Method for assisting frequency modulation of thermal power unit by double-layer coordinated control water-based organic liquid flow battery

Abstract

The application discloses a method for frequency modulation of a thermal power unit assisted by a water series organic liquid flow battery, and the method constructs a double-layer coordinated control architecture combining power optimization distribution of an upper layer with power accurate tracking of a lower layer, and comprises the following steps: S1, constructing a time-varying mathematical model of a combined frequency modulation system integrated by a load frequency control submodel of the thermal power unit and a nonlinear dynamic recursive submodel set of the water series organic liquid flow battery considering time-varying characteristics of multiple parameters; S2, constructing a dynamic output boundary prediction mechanism; S3, performing power optimization distribution of the upper layer, and calculating a difference between a total automatic generation control (AGC) instruction and a current actual output of the thermal power unit to obtain a current total power shortage demand of the combined frequency modulation system; performing priority screening on each energy storage unit in an array according to upper and lower limits of the dynamic output boundary output by the step S2, and eliminating the energy storage units without adjustment capacity in a current charging and discharging direction to determine effective participants of this rolling; and S4, performing local power accurate tracking control of the lower layer.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control and energy storage technology for new power systems, specifically relating to a secondary frequency regulation method for automatic generation control (AGC) of hydro-organic flow battery-assisted thermal power units based on two-layer coordinated control. Background Technology

[0002] With the advancement of the "dual carbon" target, the large-scale integration of new energy sources such as wind power and photovoltaics into the grid brings significant randomness and volatility, posing a severe challenge to system frequency stability. Traditional automatic generation control (AGC) frequency regulation mainly relies on thermal power units. However, thermal power units are constrained by the large thermal inertia of governors, turbines, and boiler heat storage, resulting in inherent defects such as lag in power change, slow ramp-up rate, and low regulation accuracy, making it difficult to independently adapt to the frequent and rapid power fluctuations of the grid. In recent years, the introduction of energy storage systems to assist thermal power units in coordinated frequency regulation has become an efficient means to solve the above pain points. Currently, the mainstream thermal-storage frequency regulation route mostly uses lithium-ion batteries. However, under the high-frequency charge and discharge disturbances of frequency regulation conditions, lithium batteries are prone to lithium dendrite formation, electrode pulverization, and thermal runaway risks. Their limited cycle life also seriously weakens the economic benefits throughout the entire life cycle.

[0003] In contrast, aqueous organic flow batteries, as a novel long-term energy storage technology, utilize aqueous electrolytes with no risk of combustion or explosion. They possess core advantages such as high intrinsic safety, extremely long cycle life (over 20,000 cycles), and a decoupled power-capacity design, making them highly compatible with the high safety requirements and frequent charge-discharge frequency regulation conditions of thermal power plants. However, aqueous organic flow batteries exhibit strong nonlinear and time-varying characteristics during long-term dynamic frequency regulation (such as increased internal resistance, decreased usable capacity, and efficiency degradation during operation). Furthermore, due to inconsistent or uneven initial states of the various stack units, severe imbalances in the state of charge (SOC) between units can easily occur, leading to some units prematurely reaching their operating limits and exiting regulation.

[0004] Existing energy storage power allocation strategies (such as allocation by capacity ratio or fixed boundary average allocation) do not perceive the time-varying nonlinear boundary of the battery and the differences in the SOC state of each cell. This causes the traditional proportional-integral (PI) control at the lower level to produce significant tracking lag when facing the inverter's execution hysteresis, and cannot simultaneously achieve multi-objective collaborative optimization of high-precision frequency regulation, active SOC equalization management and lifetime loss suppression. Summary of the Invention

[0005] This invention provides a dual-layer coordinated control method for frequency regulation of thermal power units assisted by aqueous organic flow batteries, in order to solve the technical problems existing in the prior art.

[0006] To achieve the above objectives, this invention provides a dual-layer coordinated control method for frequency regulation of aqueous organic flow battery-assisted thermal power units. The method constructs a dual-layer coordinated control architecture combining upper-layer power optimization allocation and lower-layer precise power tracking, and includes the following steps: S1: Construct a time-varying mathematical model for a joint frequency regulation system that integrates a thermal power unit load frequency control sub-model with a nonlinear dynamic recursive sub-model of an aqueous organic flow battery that takes into account multiple time-varying parameters; S2: Construct a dynamic output boundary prediction mechanism, collect the real-time state of charge of each energy storage unit in the array, and predict the dynamic output upper and lower limits of each independent energy storage unit in the power-time dimension in the future control time domain. S3: Execute upper-level power optimization allocation, calculate in real time the difference between the total AGC command and the current actual output of the thermal power unit to obtain the current total power deficit requirement of the joint frequency regulation system; based on the dynamic output upper and lower limit boundaries output in step S2, prioritize each energy storage unit in the array, eliminate energy storage units without adjustment capability in the current charging and discharging direction to determine the effective participants in this rolling process; construct a multi-objective optimization function that takes into account command tracking, stack life loss, and active consistency balance of state of charge, and use an improved particle swarm optimization algorithm to solve the power allocation components of the effective participants in a rolling process; S4: Executes precise local power tracking control at the lower level. After receiving the optimal power allocation command, the local controller of each independent energy storage unit triggers an adaptive composite control law that introduces feedforward dynamic self-learning compensation. This law approximates the nonlinear inverse dynamics of the converter stage online, eliminates feedback execution lag, and drives the actual output of each energy storage unit to match the allocation command.

[0007] In one embodiment of the present invention, preferably, in step S3, after each position iteration of the particle swarm optimization algorithm, a dynamic physical-level projection repair mechanism is executed to match the total grid-connected power and physical hard constraints: First, the sum of the allocated power of each energy storage unit output in the current iteration is checked. If the total amount is not equal to the total power deficit requirement, the output component of each unit is forcibly adjusted proportionally according to the current output margin ratio of each effective energy storage unit. During the adjustment process, if it is detected that the allocated power of an independent energy storage unit exceeds the dynamic output boundary hard constraint predicted in step S2 at that time node, the output of that unit is immediately limited to the dynamic boundary value, and the truncated difference power is re-distributed to the remaining available energy storage units that still have output margin, until the total grid-connected power is completely balanced. When the optimization ends, the optimal power allocation command of each effective energy storage unit is output.

[0008] In one embodiment of the present invention, preferably, in step S1, the thermal power unit load frequency control sub-model includes the thermal power unit load frequency control time-domain continuous state propagation equation as shown in the following formula:

[0009]

[0010]

[0011]

[0012] Its mechanical power output Satisfying the relation:

[0013] Actual output of the unit for: .

[0014] In the formula, - These are the internal state quantities of the speed governor, the high-pressure cylinder steam capacity, the reheat circuit, and the boiler's thermal storage inertia, respectively. For unit power command, The time constants for each stage, This represents the power percentage of the high-pressure cylinder.

[0015] Meanwhile, the change in active power command within each sampling step is subject to a hard limit on the discrete differential ramp rate:

[0016] In the formula, , Let represent the power command values ​​of the th energy storage unit at time k-1 and time k, respectively. This indicates the maximum allowable power adjustment step size within a single control cycle. The rated ramp rate of the unit, This represents the discrete sampling step size.

[0017] In one embodiment of the present invention, preferably, in step S1, the nonlinear dynamic recursive sub-model of the aqueous organic flow battery includes: The actual active power response of the energy storage unit satisfies the first-order inertial dynamic response equation:

[0018] In the formula, This is an on-site action command. The actual output power of the action. The active power conversion time constant of the converter stage is the electrochemical conversion active power within the energy storage unit. Satisfying the nonlinear efficiency conversion rule of charge and discharge segments: when hour ;when hour, In the formula, For unidirectional efficiency, The state of charge of each energy storage unit satisfies the discrete recursive equation:

[0019] In the formula, This represents the actual usable capacity. To account for self-discharge loss, the second-level self-discharge rate constant, , These represent the real-time state of charge of the i-th energy storage unit at the k-th and k+1-th control times, respectively. The terminal voltage of each energy storage unit satisfies the polarization time-varying equation:

[0020] In the formula, This represents the real-time terminal voltage of the i-th energy storage unit at time t. Open circuit voltage, For loop current, For time-varying internal resistance, The long-term time-varying degradation of energy storage units satisfies the drift equation:

[0021] In the formula, and The initial reference internal resistance and rated capacity, This is the internal resistance drift coefficient. The coefficient representing the change in capacity lifetime decay is T, where T is the sampling period. The unit-integrated active power loss equation satisfies:

[0022] In the formula, To maintain constant parasitic pump loss power, This is the rated output power.

[0023] In one embodiment of the present invention, preferably, in step S2, the maximum power prediction model for the discharge direction is:

[0024] In the formula, This represents the maximum allowable discharge power of the energy storage unit under its current state. The lower limit of SOC safe operation is defined as m, where m is the width of the output retreat zone, and the maximum power in the charging direction is determined symmetrically based on the upper limit of safety.

[0025] In one embodiment of the present invention, preferably, in step S3, the multi-objective optimization function F is a four-objective weighted function:

[0026] In the formula, , , , Assign weight coefficients to each sub-objective. , , , Each corresponds to a command to track sub-targets Lifetime loss sub-target State of Charge (SOC) Consistency Sub-Objective Sub-objective of comprehensive economic benefits of frequency modulation throughout the entire life cycle ; The State of Charge (SOC) consistency sub-objective is guided by a first-order equilibrium reward function for multi-objective solution.

[0027] In the formula, This represents the real-time state of charge of the i-th energy storage unit. The average state of charge of the current participating regulation unit is denoted as n, and the number of energy storage units participating in regulation is denoted as n.

[0028] In one embodiment of the present invention, preferably, the combination of weight coefficients in the multi-objective optimization function is determined through offline orthogonal experiment optimization, and its set value is: instruction tracking weight. =0.50, lifespan loss weight =0.15, SOC consistency weight =0.25, economic benefit weight =0.10; The rolling optimization update cycle of the upper-layer power optimization allocation is physically consistent with the discrete issuance communication cycle of the automatic generation control command of the external regional power system, and is set to 8s.

[0029] In one embodiment of the present invention, preferably, in step S4, the composite control law of the lower-level cerebellar model CMAC neural network is:

[0030] In the formula, To execute local operation commands sent to the PCS converter. The optimal power allocation command issued by the upper layer for optimization. This is the feedforward dynamic compensation term output by the CMAC neural network after online self-learning following multi-segment nonlinear quantization and associative grid mapping. For the control gain of the local proportional controller, This is the current actual output feedback of the energy storage unit.

[0031] In one embodiment of the present invention, preferably, in step S1, the nonlinear dynamic recursive sub-model of the energy storage system explicitly incorporates the parameter degradation characteristics during the long-term continuous frequency regulation service process; in the discrete state recursion, the internal parameters of its mathematical model include the time-varying internal resistance function R(t) of the individual unit, which exhibits a time-varying drift law that increases or decreases with the total duration of long-term frequency regulation operation, and the time-varying decay function of the actual usable capacity of the independent unit. The unit's time-varying internal resistance function R(t) and the actual usable capacity's time-varying decay function... The parameter drift amplitude coefficients are pre-calibrated and configured according to the cycle life decay and internal resistance increase trend of the physical battery, so as to apply the dynamic output protection boundary of the frequency modulation control system during long-term operation in steps S2 and S3.

[0032] The present invention also provides a frequency regulation control system for a water-based organic liquid flow battery-assisted thermal power unit with dual-layer coordinated control for implementing the above method, the control system comprising: Data acquisition terminal: used to collect in real time the AGC total target command of the regional power system, the current actual response output of thermal power units, and the real-time state of charge, terminal voltage and electrolyte parameters of each independent energy storage unit in the water system organic liquid flow battery array; The core of the central frequency regulation power coordination control: receives the data collected by the data acquisition terminal, embeds the time-varying mathematical model of the joint frequency regulation system, predicts the power-time dimension output boundary constraints of each unit based on the safety window, and outputs the optimal power allocation command of each effective energy storage unit by executing the improved particle swarm optimization algorithm and the physical projection repair mechanism. Energy storage unit local power tracking controller group: It is communicatively connected to the central frequency regulation power coordination control core. Each local power tracking controller controls a PCS power electronic converter. The local power tracking controller is embedded with a cerebellum model neural network composite tracking control law, which drives the actual output of the corresponding energy storage unit to follow the optimal power allocation command at the millisecond level.

[0033] Compared with traditional frequency modulation control strategies, the present invention has the following beneficial technical effects: 1. High-precision command tracking and excellent frequency support capabilities: The lower-level control introduces a CMAC neural network online learning feedforward channel to approximate the nonlinear inverse dynamics of the controlled object. Under random AGC conditions, the system's root mean square error in command tracking is reduced to 0.165 MW, and the regulation accuracy and comprehensive frequency modulation performance index K value reaches 0.991, completely solving the regulation delay and reverse deviation problems of traditional pure feedback PI control. Under AGC step command conditions, the system's tracking qualification rate reaches 83.69%, an improvement of 23.12 percentage points compared to traditional PI control; under time-varying battery parameters, the root mean square error in tracking remains at 0.165 MW, and the frequency modulation K value remains stable at 0.991, with no significant performance degradation.

[0034] 2. Superior active state of charge (SOC) balancing management: The upper-level optimization algorithm embeds a first-order balancing reward function, which enables the SOC of each energy storage unit initially distributed in an unbalanced state (e.g., standard deviation of 0.169) to converge rapidly under long-term continuous operation, reducing the SOC standard deviation at the end to below 0.065 (a reduction of about 61.5% compared to the average distribution strategy).

[0035] 3. Significantly suppresses electrolyte stress damage and improves life cycle benefits: The strategy successfully avoids chemical stress aging damage caused by deep charging and discharging in some individual stacks. While ensuring the high standard of grid performance, under continuous 1800s of random operating conditions, the battery loss cost is reduced by 6.3% (1748 yuan → 1638 yuan) compared to the fixed boundary average distribution strategy, and the overall frequency regulation benefit is increased by 1.0% (8848 yuan → 8938 yuan).

[0036] 4. Strong robustness against battery aging: The system has an embedded output prediction boundary real-time refresh mechanism. Even when faced with severe parameter drift such as increased internal resistance to 1.35 times and capacity decay in the later stages of long-term system operation, the comprehensive frequency regulation indicators can still remain stable, demonstrating excellent engineering adaptability. Attached Figure Description

[0037] 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 these drawings without creative effort.

[0038] Figure 1 A block diagram of the overall structure of the "thermal power unit-flow battery" joint frequency regulation; Figure 2 A block diagram of the mathematical model of the load frequency control subsystem of a thermal power unit; Figure 3A simulation structural block diagram of an aqueous organic flow battery energy storage subsystem; Figure 4 This is a comparison chart of the fitness evolution process between the upper-level P-AWPSO algorithm and the standard PSO algorithm; Figure 5 A comparison chart of the convergence changes in the State of Charge (SOC) of each flow battery unit under random grid fluctuation conditions; Figure 6 A comparison chart of tracking errors under time-varying parameters; Figure 7 This is a comprehensive comparison chart of five dimensions of evaluation indicators for three types of working conditions. Detailed Implementation

[0039] 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, and 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.

[0040] The technical problem to be solved by this invention is: to address the problems of tracking lag, SOC imbalance and accelerated battery aging caused by the nonlinear time-varying characteristics of the battery itself and the inconsistency of the state between units during the frequency regulation of thermal power units assisted by aqueous organic flow batteries, and to provide a frequency regulation control method that can sense the dynamic output boundary in real time, realize the optimal power allocation of multiple objectives and achieve millisecond-level accurate lag-free tracking.

[0041] First embodiment: This invention provides a dual-layer coordinated control method for frequency regulation of aqueous organic flow battery-assisted thermal power units. The method constructs a dual-layer coordinated control architecture combining upper-layer power optimization allocation and lower-layer precise power tracking, and includes the following steps: S1: Construct a time-varying mathematical model for a joint frequency regulation system that integrates a thermal power unit load frequency control sub-model with a nonlinear dynamic recursive sub-model of an aqueous organic flow battery that takes into account multiple time-varying parameters; S2: Construct a dynamic output boundary prediction mechanism, collect the real-time state of charge of each energy storage unit in the array, and predict the dynamic output upper and lower limits of each independent energy storage unit in the power-time dimension in the future control time domain. S3: Execute upper-level power optimization allocation, calculate in real time the difference between the total AGC command and the current actual output of the thermal power unit to obtain the current total power deficit requirement of the joint frequency regulation system; based on the dynamic output upper and lower limit boundaries output in step S2, prioritize each energy storage unit in the array, eliminate energy storage units without adjustment capability in the current charging and discharging direction to determine the effective participants in this rolling process; construct a multi-objective optimization function that takes into account command tracking, stack life loss, and active consistency balance of state of charge, and use an improved particle swarm optimization algorithm to solve the power allocation components of the effective participants in a rolling process; S4: Executes precise local power tracking control at the lower level. After receiving the optimal power allocation command, the local controller of each independent energy storage unit triggers an adaptive composite control law that introduces feedforward dynamic self-learning compensation. This law approximates the nonlinear inverse dynamics of the converter stage online, eliminates feedback execution lag, and drives the actual output of each energy storage unit to match the allocation command.

[0042] In one embodiment of the present invention, preferably, in step S3, after each position iteration of the particle swarm optimization algorithm, a dynamic physical-level projection repair mechanism is executed to match the total grid-connected power and physical hard constraints: First, the sum of the allocated power of each energy storage unit output in the current iteration is checked. If the total amount is not equal to the total power deficit requirement, the output component of each unit is forcibly adjusted proportionally according to the current output margin ratio of each effective energy storage unit. During the adjustment process, if it is detected that the allocated power of an independent energy storage unit exceeds the dynamic output boundary hard constraint predicted in step S2 at that time node, the output of that unit is immediately limited to the dynamic boundary value, and the truncated difference power is re-distributed to the remaining available energy storage units that still have output margin, until the total grid-connected power is completely balanced. When the optimization ends, the optimal power allocation command of each effective energy storage unit is output.

[0043] In one embodiment of the present invention, preferably, in step S1, the thermal power unit load frequency control sub-model includes the thermal power unit load frequency control time-domain continuous state propagation equation as shown in the following formula:

[0044]

[0045]

[0046]

[0047] Its mechanical power output Satisfying the relation:

[0048] Actual output of the unit for: .

[0049] In the formula, - These are the internal state quantities of the speed governor, the high-pressure cylinder steam capacity, the reheat circuit, and the boiler's thermal storage inertia, respectively. For unit power command, The time constants for each stage, This represents the power percentage of the high-pressure cylinder.

[0050] Meanwhile, the change in active power command within each sampling step is subject to a hard limit on the discrete differential ramp rate:

[0051] In the formula, , Let represent the power command values ​​of the th energy storage unit at time k-1 and time k, respectively. This indicates the maximum allowable power adjustment step size within a single control cycle. The rated ramp rate of the unit, This represents the discrete sampling step size.

[0052] In one embodiment of the present invention, preferably, in step S1, the nonlinear dynamic recursive sub-model of the aqueous organic flow battery includes: The actual active power response of the energy storage unit satisfies the first-order inertial dynamic response equation:

[0053] In the formula, This is an on-site action command. The actual output power of the action. The active power conversion time constant of the converter stage is the electrochemical conversion active power within the energy storage unit. Satisfying the nonlinear efficiency conversion rule of charge and discharge segments: when hour ;when hour, In the formula, For unidirectional efficiency, The state of charge of each energy storage unit satisfies the discrete recursive equation:

[0054] In the formula, This represents the actual usable capacity. To account for self-discharge loss, the second-level self-discharge rate constant, , These represent the real-time state of charge of the i-th energy storage unit at the k-th and k+1-th control times, respectively. The terminal voltage of each energy storage unit satisfies the polarization time-varying equation:

[0055] In the formula, This represents the real-time terminal voltage of the i-th energy storage unit at time t. Open circuit voltage, For loop current, For time-varying internal resistance, The long-term time-varying degradation of energy storage units satisfies the drift equation:

[0056] In the formula, and The initial reference internal resistance and rated capacity, This is the internal resistance drift coefficient. The coefficient representing the change in capacity lifetime decay is T, where T is the sampling period. The unit-integrated active power loss equation satisfies:

[0057] In the formula, To maintain constant parasitic pump loss power, This is the rated output power.

[0058] In one embodiment of the present invention, preferably, in step S2, the maximum power prediction model for the discharge direction is:

[0059] In the formula, This represents the maximum allowable discharge power of the energy storage unit under its current state. The lower limit of SOC safe operation is defined as m, where m is the width of the output retreat zone, and the maximum power in the charging direction is determined symmetrically based on the upper limit of safety.

[0060] In one embodiment of the present invention, preferably, in step S3, the multi-objective optimization function F is a four-objective weighted function:

[0061] In the formula, , , , Assign weight coefficients to each sub-objective. , , , Each corresponds to a command to track sub-targets Lifetime loss sub-target State of Charge (SOC) Consistency Sub-Objective Sub-objective of comprehensive economic benefits of frequency modulation throughout the entire life cycle ; The State of Charge (SOC) consistency sub-objective is guided by a first-order equilibrium reward function for multi-objective solution.

[0062] In the formula, This represents the real-time state of charge of the i-th energy storage unit. The average state of charge of the current participating regulation unit is denoted as n, and the number of energy storage units participating in regulation is denoted as n.

[0063] In one embodiment of the present invention, preferably, the combination of weight coefficients in the multi-objective optimization function is determined through offline orthogonal experiment optimization, and its set value is: instruction tracking weight. =0.50, lifespan loss weight =0.15, SOC consistency weight =0.25, economic benefit weight =0.10; The rolling optimization update cycle of the upper-layer power optimization allocation is physically consistent with the discrete issuance communication cycle of the automatic generation control command of the external regional power system, and is set to 8s.

[0064] In one embodiment of the present invention, preferably, in step S4, the composite control law of the lower-level cerebellar model CMAC neural network is:

[0065] In the formula, To execute local operation commands sent to the PCS converter. The optimal power allocation command issued by the upper layer for optimization. This is the feedforward dynamic compensation term output by the CMAC neural network after online self-learning following multi-segment nonlinear quantization and associative grid mapping. For the control gain of the local proportional controller, This is the current actual output feedback of the energy storage unit.

[0066] In one embodiment of the present invention, preferably, in step S1, the nonlinear dynamic recursive sub-model of the energy storage system explicitly incorporates the parameter degradation characteristics during the long-term continuous frequency regulation service process; in the discrete state recursion, the internal parameters of its mathematical model include the time-varying internal resistance function R(t) of the individual unit, which exhibits a time-varying drift law that increases or decreases with the total duration of long-term frequency regulation operation, and the time-varying decay function of the actual usable capacity of the independent unit. The unit's time-varying internal resistance function R(t) and the actual usable capacity's time-varying decay function... The parameter drift amplitude coefficients are pre-calibrated and configured according to the cycle life decay and internal resistance increase trend of the physical battery, so as to apply the dynamic output protection boundary of the frequency modulation control system during long-term operation in steps S2 and S3.

[0067] The present invention also provides a frequency regulation control system for a water-based organic liquid flow battery-assisted thermal power unit with dual-layer coordinated control for implementing the above method, the control system comprising: Data acquisition terminal: used to collect in real time the AGC total target command of the regional power system, the current actual response output of thermal power units, and the real-time state of charge, terminal voltage and electrolyte parameters of each independent energy storage unit in the water system organic liquid flow battery array; The core of the central frequency regulation power coordination control: receives the data collected by the data acquisition terminal, embeds the time-varying mathematical model of the joint frequency regulation system, predicts the power-time dimension output boundary constraints of each unit based on the safety window, and outputs the optimal power allocation command of each effective energy storage unit by executing the improved particle swarm optimization algorithm and the physical projection repair mechanism. Energy storage unit local power tracking controller group: It is communicatively connected to the central frequency regulation power coordination control core. Each local power tracking controller controls a PCS power electronic converter. The local power tracking controller is embedded with a cerebellum model neural network composite tracking control law, which drives the actual output of the corresponding energy storage unit to follow the optimal power allocation command at the millisecond level.

[0068] Second embodiment: This invention constructs a two-layer coordinated control architecture that combines upper-layer power optimization allocation with lower-layer precise power tracking, specifically including the following steps: Step 1: Construct a nonlinear time-varying mathematical model for the joint frequency regulation system of "thermal power unit-water system organic flow battery".

[0069] 1. Establish the time-domain continuous state propagation equation for the load frequency control of thermal power units, as shown in the following formula (1):

[0070]

[0071]

[0072] (1) Its mechanical power output satisfies the following relationship:

[0073] The actual output of the unit is: .

[0074] In the formula, - These are the internal state quantities of the speed governor, the high-pressure cylinder steam capacity, the reheat circuit, and the boiler's thermal storage inertia, respectively. For unit power command, The time constants for each stage, This represents the power percentage of the high-pressure cylinder.

[0075] Meanwhile, the change in active power command of the unit within each sampling step is subject to a hard limit constraint of the discrete differential ramp rate, as shown in the following formula (2): (2) In the formula, The rated ramp rate of the unit, This represents the discrete sampling step size.

[0076] 2. Establish a nonlinear time-varying model for an aqueous organic flow battery that takes into account electrolyte mass transfer, dynamic state of charge (SOC), time-varying internal resistance, and pump power loss.

[0077] Considering the transient response characteristics and execution hysteresis of the power electronic converter (PCS) in actual operation, a first-order inertial dynamic propagation equation for the active power response of the energy storage unit converter is constructed as shown in Equation 3.

[0078] (3) In the formula, The local active power operation command issued to the energy storage unit by the lower-level control calculation. This refers to the actual operating output power of the energy storage unit. The active power execution time constant of the converter stage. Considering the efficiency difference of the internal nonlinear chemical energy-electric energy conversion of the aqueous organic electrolyte under charging and discharging states, the active power of the internal electrochemical conversion of the energy storage unit is constructed as shown in the following formula (4). Segmentation conversion rules: (4) In the formula, The unidirectional charge-discharge efficiency of the aqueous organic flow battery is given. The discrete-time recursive relationship of the State of Charge (SOC) of each independent energy storage unit is shown in the following formula (5): (5) In the formula, Let i be the state of charge of the i-th energy storage unit at step k. For discrete sampling step size, This represents the current actual available capacity of the independent unit. To account for self-discharge losses, the second-level self-discharge rate constant is rigorously calibrated based on the 24-hour voltage retention rate (97.44%) measured in the open-circuit standby test. The calculation method is as follows: .

[0079] Terminal voltage of each independent energy storage unit The following formula (6) is used to determine the internal nonlinear polarization and resistance characteristics: (6) In the formula, The battery open-circuit voltage varies with real-time state of charge. For single-loop current, The internal resistance is time-varying. Considering the time-varying degradation characteristics of electrochemical performance parameters during long-term continuous frequency modulation operation of the system, the linear drift law of the increase in internal resistance and the decrease in actual usable capacity with the total operating time is shown in the following formula (7): (7) In the formula, and These represent the initial reference internal resistance coefficient and the initial rated usable capacity of the energy storage unit, respectively. , These are the coefficients for internal resistance drift and capacity decay, respectively, and T is the total operating time. The time-varying drift coefficient of the internal resistance during long-term operation. This is the coefficient representing the magnitude of change in available capacity lifetime. This refers to the total duration of continuous service.

[0080] The combined internal active power loss of each independent energy storage unit during actual operation Defined by the following formula (8): (8) In the formula, To maintain a constant parasitic pump loss power in a single-pile electrolyte circulation mass transfer, This refers to the rated output power of a single energy storage unit.

[0081] Step 2: Construct a dynamic output boundary prediction mechanism to determine the upper and lower limits of the battery's dynamic output, which are updated in real time according to the operating conditions.

[0082] Within each optimization cycle, based on the real-time state of charge of each unit, electrolyte operating parameters, and stack operating conditions, the maximum charge and discharge capacity of the battery in the power-time dimension within the future control time domain is predicted. The maximum power prediction model in the discharge direction is shown in the following formula (9): (9) In the formula, The lower limit for safe operation of the SOC is defined by m, which represents the width of the output retreat zone. The maximum power in the charging direction is determined by the upper limit of safety in this symmetrical form, with dynamic hard constraint boundaries output in real time.

[0083] Step 3: Perform upper-level priority-based adaptive weighted particle swarm optimization (P-AWPSO) power allocation.

[0084] 1. Introduce priority screening logic: Based on the maximum power prediction results in step two, before the start of the rolling cycle, remove energy storage units that have no adjustment capability in the charging and discharging direction (i.e., close to the SOC safe operating limit) and only retain units with adjustable output capability to participate in this optimization allocation.

[0085] 2. Construct a multi-objective comprehensive optimization objective function: By balancing the priorities of different objectives through adaptive weighting factors, a comprehensive objective function as shown in Formula 10 is constructed.

[0086] (10) In the formula, , , , Assign weight coefficients to each sub-objective.

[0087] In the synthesis objective function, the instruction tracking sub-objective function As shown in the following formula (11), it is used to measure the deviation between the total battery output and the actual response output deficit of the thermal power unit: (11) In the formula, The battery needs to compensate for the total shortfall after subtracting the actual output of the thermal power unit from the total AGC command of the power grid.

[0088] State of Charge (SOC) Consistency Sub-Objective Function A first-order equalization reward function is used to guide high-SOC cells to discharge more and low-SOC cells to charge more, as shown in the following formula (12): (12) In the formula, The average state of charge of the current participating regulation unit is denoted as n, and the number of energy storage units participating in regulation is denoted as n.

[0089] Lifetime loss sub-objective function Includes the cumulative throughput of each internal unit The dynamic weight penalty is applied and a nonlinear asymmetric weight penalty is implemented, as shown in the following formula (13): (13) In the formula, To control the power distribution components, For dynamic lifetime penalty weights, To prevent tiny positive numbers with a denominator of zero.

[0090] The comprehensive economic penalty sub-objective function throughout the entire life cycle Introducing the costs of electricity market mileage compensation and capacity degradation loss, and taking a negative number as the minimum penalty term, as shown in the following formula (14):

[0091] (14) In the formula, This refers to the net profit within a single rolling cycle. The unit price for FM mileage compensation. This is the unit price for life degradation loss of the flow battery calculated from the initial construction cost.

[0092] 3. Rolling Update Optimization: During the optimization process, the particle velocity and position are updated as shown in the following formula (15): (15) in, The particle's current speed. For inertial weights, This refers to individual learning factors (cognitive coefficients). This is the social learning factor (social coefficient). , A random number uniformly distributed between [0,1]. This is the best historical position found by the particle itself. This represents the globally optimal position found across the entire particle swarm. This represents the current position of the particle.

[0093] The algorithm's inertia weights adopt an adaptive linear decreasing rule, as shown in the following formula (16), which takes into account both the early global search and the later local convergence: (16) In the formula, , These are the upper and lower limits of the inertia weight. , This represents the current iteration number and the maximum iteration number.

[0094] After each particle position iteration, dynamic physical-level projection repair must be performed to forcibly match the total grid connection and physical boundary constraints. The repair logic is as follows: First, verify the sum of the power allocated to each energy storage unit output by the current particle. If the total power is not equal to the current total power deficit requirement of the system... Then, according to the current output margin ratio of each effective energy storage unit, the output components of each unit are forcibly balanced proportionally. During the adjustment process, if it is detected that the allocated power of an independent energy storage unit exceeds the dynamic output boundary hard constraint predicted in step two, the output of the unit is immediately limited to the dynamic boundary value, and the power difference generated by the cut-off is redistributed to the remaining available energy storage units that still have output margin, until the grid-connected power conservation requirements are fully met.

[0095] Step 4: Execute the lower-level compound tracking control based on the cerebellar model (CMAC) neural network.

[0096] After receiving the power command allocated by the upper layer, the local controller of each energy storage unit adopts a composite control scheme combining a piecewise nonlinear quantization CMAC neural network and a proportional controller for millisecond-level tracking. The control law is shown in the following formula (17): (17) In the formula, To execute control commands, The power command assigned to this unit by the upper layer. For CMAC feedforward dynamic compensation output, For proportional gain, This provides feedback on the actual output of the unit. The CMAC weights are updated online based on the tracking error, and a leakage term is introduced to suppress integral saturation. The combination of feedforward and proportional inputs eliminates the dynamic lag in the power electronic PCS execution loop.

[0097] The internal weight array of the cerebellar model neural network tracks the error according to the current instruction within the sampling step. Discrete online correction is performed, and a leakage term is introduced to suppress integral saturation. The discrete iterative update law of the grid weights is shown in the following formula (18): C(18) In the formula, To prevent leakage of saturation weight coefficients, The weights of the currently active grid. For feedforward online learning rate, To generalize the number of associative units, C represents the currently triggered set of associative memory spaces. Through online updates of the weights, the feedforward channel can offset the dynamic hysteresis of the converter execution process in real time.

[0098] Third embodiment: Figure 1 This is a block diagram of the overall structure of the "thermal power unit-flow battery" joint frequency regulation system. Figure 2 This is a block diagram of the mathematical model for the load frequency control subsystem of a thermal power unit. Figure 3This is a simulation structural block diagram of an aqueous organic flow battery energy storage subsystem. Figure 4 This is a comparison chart of the fitness evolution process between the upper-level P-AWPSO algorithm and the standard PSO algorithm. Figure 5 This is a comparison chart showing the convergence changes of the State of Charge (SOC) of each flow battery cell under random grid fluctuation conditions. Figure 6 This is a comparison chart of tracking errors under time-varying parameters. Figure 7 This is a comprehensive comparison chart of five dimensions of evaluation indicators for three types of working conditions.

[0099] The joint frequency modulation scenario in this embodiment is based on typical secondary frequency modulation requirements, and the specific simulation environment is built on the MATLAB R2024a and Simulink platforms. The continuous system uses a variable step size ode45 solver, the sampling step size Δt of the discrete coordination control link is 0.5 s, and the AGC instruction rolling allocation cycle is 8 s.

[0100] Implementation Example: Basic System Parameter Configuration The key parameters of the joint frequency modulation system in this embodiment are calibrated as follows: 1. The capacity of the thermal power unit is referenced to the 660 MW supercritical reheat condensing coal-fired unit of the Huaneng Luoyuan Power Plant hybrid energy storage frequency regulation demonstration project. Governor time constant. Take 0.20 s as the vapor volume time constant. Take 0.30 s as the reheat time constant. Taking 10.0 s as an example, the power ratio of the high-pressure cylinder is... Take 0.30, boiler thermal storage inertia The time constant is set to 8.0 s, the equivalent inertial time constant H is set to 5.0 s, and the load damping coefficient D is set to 1.00. The rated ramp rate of the unit is strictly calibrated to 0.242 MW / s (2.2%Pn / min).

[0101] 2. The aqueous organic redox flow battery energy storage system has a total power of 40 MW (≈6%Pn), consisting of 5 independent energy storage units, each with a rated power of 8 MW and a rated capacity of 3 MWh. The round-trip energy efficiency is set at 80.01% based on the average measured value in the laboratory, and the unidirectional efficiency of each unit is taken as the square root of the average value. The safe operating range of SOC is limited to 0.10 ~ 0.90, with an expected value of 0.50. The initial state of the 5 independent units simulates the non-uniformity in actual conditions, with initial SOCs set as follows: 0.55, 0.30, 0.65, 0.25, and 0.50, respectively. The pump loss and parasitic losses during single-unit operation are set at 0.16 MW, the time constant of the PCS converter execution loop is set at 0.20 s, and the converter efficiency is set at 98.5%.

[0102] In this embodiment, all the baseline parameters of the model are derived from the physical measurement data of the characteristic test of the laboratory aqueous organic redox flow battery: the round-trip energy efficiency η is taken as the measured average of 80.01% under 25℃ temperature control after 18 charge-discharge cycles (the square root of the average value of the unidirectional efficiency is taken); the self-discharge rate parameter is strictly equivalently discretized from the 24-hour voltage retention rate of 97.44% measured in the long-term open-circuit test; the initial temperature and characteristics of each stack module during operation are referenced from the measured range of the positive and negative electrode liquid inlet and outlet temperatures of 10 independent modules (23.9℃~28.4℃).

[0103] In this embodiment, the rolling allocation cycle of AGC commands is set to 8 seconds, which, in legal terms, is precisely consistent with the physical issuance rhythm of AGC secondary frequency regulation commands under the two detailed rules of typical regional power grids; the sub-objective weights in the multi-objective comprehensive optimization function are adaptively set to command tracking. =0.50, lifespan loss =0.15, SOC consistency =0.25, economic benefits =0.10, this set of weights was determined through rigorous orthogonal experiment optimization, prioritizing the physical guarantee of the system's second-level high frequency modulation accuracy, while also taking into account the SOC consistency balance among the internal units and the economy throughout the entire life cycle.

[0104] In operating condition 3, the battery internal resistance drift coefficient The capacity decay coefficient is set to 0.35 (the internal resistance at the end of operation is 1.35 times the initial value). A value of 0.04 (the final usable capacity is 0.96 times the initial value) was used to verify the robustness of the strategy after battery aging.

[0105] Detailed control of the iteration timing process: After the dispatching side issues the secondary AGC frequency adjustment command, the system operates in a closed loop according to the following process: 1. Unit Output Status Acquisition and Shortfall Extraction: The thermal power unit subsystem first responds to the AGC command. Faced with a sudden step command with an amplitude of +25 MW (condition 1), the thermal power unit is constrained by a slow ramp rate of 2.2%Pn / min, resulting in a significant lag in output response in the initial few seconds. The system uses the difference requirement module to calculate the difference between the AGC command and the actual output of the generator unit in real time, extracting the total shortfall requirement that needs to be quickly compensated by the aqueous flow battery. .

[0106] 2. Dynamic Prediction and Limitation of Energy Storage Unit Output: The central power distribution mechanism for energy storage collects the real-time State of Charge (SOC) of each energy storage unit every 8 seconds. Based on the maximum power prediction model for the discharge direction defined by Formula 9, after determining that the current direction belongs to the discharge compensation direction, the maximum charge and discharge constraints of each unit in the future time domain are updated in real time. For example, in this embodiment, since the initial SOC of unit 4 (0.25) is relatively close to the lower operating limit, its maximum output upper limit will be actively linearly reduced according to Formula 9, providing an adaptive protection boundary for upper-level optimization output.

[0107] 3. Upper-level Particle Swarm Optimal Power Allocation (P-AWPSO): In the allocation pool, if the predicted power boundary of a certain cell (such as a cell that has reached the safety lower limit) in the current direction is 0, a priority screening mechanism is triggered to directly remove it from the optimization calculation of the current cycle. For the remaining cells with output margin, an improved particle swarm optimization is initiated. The weights of the multi-objective function are set to command tracking. =0.50, lifespan loss =0.15, SOC consistency =0.25, economic benefits =0.10. The population size is set to 25, and the maximum number of iterations is 40. The inertia weight w of the particles follows (Equation 16), adaptively decreasing from 0.9 to 0.4 as the iteration progresses. Projection repair is performed in each optimization iteration to forcibly match the total grid connection. Finally, at the end of the rolling cycle, the optimal power command components of the 5 energy storage units are calculated and issued. .

[0108] 4. Millisecond-level power composite tracking control of the lower-level converter: The local execution controllers of the five energy storage units receive their respective optimized allocation commands in real time. Then, the CMAC neural network feedforward controller is triggered locally. The CMAC neural network quantization level is set to 100, the number of generalization units is set to 8, the online learning rate is set to 0.10, and the proportional feedback gain is set to... Set to 0.6. Within each fast sampling step of 0.5 seconds, CMAC calculates the feedforward dynamic compensation value used to offset the first-order execution delay (time constant 0.20 s) of the converter PCS through the normalized input. The compensation value is directly superimposed with the proportional feedback result to obtain the execution control law. (Formula 17) The actual power of the driving flow battery precisely matches the command without significant lag or overshoot. .

[0109] The actual output of the five aqueous organic flow battery units is ultimately physically superimposed with the output of the thermal power unit at the grid connection node, resulting in a highly precise combined total output that completely eliminates the reverse regulation and regulation deviation of traditional thermal power units. Based on 1800 seconds of random continuous operating condition testing, the method of this invention successfully achieved a comprehensive tracking qualification rate of 99.97%, with a frequency modulation K-value as high as 0.991. Furthermore, it successfully converged the unbalanced state of charge of the units to below a standard deviation of 0.065 at the end of operation, achieving the optimal balance between technical performance, safety boundary management, and economic benefits throughout the entire life cycle.

[0110] Comparative analysis of strategy defects through quantitative demonstration and robust data verification: To fully demonstrate the substantial and significant advancements of the method of this invention under multi-condition random frequency regulation, two industry-standard strategies were introduced to conduct a continuous 1800-second parallel test of random power grid fluctuations under identical hardware and initial environmental parameters: (1) Quantitative verification shows that if the existing fixed boundary average allocation strategy is adopted, it cannot achieve fine perception of the differences in the state of charge of each energy storage unit. As a result, after the system has been running continuously for 1800s, the standard deviation of SOC among the energy storage units is still stubbornly maintained at the initial severe imbalance of 0.169. This imbalance triggers a chain reaction of deterioration, causing about 20% of the participating units to reach the safe operating hard limit of 0.10 or 0.90 in advance. Due to the electrochemical stress aging damage caused by deep charging and deep discharging, they are forced to withdraw from grid regulation in advance. At the same time, if the lower-level control adopts the traditional PI pure feedback control (such as proportional 0.7, integral 0.3), there is obvious phase lag and reverse regulation when facing rapid fluctuation AGC commands. Its tracking root mean square error is as high as 1.423 MW, and the comprehensive K value for frequency regulation assessment is only 0.690, which cannot meet the qualification rate requirements of modern high-standard power grid for frequency regulation ancillary services.

[0111] (2) In contrast, because the present invention embeds a first-order equilibrium reward function in the upper-level allocation, the five energy storage units in a severely unbalanced state spontaneously converge to the healthy mean during the operation process, and the standard deviation of the end SOC drops to 0.065 (which is significantly reduced by about 61.5% compared to the average allocation strategy). This successfully protects each stack in a good operating range close to 0.50, thereby reducing the overall battery depreciation cost from RMB 1,748 in the average allocation to RMB 1,638 (a reduction of about 6.3%). At the same time, the introduction of the lower-level CMAC feedforward reduces the root mean square error of the system's random tracking to 0.165 MW, and the regulation accuracy and comprehensive frequency regulation K value climb to 0.991. Finally, the comprehensive mileage benefit of frequency regulation is as high as RMB 8,938 (an increase of 1.0% compared to the average allocation strategy).

[0112] (3) In order to further conduct a rigorous robustness limit test (condition 3) on the present invention, a severe nonlinear parameter drift caused by battery aging was forcibly introduced into the random fluctuation command: The amplitude is trend-mapped based on the measured cycle life decay law of the system (5.33% decay every 20,000 cycles), and the internal resistance degradation drift is set to 1.35 times the initial reference value in the later stage. =0.35), usable capacity reduced by 4% ( =0.04). Test results show (e.g.) Figure 6 As shown in the figure, under such harsh time-varying operating conditions, thanks to the real-time refresh perception of the dynamic output boundary by the upper-layer rolling optimization and the disturbance resistance of the lower-layer network self-learning feedforward, the root mean square error of the system's command tracking remains stable at 0.165 MW, the frequency modulation K value remains stable at a high level of 0.991, and the tracking error waveform is firmly anchored within ±0.5MW, with no significant performance degradation, demonstrating excellent robustness in engineering field and a solid creative barrier.

[0113] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0114] Those skilled in the art will understand that the modules in the apparatus of the embodiments can be distributed in the apparatus of the embodiments as described in the embodiments, or they can be located in one or more devices different from this embodiment with corresponding changes. The modules of the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for frequency regulation of a thermal power unit assisted by a water-based organic flow battery with dual-layer coordinated control, characterized in that, The method constructs a two-layer coordinated control architecture that combines upper-layer power optimization allocation with lower-layer precise power tracking, and includes the following steps: S1: Construct a time-varying mathematical model for a joint frequency regulation system that integrates a thermal power unit load frequency control sub-model with a nonlinear dynamic recursive sub-model of an aqueous organic flow battery that takes into account multiple time-varying parameters; S2: Construct a dynamic output boundary prediction mechanism, collect the real-time state of charge of each energy storage unit in the array, and predict the dynamic output upper and lower limits of each independent energy storage unit in the power-time dimension in the future control time domain. S3: Execute upper-level power optimization allocation, calculate in real time the difference between the total AGC command and the current actual output of the thermal power unit to obtain the current total power deficit requirement of the joint frequency regulation system; based on the dynamic output upper and lower limit boundaries output in step S2, prioritize each energy storage unit in the array, eliminate energy storage units without adjustment capability in the current charging and discharging direction to determine the effective participants in this rolling process; construct a multi-objective optimization function that takes into account command tracking, stack life loss, and active consistency balance of state of charge, and use an improved particle swarm optimization algorithm to solve the power allocation components of the effective participants in a rolling process; S4: Executes precise local power tracking control at the lower level. After receiving the optimal power allocation command, the local controller of each independent energy storage unit triggers an adaptive composite control law that introduces feedforward dynamic self-learning compensation. This law approximates the nonlinear inverse dynamics of the converter stage online, eliminates feedback execution lag, and drives the actual output of each energy storage unit to match the allocation command.

2. The method for frequency regulation of a hydro-organic flow battery-assisted thermal power unit with dual-layer coordinated control according to claim 1, characterized in that, In step S3, after each position iteration of the particle swarm optimization algorithm, a dynamic physical-level projection repair mechanism is executed to match the total grid-connected power and physical hard constraints: First, the sum of the allocated power of each energy storage unit output in the current iteration is checked. If the total amount is not equal to the total power deficit requirement, the output components of each unit are forcibly adjusted proportionally according to the current output margin ratio of each effective energy storage unit. During the adjustment process, if it is detected that the allocated power of an independent energy storage unit exceeds the dynamic output boundary hard constraint predicted in step S2 at that time node, the output of that unit is immediately limited to the dynamic boundary value, and the truncated difference power is re-distributed to the remaining available energy storage units that still have output margin, until the total grid-connected power is completely balanced. When the optimization ends, the optimal power allocation command for each effective energy storage unit is output.

3. The method for frequency regulation of a hydroelectric organic flow battery-assisted thermal power unit with dual-layer coordinated control according to claim 1, characterized in that, In step S1, the thermal power unit load frequency control sub-model includes the thermal power unit load frequency control time-domain continuous state propagation equation as shown in the following formula: Its mechanical power output Satisfying the relation: Actual output of the unit for: ; In the formula, - These are the internal state quantities of the speed governor, the high-pressure cylinder steam capacity, the reheat circuit, and the boiler's thermal storage inertia, respectively. For unit power command, The time constants for each stage, This represents the power percentage of the high-pressure cylinder. Meanwhile, the change in active power command within each sampling step is subject to a hard limit on the discrete differential ramp rate: In the formula, , Let represent the power command values ​​of the th energy storage unit at time k-1 and time k, respectively. This indicates the maximum allowable power adjustment step size within a single control cycle. The rated ramp rate of the unit, This represents the discrete sampling step size.

4. The method for frequency regulation of a hydroelectric organic flow battery-assisted thermal power unit with dual-layer coordinated control according to claim 1, characterized in that, In step S1, the nonlinear dynamic recursive sub-model of the aqueous organic flow battery includes: The actual active power response of the energy storage unit satisfies the first-order inertial dynamic response equation: In the formula, This is an on-site action command. The actual output power of the action. The active power execution time constant of the converter stage, and the active power converted by electrochemical conversion within the energy storage unit. Satisfying the nonlinear efficiency conversion rule of charge and discharge segments: when hour ;when hour, In the formula, For unidirectional efficiency, The state of charge of each energy storage unit satisfies the discrete recursive equation: In the formula, This represents the actual usable capacity. To account for self-discharge loss, the second-level self-discharge rate constant, , These represent the real-time state of charge of the i-th energy storage unit at the k-th and k+1-th control times, respectively. The terminal voltage of each energy storage unit satisfies the polarization time-varying equation: In the formula, This represents the real-time terminal voltage of the i-th energy storage unit at time t. Open circuit voltage, For loop current, For time-varying internal resistance, The long-term time-varying degradation of energy storage units satisfies the drift equation: In the formula, and The initial reference internal resistance and rated capacity, This is the internal resistance drift coefficient. The coefficient representing the change in capacity lifetime decay is T, where T is the sampling period. The unit-integrated active power loss equation satisfies: In the formula, To maintain constant parasitic pump loss power, This is the rated output power.

5. The method for frequency regulation of a hydroelectric organic flow battery-assisted thermal power unit with dual-layer coordinated control according to claim 1, characterized in that, In step S2, the maximum power prediction model for the discharge direction is: In the formula, This represents the maximum allowable discharge power of the energy storage unit under its current state. The lower limit of SOC safe operation is defined as m, where m is the width of the output retreat zone, and the maximum power in the charging direction is determined symmetrically based on the upper limit of safety.

6. The method for frequency regulation of a hydro-organic flow battery-assisted thermal power unit with dual-layer coordinated control according to claim 1, characterized in that, In step S3, the multi-objective optimization function F is a four-objective weighted function: In the formula, , , , Assign weight coefficients to each sub-objective. , , , Each corresponds to a command to track sub-targets Lifetime loss sub-target State of Charge (SOC) Consistency Sub-Objective Sub-objective of comprehensive economic benefits of frequency modulation throughout the entire life cycle ; The State of Charge (SOC) consistency sub-objective is guided by a first-order equilibrium reward function for multi-objective solution. In the formula, This represents the real-time state of charge of the i-th energy storage unit. The average state of charge of the current participating regulation unit is denoted as n, and the number of energy storage units participating in regulation is denoted as n.

7. The method for frequency regulation of a hydro-organic flow battery-assisted thermal power unit with dual-layer coordinated control according to claim 1 or 6, characterized in that, The combination of weight coefficients in the multi-objective optimization function is determined through offline orthogonal experiment optimization, and its set values ​​are: instruction tracking weights. =0.50, lifespan loss weight =0.15, SOC consistency weight =0.25, economic benefit weight =0.10; The rolling optimization update cycle of the upper-layer power optimization allocation is physically consistent with the discrete issuance communication cycle of the automatic generation control command of the external regional power system, and is set to 8s.

8. The method for frequency regulation of a hydroelectric organic flow battery-assisted thermal power unit with dual-layer coordinated control according to claim 1, characterized in that, In step S4, the composite control law of the lower-level cerebellar model CMAC neural network is: In the formula, To execute local operation commands sent to the PCS converter. The optimal power allocation command issued by the upper layer for optimization. This is the feedforward dynamic compensation term output by the CMAC neural network after online self-learning following multi-segment nonlinear quantization and associative grid mapping. For the control gain of the local proportional controller, This is the current actual output feedback of the energy storage unit.

9. The method for frequency regulation of a hydroelectric organic flow battery-assisted thermal power unit with dual-layer coordinated control according to claim 1 or 4, characterized in that, In step S1, the nonlinear dynamic recursive sub-model of the energy storage system explicitly incorporates the parameter degradation characteristics during the long-term continuous frequency regulation service process. In the discrete-state recursion, the internal parameters of its mathematical model include the time-varying internal resistance function R(t) of each individual cell, which exhibits a time-varying drift law that increases or decreases with the total duration of long-term frequency regulation operation, and the time-varying decay function of the actual usable capacity of each individual cell. The unit's time-varying internal resistance function R(t) and the actual usable capacity's time-varying decay function... The parameter drift amplitude coefficients are pre-calibrated and configured according to the cycle life decay and internal resistance increase trend of the physical battery, so as to apply the dynamic output protection boundary of the frequency modulation control system during long-term operation in steps S2 and S3.

10. A frequency regulation control system for a water-based organic flow battery-assisted thermal power unit with dual-layer coordinated control for implementing the method of any one of claims 1-9, characterized in that, The control system includes: Data acquisition terminal: used to collect in real time the AGC total target command of the regional power system, the current actual response output of thermal power units, and the real-time state of charge, terminal voltage and electrolyte parameters of each independent energy storage unit in the water system organic liquid flow battery array; The core of the central frequency regulation power coordination control: receives the data collected by the data acquisition terminal, embeds the time-varying mathematical model of the joint frequency regulation system, predicts the power-time dimension output boundary constraints of each unit based on the safety window, and outputs the optimal power allocation command of each effective energy storage unit by executing the improved particle swarm optimization algorithm and the physical projection repair mechanism. Energy storage unit local power tracking controller group: It is communicatively connected to the central frequency regulation power coordination control core. Each local power tracking controller controls a PCS power electronic converter. The local power tracking controller is embedded with a cerebellum model neural network composite tracking control law, which drives the actual output of the corresponding energy storage unit to follow the optimal power allocation command at the millisecond level.

Images