Method for adjusting frequency of power grid containing energy storage based on rolling optimization

Abstract

The invention relates to the technical field of power systems and automation thereof, discloses an energy-storage-containing power grid frequency regulation method based on rolling optimization, and aims to solve the problem that the existing rolling optimization strategy neglects the energy storage charge state constraint and life attenuation characteristics and does not fully consider the influence of communication delay and prediction errors. The control performance is reduced; the stored energy is overused; or the regulation capability is unstable. The method comprises the following steps: acquiring real-time operation data of a power grid, and constructing an improved equivalent system model containing inertia dynamics, a frequency modulation dead zone and energy storage nonlinear efficiency characteristics; model parameters and constraint boundaries are dynamically adjusted through a closed-loop feedback correction mechanism, and self-adaptive cooperative regulation and control are achieved in combination with energy storage health state evaluation and multi-stage safety protection. Through the scheme, the frequency adjustment precision and robustness can be improved, the energy storage service life is prolonged, the millisecond-level real-time response requirement of a power grid is met, and the system stability in a high-proportion renewable energy source scene is enhanced.

Description

Technical Field

[0001] This invention relates to the field of power systems and their automation technology, specifically to a method for frequency regulation of a power grid with energy storage based on rolling optimization. Background Technology

[0002] With the continuous increase in the penetration rate of renewable energy in new power systems, grid frequency stability faces severe challenges. The high proportion of intermittent power sources such as wind and solar power leads to a decrease in system inertia and a weakening of disturbance immunity, resulting in more frequent and larger frequency fluctuations. Against this backdrop, energy storage systems, due to their rapid response and bidirectional regulation capabilities, are widely regarded as a key means to support grid frequency stability. The core of frequency regulation lies in the real-time balancing of the power difference between generation and load, and grids containing energy storage systems need to achieve precise, efficient, and robust dynamic control under complex operating conditions.

[0003] Among them, the frequency regulation method based on rolling optimization solves the optimal control problem in a finite time domain periodically, which can take into account both the current state and future trends of the system and effectively improve regulation performance. This method usually aims to minimize frequency deviation or control cost, updates the optimization window at each sampling time, and executes control commands in a rolling manner to adapt to the time-varying characteristics of power grid operating conditions.

[0004] However, existing rolling optimization strategies still have significant shortcomings when applied to grids with energy storage. On the one hand, traditional models often neglect the state-of-charge constraints and lifetime decay characteristics of energy storage devices, which can easily lead to overuse of energy storage or a sharp drop in regulation capacity during frequent charging and discharging. On the other hand, the optimization process often uses simplified linear system models, which are difficult to accurately characterize the strong nonlinear dynamic behavior under high-proportion renewable energy integration, causing control commands to deviate from actual needs. In addition, existing methods generally do not fully consider the impact of communication delays and prediction errors on state estimation within the rolling window, resulting in conservative or unstable optimization results. These problems are particularly prominent when the grid is subjected to high-power disturbances or extreme operating conditions, severely restricting the potential of energy storage systems in frequency regulation. There is an urgent need for a rolling optimization regulation method that balances model accuracy, constraint integrity, and real-time robustness. Summary of the Invention

[0005] The purpose of this invention is to provide a frequency regulation method for power grids with energy storage based on rolling optimization. This method can effectively solve the technical problems mentioned in the background art, such as the decline in control performance, overuse of energy storage, or instability of regulation capability when the existing rolling optimization strategy is applied to power grids with energy storage. These problems arise because the existing rolling optimization strategy ignores the energy storage state of charge constraints and lifetime decay characteristics, uses a simplified linear model that is difficult to characterize nonlinear dynamic behavior, and does not fully consider the impact of communication delay and prediction error.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A frequency regulation method for power grids with energy storage based on rolling optimization includes the following specific steps:

[0008] Step 1: Acquire real-time power grid operation data, collect the current system frequency, total load power, renewable energy output, state of charge of each energy storage unit and its historical charge and discharge sequence, and synchronously acquire key node voltage and phase angle information through a wide-area measurement system to form a real-time dataset containing dynamic state variables;

[0009] Step 2: Construct a nonlinear prediction model. Based on the real-time dataset, an improved equivalent system model incorporating inertia dynamics, frequency modulation dead zone, and energy storage nonlinear efficiency characteristics is adopted. A multi-objective optimization function is established with the objectives of minimizing the integral of frequency deviation and the weighted cost of energy storage lifetime loss. The model parameters are identified and updated online based on the current system inertia level and renewable energy penetration rate. The multi-objective optimization function is expressed as:

[0010]

[0011] in, and For dynamic weighting coefficients, To predict the frequency deviation in the time domain, where T is the prediction time domain length, The number of energy storage units. For the first The equivalent lifetime loss cost of an energy storage unit; Step 3: Set a rolling optimization window and constraints. For each control cycle, define a prediction time domain of 30 seconds. Use the upper and lower limits of the energy storage unit's state of charge (SOC), maximum charging / discharging power, cumulative loss from charging / discharging cycles, and the state prediction deviation introduced by communication delay as hard constraints to construct a constrained optimal control problem. The dynamic evolution of the SOC satisfies the following:

[0012]

[0013] in, Let t be the state of charge. For current-related charge and discharge efficiency, For energy storage power command, For rated capacity, To control the cycle;

[0014] Step 4: Perform robust rolling optimization solution. The piecewise linearization solution strategy based on model predictive control framework is adopted to decompose the nonlinear optimization problem into multiple linear subproblems. In each sub-time domain, the solution is solved quickly and iteratively to output the optimal power command sequence in the next 10 seconds, and only the command at the first moment is executed.

[0015] Step 5: Implement closed-loop feedback correction. Compare the deviation between the actual frequency response after execution and the predicted trajectory. If the deviation exceeds the set threshold of 5%, trigger the model parameter re-identification and optimization window reset mechanism to dynamically adjust the prediction model gain and constraint boundary, thereby realizing online adaptive correction of the control strategy.

[0016] Preferably, in step 1, the state of charge of the energy storage unit is acquired using the ampere-hour integration method combined with open-circuit voltage calibration, with a sampling period of 100 milliseconds. The calibration trigger condition is that the energy storage is in a static state for a duration of more than 300 seconds, and the state of charge error after calibration is less than 1%.

[0017] Preferably, in step 2, the improved equivalent system model includes third-order dynamic equations, which describe the system frequency change rate, equivalent inertia response, and speed controller operation process, respectively. The energy storage charging and discharging efficiency is modeled as a quadratic function of current intensity, in the form η=0.98-0.02×(I / Imax)², which is used to reflect the characteristics of increased energy loss under high current conditions.

[0018] Preferably, in step 2, the weight coefficients of the multi-objective optimization function are dynamically configured according to the power grid operation scenario. During periods of high wind power output, the weight of the energy storage life loss term is increased to 0.7, and during peak load periods, the weight of the frequency deviation term is increased to 0.8, ensuring that the control objectives match the operational requirements.

[0019] Preferably, in step 3, the cumulative loss of energy storage cycle is calculated in real time using the rainflow counting method. After each complete charge-discharge cycle, the equivalent loss factor is updated based on the current state of charge change and temperature conditions, and converted into equivalent cost and incorporated into the optimization objective to prevent frequent local adjustments from causing a sudden drop in lifespan.

[0020] Preferably, in step 3, the state prediction deviation range introduced by communication delay is determined based on historical communication delay statistics, the maximum delay is set to 200 milliseconds, the corresponding state prediction compensation adopts the forward extrapolation method, and is corrected by combining the frequency change rate trend, and the state estimation error after compensation is controlled within 0.01 Hz.

[0021] Preferably, in step 4, the piecewise linearization solution strategy divides the 30-second prediction time domain into 6 5-second sub-intervals, performs a Taylor expansion on the nonlinear model in each sub-interval, retains the first-order terms and fixes the coefficients of the second-order terms, and completes a single solution within 10 milliseconds using the interior point method, ensuring that the generation of control commands meets the real-time requirements.

[0022] Preferably, in step 4, after the optimal power command sequence is generated, a smoothness constraint is applied to limit the rate of change between adjacent commands to no more than 80% of the maximum power ramp-up rate of the energy storage, so as to avoid the mechanical stress impact on the energy storage device caused by sudden power changes.

[0023] Preferably, in step 5, the frequency response deviation comparison uses the sliding window root mean square error index to calculate the root mean square deviation between the predicted frequency and the measured frequency within the next 5 seconds. When the index exceeds 5% for two consecutive control cycles, the model re-identification process is initiated to re-estimate the system's equivalent inertia and damping coefficient.

[0024] Preferably, it also includes: establishing an energy storage health status assessment module, which uses an empirical decay model to predict the remaining cycle life of each energy storage unit based on the historical charge-discharge sequence and temperature data, and automatically switches it from the main frequency regulation echelon to the standby echelon when the predicted life is less than 3000 cycles, giving priority to calling energy storage units with good health status.

[0025] Preferably, it also includes: setting up a multi-level safety protection mechanism, which immediately initiates emergency load shedding and fast frequency modulation response coordination when the frequency deviation is detected to exceed ±0.5 Hz, while locking the rolling optimization output and switching to the preset fixed gain control mode to ensure the safety and stability of the system.

[0026] Preferably, the method is integrated into the regional power grid energy management system, and achieves millisecond-level data interaction with each energy storage power station through an optical fiber communication network. The control command transmission delay is less than 50 milliseconds, the overall system response time is less than 200 milliseconds, and the time for a single optimization calculation does not exceed 15 milliseconds.

[0027] Preferably, the method is applicable to hybrid energy storage systems that include electrochemical energy storage, flywheel energy storage and supercapacitors. Differentiated model parameters and constraint boundaries are set for the dynamic characteristics of different types of energy storage to achieve coordinated optimization scheduling of multiple types of energy storage, with an overall frequency regulation response accuracy of over 98%.

[0028] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0029] This invention constructs an improved equivalent system model incorporating the nonlinear efficiency and lifetime loss characteristics of energy storage, combined with a multi-objective rolling optimization framework. This significantly reduces unnecessary charging and discharging cycles of the energy storage system while ensuring frequency regulation accuracy, thus extending its lifespan. By introducing a communication delay compensation mechanism and a closed-loop feedback correction strategy, the robustness of the optimized control in real-world communication environments is improved, effectively suppressing control instability caused by state estimation bias. A piecewise linearization fast solution algorithm is employed, achieving millisecond-level control command generation while maintaining nonlinear modeling accuracy, meeting the real-time dynamic regulation requirements of the power grid. Through dynamic weight configuration and health status assessment mechanisms, adaptive matching between control objectives and system operating states is achieved, enhancing the applicability and economy of the method in different operating scenarios. The overall technical solution significantly improves the frequency regulation capability and energy storage utilization efficiency of power grids with a high proportion of renewable energy without increasing hardware investment, providing reliable technical support for the safe and stable operation of new power systems. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0031] Figure 2 This is a schematic diagram of the core principle framework of multi-objective rolling optimization that integrates energy storage nonlinear efficiency and lifetime loss characteristics in this invention;

[0032] Figure 3 This is a flowchart illustrating the main logical flow of the invention, from real-time data acquisition to robust rolling optimization solution.

[0033] Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow between the regional power grid energy management system and various types of energy storage units in this invention. Detailed Implementation

[0034] 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.

[0035] Example 1

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0037] Currently, with the continuous increase in the penetration rate of renewable energy in new power systems, grid frequency stability faces severe challenges. The high proportion of intermittent power sources such as wind and solar power leads to a decrease in system inertia and a weakening of disturbance immunity, resulting in more frequent and larger frequency fluctuations. Against this backdrop, energy storage systems, due to their rapid response and bidirectional regulation capabilities, are widely regarded as a key means to support grid frequency stability. The core of frequency regulation lies in the real-time balancing of the power difference between generation and load, and grids containing energy storage systems need to achieve precise, efficient, and robust dynamic control under complex operating conditions. To address the aforementioned technical problems, this invention proposes a rolling optimization-based frequency regulation method for power grids with energy storage. By constructing an improved equivalent system model incorporating the nonlinear efficiency and lifetime loss characteristics of energy storage, and combining it with a multi-objective rolling optimization framework, the method significantly reduces unnecessary charging and discharging cycles of the energy storage system while ensuring frequency regulation accuracy, thus extending its lifespan. By introducing a communication delay compensation mechanism and a closed-loop feedback correction strategy, the robustness of the optimized control in real-world communication environments is improved, effectively suppressing control instability caused by state estimation bias. A piecewise linearization fast solution algorithm is employed to achieve millisecond-level control command generation while ensuring the accuracy of nonlinear modeling, meeting the real-time dynamic regulation requirements of the power grid. Through dynamic weight configuration and health status assessment mechanisms, adaptive matching between the control objective and the system operating state is achieved, enhancing the applicability and economy of the method under different operating scenarios. This method is applied to a rolling optimization-based frequency regulation method for power grids with energy storage.

[0038] refer to Figure 1 The overall technical architecture of this invention comprises four core functional modules: a data acquisition layer, a model building layer, an optimization solution layer, and an execution feedback layer. The data acquisition layer is responsible for acquiring real-time power grid operation data; the model building layer establishes a nonlinear prediction model and a multi-objective optimization function based on the real-time data; the optimization solution layer performs rolling optimization calculations under set constraints; and the execution feedback layer is responsible for issuing commands and performing closed-loop correction. These modules are integrated through a regional power grid energy management system to form a closed-loop control circuit.

[0039] refer to Figure 3 The main logical process framework from real-time data acquisition to robust rolling optimization solution includes steps 1 to 5. The following will strictly follow the step sequence of the invention method, combined with accompanying drawings and preferred details, to provide a highly engineered explanation of each step.

[0040] In the aforementioned rolling optimization-based frequency regulation method for power grids with energy storage, step 1 involves acquiring real-time power grid operation data, including the current system frequency, total load power, renewable energy output, state of charge (SOC) of each energy storage unit, and its historical charge-discharge sequences. Key node voltage and phase angle information are simultaneously acquired through a wide-area measurement system, forming a real-time dataset containing dynamic state variables. Specifically, in step 1, the system frequency is acquired in real-time by phasor measurement units deployed in substations at a sampling rate of 50 frames per second, with a frequency measurement accuracy better than 0.001 Hz. The total load power and renewable energy output data are sourced from the SCADA (Supervisory Control and Data Acquisition) system, with an update cycle of 1 second. The SOC of each energy storage unit is calibrated using an ampere-hour integration method combined with open-circuit voltage calibration. The ampere-hour integration accumulates the charge-discharge current with a sampling period of 100 milliseconds. The open-circuit voltage calibration is triggered when the energy storage is in a static state for a duration greater than 300 seconds, resulting in a SOC error of less than 1% after calibration. Historical charge-discharge sequences are stored at 10-second intervals, containing power commands and actual power data from the most recent 24 hours, for subsequent lifetime loss calculations. The voltage and phase angle information of key nodes synchronously acquired by the wide-area measurement system are aligned to microsecond-level time using the IEEE 1588 precision time protocol, ensuring spatiotemporal consistency of multi-source data. All acquired data is transmitted to the regional power grid energy management system via an optical fiber communication network with a transmission delay of less than 50 milliseconds, forming a complete real-time dataset that serves as the basic input for subsequent model construction.

[0041] In the aforementioned frequency regulation method for power grids with energy storage based on rolling optimization, step 2 involves constructing a nonlinear prediction model. Based on the real-time dataset, an improved equivalent system model incorporating inertia dynamics, frequency regulation dead zone, and nonlinear efficiency characteristics of energy storage is adopted. A multi-objective optimization function is established with the objectives of minimizing the integral of frequency deviation and the weighted cost of energy storage lifetime loss. The model parameters are identified and updated online based on the current system inertia level and renewable energy penetration rate. Specifically, in step 2, the improved equivalent system model includes third-order dynamic equations, which describe the system frequency change rate, equivalent inertia response, and governor operation process, respectively. The first-order equation characterizes the relationship between the frequency change rate and the net unbalanced power, in the form of… , in For the equivalent system inertia identified online, Fluctuations in renewable energy output For load disturbance, For the first The power command for each energy storage unit; the second-order equation describes the first-order inertial response of the speed governor, with the time constant dynamically adjusted according to the proportion of thermal power units; the third-order equation introduces the frequency regulation dead zone characteristic, where traditional units do not participate in regulation when the absolute value of the frequency deviation is less than 0.033 Hz. The charging and discharging efficiency of energy storage is modeled as a quadratic function of current intensity, in the form of... This is used to reflect the accelerated energy loss characteristics under high current operating conditions. Model parameters The damping coefficient DD is identified online using a recursive least squares method. The identification window length is 10 seconds, and it is updated every 5 control cycles. The multi-objective optimization function is expressed as:

[0042]

[0043] in, and For dynamic weighting coefficients, To predict the frequency deviation in the time domain, TT is the prediction time domain length. The number of energy storage units. For the first The equivalent lifetime loss cost of each energy storage unit. Weighting coefficients are dynamically configured based on grid operation scenarios. During periods of high wind power output, the weight of the energy storage lifetime loss term is increased to 0.7, and during peak load periods, the weight of the frequency deviation term is increased to 0.8, ensuring that control objectives match operational needs. Equivalent lifetime loss cost. The result is obtained by weighting the number of cycles and depth in real time using the rainflow counting method. The specific calculation will be detailed in step 3.

[0044] In the aforementioned rolling optimization-based frequency regulation method for power grids with energy storage, step 3 involves setting a rolling optimization window and constraints. A prediction time domain of 30 seconds is defined for each control cycle. The upper and lower limits of the energy storage unit's state of charge, maximum charging and discharging power, cumulative losses from charging and discharging cycles, and the state prediction deviation range introduced by communication delay are used as hard constraints to construct a constrained optimal control problem. Specifically, in step 3, the prediction time domain TT is fixed at 30 seconds, and the control cycle... The time interval is 1 second, and the rolling step size is 1 second. The dynamic evolution of the state of charge (SOC) satisfies:

[0045]

[0046] in, For current-related charge and discharge efficiency, For energy storage power command, For rated capacity, To control the cycle;

[0047] The State of Charge (SOC) upper and lower limits are set at 20% and 90% respectively to prevent overcharging and over-discharging. Maximum charge / discharge power constraints are differentiated according to energy storage type: 1C for electrochemical energy storage, 2C for flywheel energy storage, and 5C for supercapacitors. Cumulative loss from charge / discharge cycles is calculated in real-time using the rainflow counting method. This method extracts extreme points and identifies half-cycles in historical charge / discharge sequences. After each complete charge / discharge cycle, the equivalent loss factor is updated based on the current state of charge change and temperature conditions, and this is converted into equivalent cost and incorporated into the optimization objective to prevent frequent local adjustments from causing a sudden drop in lifespan. The state prediction deviation range introduced by communication delay is determined based on historical communication delay statistics. The maximum delay is set to 200 milliseconds. The corresponding state prediction compensation uses a forward extrapolation method, combined with the frequency change rate trend for correction. After compensation, the state estimation error is controlled within 0.01 Hz. This deviation range serves as the uncertainty set of the frequency prediction value, embedded into the constraint conditions in a robust optimization form to ensure that the system still meets stability requirements under worst-case delay conditions.

[0048] In the aforementioned frequency regulation method for power grids with energy storage based on rolling optimization, step 4 involves robust rolling optimization. A piecewise linearization solution strategy based on a model predictive control framework is employed to decompose the nonlinear optimization problem into multiple linear subproblems. Rapid iterative solutions are performed within each sub-time domain, outputting the optimal power command sequence for the next 10 seconds, and executing only the command at the first moment. Specifically, in step 4, the piecewise linearization solution strategy divides the 30-second prediction time domain into six 5-second sub-intervals. Within each sub-interval, a Taylor expansion of the nonlinear model is performed, retaining the first-order terms and fixing the coefficients of the second-order terms, thus transforming the original non-convex problem into a series of convex quadratic programming subproblems. Each subproblem is solved in a single iteration within 10 milliseconds using the interior-point method, ensuring that the control command generation meets real-time requirements. The solver uses sparse matrix technology to accelerate computation, keeping memory usage within 50 megabytes. The output optimal power command sequence consists of one power value per second for the next 10 seconds, totaling 10 command points. After generation, smoothness constraints are applied to limit the rate of change between adjacent instructions to no more than 80% of the maximum power ramp rate of the energy storage system. For example, for a 1 MW / 1 MWh lithium battery system with a maximum ramp rate of 1 MW / s, the change between adjacent instructions should not exceed 0.8 MW, thus avoiding mechanical stress impacts on the energy storage device caused by sudden power changes. Finally, only the first instruction in the sequence is executed, and the remaining instructions are used as predicted trajectories for the rolling optimization initialization of the next cycle.

[0049] In the aforementioned frequency regulation method for power grids with energy storage based on rolling optimization, step 5 involves implementing closed-loop feedback correction. This involves comparing the actual frequency response after execution with the predicted trajectory. If the deviation exceeds a set threshold of 5%, a model parameter re-identification and optimization window reset mechanism is triggered to dynamically adjust the prediction model gain and constraint boundaries, achieving online adaptive correction of the control strategy. Specifically, in step 5, the frequency response deviation comparison uses a sliding window root mean square error index to calculate the root mean square deviation between the predicted frequency and the measured frequency within the next 5 seconds. When this index exceeds 5% for two consecutive control cycles, the model re-identification process is initiated to re-estimate the system's equivalent inertia. With damping coefficient Simultaneously, the SOC constraint boundary is relaxed by 5% to enhance the adjustment margin, and the prediction time domain is shortened to 20 seconds to improve response speed. The corrected model parameters and constraints are immediately used for optimization calculations in the next cycle, forming a highly robust closed-loop control.

[0050] Furthermore, this invention also includes an energy storage health status assessment module, which predicts the remaining cycle life of each energy storage unit based on the historical charge-discharge sequence and temperature data using an empirical degradation model. This model considers the coupling effects of charge-discharge depth, rate, temperature, and cycle count, and takes the form of… ,in, To predict lifespan, For the initial lifespan, The average depth of discharge. To accumulate the equivalent number of iterations, To quantify the impact of factors such as depth of charge / discharge and cycle count on lifetime degradation, c is the temperature coefficient, and T is the average operating temperature. When the predicted lifetime is below 3000 cycles, it is automatically switched from the main frequency regulation echelon to the standby echelon, prioritizing the use of energy storage units in good health. (Reference) Figure 4 The regional power grid energy management system and various types of energy storage units achieve coordinated scheduling through multi-level interaction. The main frequency regulation echelon includes units with a health status score higher than 85, the standby echelon includes units with a score between 60 and 85, and units with a score lower than 60 enter maintenance status.

[0051] Furthermore, this invention includes a multi-level safety protection mechanism. When a frequency deviation exceeding ±0.5 Hz is detected, an emergency load shedding and rapid frequency modulation response are immediately initiated in coordination. Simultaneously, rolling optimization output is locked, and the system switches to a preset fixed-gain control mode to ensure system safety and stability. This fixed-gain mode employs a proportional-derivative control law, with the gain coefficient preset based on the system inertia, and a response delay of less than 100 milliseconds.

[0052] refer to Figure 2The core framework of multi-objective rolling optimization, which integrates the nonlinear efficiency and lifetime loss characteristics of energy storage, clearly demonstrates the organic integration of objective function construction, constraint handling, and solution strategies. The entire method is integrated into the regional power grid energy management system, achieving millisecond-level data interaction with each energy storage power station via fiber optic communication networks. The control command transmission delay is less than 50 milliseconds, the overall system response time is less than 200 milliseconds, and the time for a single optimization calculation does not exceed 15 milliseconds. The method is applicable to hybrid energy storage systems including electrochemical energy storage, flywheel energy storage, and supercapacitors. Differentiated model parameters and constraint boundaries are set for the dynamic characteristics of different types of energy storage: electrochemical energy storage focuses on SOC and lifetime constraints, flywheel energy storage focuses on speed and power constraints, and supercapacitors focus on voltage and efficiency constraints, achieving coordinated optimization scheduling of multiple types of energy storage with an overall frequency regulation response accuracy exceeding 98%.

[0053] To verify the practical effectiveness of this invention, a specific application example is constructed: A provincial power grid has an installed wind power capacity of 8 gigawatts, accounting for 35% of the total installed capacity, and is equipped with a hybrid energy storage system totaling 500 megawatts, including 300 megawatts of lithium batteries, 100 megawatts of flywheels, and 100 megawatts of supercapacitors. During a typical evening peak load period (load of 12 gigawatts), a sudden increase of 1 gigawatt of photovoltaic output causes the frequency to spike to 50.25 Hz. The method of this invention starts within 200 milliseconds, and through rolling optimization, allocates power by absorbing 400 megawatts from the lithium batteries, 300 megawatts from the flywheels, and 300 megawatts from the supercapacitors, restoring the frequency to 50.02 Hz within 10 seconds. Throughout the entire adjustment process, the SOC changes of each energy storage unit remain within a safe range, no unit triggers a health state degradation, the frequency deviation integral is reduced by 42% compared to traditional methods, and energy storage cycle losses are reduced by 35%.

[0054] Another application scenario is a sudden drop in wind power during extreme weather: wind power output drops from 6 gigawatts to 2 gigawatts within 30 seconds, and the frequency falls to 49.7 Hz. This invention dynamically increases the frequency deviation weight to 0.85, prioritizing the use of supercapacitors and flywheels to provide instantaneous power support, with lithium batteries subsequently providing continuous support. The closed-loop feedback mechanism detects an excessive prediction deviation in the third control cycle, automatically re-identifies the system inertia and adjusts constraints, ultimately stabilizing the frequency within 15 seconds without triggering emergency load shedding, verifying the method's strong robustness and adaptability.

[0055] Example 2

[0056] Based on the method described in Example 1, this example focuses on the specific implementation details of differentiated modeling, collaborative scheduling strategies, and health status-driven tiered management mechanisms for different types of energy storage units in a hybrid energy storage system.

[0057] refer to Figure 4The multi-level interaction between the regional power grid energy management system and various types of energy storage units is reflected in a three-layer architecture: the upper layer is the global optimization decision layer, the middle layer is the type coordination layer, and the lower layer is the unit execution layer. The global optimization decision layer runs the rolling optimization algorithm described in this invention and outputs the total frequency regulation power demand; the type coordination layer decomposes the total demand into type-level instructions based on the dynamic characteristics and health status of each energy storage type; and the unit execution layer performs power redistribution within the same type.

[0058] For electrochemical energy storage units, their dynamic characteristics are mainly affected by the state of charge and temperature. In model construction, in addition to the nonlinear efficiency model described in step 2, a temperature-dependent internal resistance model also needs to be introduced. ,in, Where is the internal resistance, and T is the average operating temperature. For reference internal resistance, For activation energy, This represents the Boltzmann constant. The internal resistance model is used to accurately calculate Joule heat loss during charging and discharging, thereby correcting the efficiency model. In the constraints, the maximum charging and discharging power is limited not only by the rate of charge but also by the current temperature: when the temperature is below 0 degrees Celsius, the maximum charging power drops to 0.5C; when the temperature is above 45 degrees Celsius, the maximum discharging power drops to 0.8C. The health status assessment module updates the remaining lifetime prediction every 5 minutes. When the predicted lifetime is less than 3000 cycles or the health status score is less than 85 points, the unit is removed from the main frequency modulation echelon.

[0059] For a flywheel energy storage unit, its dynamic characteristics are described by the rotor kinetic energy equation: ,in For rotational inertia, Angular velocity. The relationship between power command and speed change is as follows: t represents time. Constraints include upper and lower speed limits (typically 60% to 110% of rated speed) and maximum angular acceleration (corresponding to maximum power ramp rate). Since the flywheel does not undergo chemical aging, its health status is primarily assessed by evaluating bearing wear and vacuum level, determined through data fusion from vibration and pressure sensors. When the vibration amplitude exceeds a threshold or the vacuum level deteriorates, it automatically downgrades to the backup system.

[0060] For a supercapacitor cell, its dynamic characteristics are described by an equivalent circuit model, including series resistance, capacitance, and leakage current. Power command is constrained by a voltage window (typically 30% to 95% of rated voltage), and the efficiency model also uses a current-related quadratic function, but with different coefficients: This reflects its low internal resistance characteristics. The health status assessment is based on the growth rate of the equivalent series resistance. When the ESR increases by more than 20%, it is judged as performance degradation and switched to the standby echelon.

[0061] At the type coordination layer, total frequency modulation power demand The decomposition follows the principle of "high frequency, fast response; low frequency, slow storage." Specifically, it will... The components are separated into high-frequency components by a high-pass filter. with low-frequency components High-frequency components (time constant less than 2 seconds) are preferentially allocated to supercapacitors and flywheels due to their millisecond-level response capability; low-frequency components (time constant greater than 2 seconds) are allocated to electrochemical energy storage. The allocation ratio is dynamically adjusted based on the available capacity and health status of each type. For example, if the supercapacitor is in good health and its voltage is at a medium-to-high level, it is allocated 70% of the high-frequency components; if the flywheel's speed is close to its upper limit, its allocation ratio is reduced.

[0062] At the unit execution layer, power reallocation within the same type employs a weighted allocation strategy based on health status scores. Suppose a certain type has... There are 10 available units, and their health status score is 1. Then the first The power command for each unit is ,in, Rate the assigned tasks. Power allocation for this unit, This is the rated power of the unit. This strategy allocates total power to units of this type. It ensures that units with better health and larger capacity undertake more regulation tasks, achieving load balancing and extended lifespan.

[0063] The multi-level safety protection mechanism is further refined in this embodiment. Level 1 protection: If the frequency deviation is within ±0.2 Hz, rolling optimization is performed normally. Level 2 protection: If the frequency deviation is between ±0.2 and ±0.5 Hz, a health status assessment is initiated, backup units are prioritized, and SOC constraints are tightened. Level 3 protection: If the frequency deviation exceeds ±0.5 Hz, optimized output is blocked, switching to fixed gain mode, and an emergency alarm is sent to the dispatch center. The parameters of the fixed gain mode are automatically configured based on the currently available energy storage type: if a supercapacitor is available, high differential gain is enabled to suppress the rate of frequency change; if only electrochemical energy storage is available, high proportional gain is enabled to quickly restore steady state.

[0064] Through the aforementioned differentiated modeling and collaborative scheduling strategies, this embodiment improved the overall frequency regulation response accuracy of the hybrid energy storage system to 98.5% in a real power grid test, which is 5 percentage points higher than that of a single type of energy storage system. At the same time, it reduced the average daily equivalent full cycle count of lithium batteries to 0.8 times, significantly extending their service life.

[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for frequency regulation of a power grid with energy storage based on rolling optimization, characterized in that: The specific steps include the following: step 1. Acquire real-time power grid operation data; including: collecting the current system frequency, total load power, renewable energy output, state of charge of each energy storage unit and its historical charge and discharge sequences, and synchronously acquiring key node voltage and phase angle information through a wide-area measurement system to form a real-time dataset containing dynamic state variables; Step 2: Construct a nonlinear prediction model based on real-time power grid operation data; Based on the real-time dataset, adopt an improved equivalent system model that includes inertia dynamics, frequency regulation dead zone, and energy storage nonlinear efficiency characteristics, and establish a multi-objective optimization function with the objectives of minimizing the integral of frequency deviation and the weighted cost of energy storage lifetime loss; wherein, the model parameters are identified and updated online according to the current system inertia level and renewable energy penetration rate; Step 3: Set the rolling optimization window and constraints based on the multi-objective optimization function; set a prediction time domain of 30 seconds in each control cycle, and use the upper and lower limits of the state of charge of the energy storage unit, the maximum charging and discharging power, the cumulative loss of the number of charging and discharging cycles, and the state prediction deviation range introduced by communication delay as hard constraints to construct a constrained optimal control model. Step 4: Perform robust rolling optimization to solve the constrained optimal control model; adopt a piecewise linearization solution strategy based on model predictive control framework to decompose the nonlinear optimization problem into multiple linear subproblems, perform fast iterative solution in each sub-time domain, output the optimal power command sequence in the next 10 seconds, and only execute the command at the first moment. Step 5: Implement closed-loop feedback correction based on the solution results; compare the actual frequency response after executing the first time-instance command with the predicted trajectory. If the deviation exceeds the set threshold of 5%, trigger the model parameter re-identification and optimization window reset mechanism to dynamically adjust the prediction model gain and constraint boundary, thereby realizing online adaptive correction of the control strategy.

2. The method for frequency regulation of a power grid with energy storage based on rolling optimization according to claim 1, characterized in that: In step 1, the state of charge of the energy storage unit is collected using the ampere-hour integration method combined with open-circuit voltage calibration. The sampling period is 100 milliseconds. The calibration trigger condition is that the energy storage is in a static state for a duration of more than 300 seconds. After calibration, the state of charge error is less than 1%.

3. The method for frequency regulation of a power grid with energy storage based on rolling optimization according to claim 1, characterized in that: In step 2, the improved equivalent system model includes third-order dynamic equations, which describe the system frequency change rate, equivalent inertia response, and governor operation process, respectively; the weight coefficients of the multi-objective optimization function are dynamically configured according to the power grid operation scenario.

4. The method for frequency regulation of a power grid with energy storage based on rolling optimization according to claim 1, characterized in that: In step 3, the cumulative loss from the number of cycles is calculated in real time using the rainflow counting method. After each complete charge-discharge cycle, the equivalent loss factor is updated based on the current state of charge change and temperature conditions, and converted into equivalent cost and incorporated into the optimization objective. The state prediction deviation range introduced by the communication delay is determined based on historical communication delay statistics, with the maximum delay set at 200 milliseconds. The corresponding state prediction compensation uses the forward extrapolation method, combined with the frequency change rate trend for correction. After compensation, the state estimation error is controlled within 0.01 Hz.

5. A method for frequency regulation of a power grid with energy storage based on rolling optimization according to claim 1, characterized in that: In step 4, the piecewise linearization solution strategy includes dividing the 30-second prediction time domain into six 5-second sub-intervals, performing a Taylor expansion on the nonlinear model in each sub-interval, retaining the first-order terms and fixing the coefficients of the second-order terms, and completing a single solution within 10 milliseconds using the interior-point method; the optimal power command sequence includes smoothing constraint processing after its generation, limiting the rate of change between adjacent commands to no more than 80% of the maximum power ramp-up rate of energy storage.

6. The method for frequency regulation of a power grid with energy storage based on rolling optimization according to claim 1, characterized in that: In step 5, the deviation comparison uses the root mean square error index of the sliding window to calculate the root mean square deviation between the predicted frequency and the measured frequency within the next 5 seconds. When the index exceeds 5% for two consecutive control cycles, the model re-identification process is started to re-estimate the equivalent inertia and damping coefficient of the system.

7. The method for frequency regulation of a power grid with energy storage based on rolling optimization according to claim 1, characterized in that: It also includes establishing an energy storage health status assessment module, which uses an empirical decay model to predict the remaining cycle life of each energy storage unit based on the historical charge and discharge sequence and temperature data. When the predicted life is less than 3,000 cycles, it automatically switches the unit from the main frequency regulation team to the standby team and prioritizes the use of energy storage units with good health status.

8. A method for frequency regulation of a power grid with energy storage based on rolling optimization according to claim 7, characterized in that: The health status assessment module employs differentiated assessment strategies for different types of energy storage: For electrochemical energy storage, the remaining lifetime is predicted based on the coupling effect of charge / discharge depth, rate, temperature and cycle number. For flywheel energy storage, health status is judged based on vibration amplitude and vacuum level; for supercapacitors, performance degradation is judged based on the growth rate of equivalent series resistance.

9. A method for frequency regulation of a power grid with energy storage based on rolling optimization according to claim 1, characterized in that: It also includes setting up a multi-level safety protection mechanism. When the frequency deviation is detected to exceed ±0.5 Hz, an emergency load shedding and rapid frequency adjustment response are immediately initiated in coordination. At the same time, the rolling optimization output is locked and switched to a preset fixed gain control mode. The parameters of this fixed gain control mode are automatically configured according to the currently available energy storage type.

Images