraints, the control mode variable u iUsing these as decision variables, a multi-objective optimization model is constructed with the objectives of "minimum frequency deviation + minimum control cost" to solve for the regulation power ΔP of each energy storage unit. i Or trip command;

[0155] The control command execution and response closed-loop module is used to issue commands to the units in the adjustable energy storage set based on the optimization model solution and the optimization results of the power allocation module: if u i If u = 1, then power adjustment ΔPi is executed. i If the value is 0, a tripping operation is performed, forming a closed loop of "state awareness-decision-execution".

[0156] This invention has the following characteristics:

[0157] 1) This invention can be used for emergency regulation of UHV receiving-end power grids in the event of a large power deficit caused by DC faults, filling the technical gap of lacking a flexible control mode switching mechanism in the process of energy storage participating in load shedding control.

[0158] 2) This invention constructs a neural network-based energy storage SOC estimation model to dynamically identify the operating status of energy storage units, avoiding the control errors caused by traditional static SOC judgment methods. This not only improves the reliability of control commands but also significantly improves the scheduling efficiency of energy storage resources.

[0159] 3) This invention uses the dual objectives of frequency stability and control cost as optimization criteria to establish a multi-objective optimization regulation model. While meeting the system's frequency support requirements, it minimizes the cost of energy storage regulation, achieving a balance between the technical feasibility and economic rationality of energy storage participating in load shedding control.

[0160] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for energy storage load shedding control based on SOC estimation and mode adaptation, characterized in that, Includes the following steps: (1) Fault detection and regulation requirement assessment: Real-time acquisition of power grid frequency and calculation of frequency deviation Δf(t). When Δf(t) exceeds the warning threshold Δf th At that time, the energy storage regulation mechanism is triggered; simultaneously, based on the grid inertia constant H and the system capacity S sys Calculate the regulation power deficit This serves as the target benchmark for subsequent control strategies. (2) Energy storage state estimation and resource selection based on neural network: Construct a long short-term memory network SOC estimation model, using the voltage V of the energy storage unit as the basis for the estimation. i (t), Current I i (t), temperature T i (t) and historical time series data are used as inputs, and the real-time SOC prediction value of each energy storage unit is output. Based on the adjustable minimum SOC threshold min and adjustable maximum SOC threshold SOC max Select adjustable energy storage sets This serves as a boundary condition for resource scheduling. (3) Adaptive selection of control mode: The control mode variable u is determined based on the frequency deviation Δf(t) in step (1) and the SOC prediction value in step (2). i When the frequency drops sharply and the response time requirement is extremely high, trip control is selected; when the frequency change is mild and the energy storage SOC is sufficient, power regulation mode is selected. (4) Optimization model solution and power allocation: Adjusting the power deficit in step (1) For the objective, the adjustable energy storage set in step (2) is the resource constraint, and the control mode variable u in step (3) is... i Using these as decision variables, a multi-objective optimization model is constructed with the objectives of "minimum frequency deviation + minimum control cost" to obtain the regulation power ΔP of each energy storage unit. i Or trip command; (5) Control command execution and response closed loop: Based on the optimization results of step (4), commands are issued to the units in the adjustable energy storage set: if u i If u = 1, then power adjustment ΔPi is executed. i If the value is 0, a tripping operation is performed, forming a closed loop of "state awareness-decision-execution".

2. The method as described in claim 1, characterized in that: In step (1), based on the grid inertia constant H and the system capacity S sys Calculate the regulation power deficit ,include: ； Where H represents the power grid inertia constant, S sys This refers to the system capacity.

3. The method as described in claim 1, characterized in that: In step (2), the input vector of the Long Short-Term Memory Network SOC estimation model It contains time-series data with at least three dimensions, namely voltage, current, and temperature, and is trained using the mean squared error (MSE) loss function to ensure that the state of charge (SOC) estimation error is ≤5%; the adjustable energy storage set only contains data that meets the SOC requirements. min ≤SOC_i(t)≤SOC max The energy storage unit, in which SOC min The value range is 20%-30%, SOC max The value range is 85%-95%.

4. The method as described in claim 1, characterized in that: The objective function of the multi-objective optimization model described in step (4) is as follows: ； in: λ1, λ2: Weighting coefficients; : The tripping cost of the i-th device; Unit power regulation cost; ΔP i : Adjusting power variables.

5. The method as described in claim 4, characterized in that: The multi-objective optimization model described in step (4) includes the following constraints: (1) The total power meets the requirements: （9）； Where a i For the controllability of energy storage unit i; (2) Adjusting power boundary constraints: （10）； (3) Trip power limit: （11）； (4) Control mode constraints: （12）。 6. A storage load shedding control device based on SOC estimation and mode adaptation, characterized in that, include: The fault detection and regulation demand assessment module is used to collect the power grid frequency in real time and calculate the frequency deviation Δf(t). When Δf(t) exceeds the warning threshold Δf th At that time, the energy storage regulation mechanism is triggered; simultaneously, based on the grid inertia constant H and the system capacity S sys Calculate the regulation power deficit This serves as the target benchmark for subsequent control strategies. A neural network-based energy storage state estimation and resource selection module is used to construct a long short-term memory network SOC estimation model, using the voltage V of the energy storage unit as the basis for the estimation. i (t), Current I i (t), temperature T i (t) and historical time series data are used as inputs, and the real-time SOC prediction value of each energy storage unit is output. Based on the adjustable minimum SOC threshold min and adjustable maximum SOC threshold SOC max Select adjustable energy storage sets This serves as a boundary condition for resource scheduling. The adaptive control mode selection module is used to determine the control mode variable u based on the frequency deviation Δf(t) and the SOC prediction value. i When the frequency drops sharply and the response time requirement is extremely high, trip control is selected; when the frequency change is mild and the energy storage SOC is sufficient, power regulation mode is selected. The optimization model solution and power allocation module is used to adjust power deficit. As a target, adjustable energy storage clusters For resource constraints, the control mode variable u i Using these as decision variables, a multi-objective optimization model is constructed with the objectives of "minimum frequency deviation + minimum control cost" to obtain the regulation power ΔP of each energy storage unit. i Or trip command; The control command execution and response closed-loop module is used to issue commands to the units in the adjustable energy storage set based on the optimization model solution and the optimization results of the power allocation module: if u i If u = 1, then power adjustment ΔPi is executed. i If the value is 0, a tripping operation is performed, forming a closed loop of "state awareness-decision-execution".

7. The apparatus as claimed in claim 6, characterized in that: The basis is the power grid inertia constant H and the system capacity S. sys Calculate the regulation power deficit ,include: ； Where H represents the power grid inertia constant, S sys This refers to the system capacity.

8. The apparatus as claimed in claim 6, characterized in that: The input vector of the Long Short-Term Memory Network SOC estimation model It contains time-series data with at least three dimensions, namely voltage, current, and temperature, and is trained using the mean squared error (MSE) loss function to ensure that the state of charge (SOC) estimation error is ≤5%; the adjustable energy storage set only contains data that meets the SOC requirements. min ≤SOC_i(t)≤SOC max The energy storage unit, in which SOC min The value range is 20%-30%, SOC max The value range is 85%-95%.

9. The apparatus as claimed in claim 6, characterized in that: The objective function of the multi-objective optimization model is as follows: ； in: λ1, λ2: Weighting coefficients; : The tripping cost of the i-th device; Unit power regulation cost; ΔP i : Adjusting power variables.

10. The apparatus as claimed in claim 9, characterized in that: The multi-objective optimization model includes the following constraints: (1) The total power meets the requirements: （9）； Where a i For the controllability of energy storage unit i; (2) Adjusting power boundary constraints: （10）； (3) Trip power limit: （11）； (4) Control mode constraints: （12）。

Images