A hierarchical cooperative scheduling optimization method and system for distributed energy storage

Abstract

This invention relates to the field of power system energy storage dispatch and control technology, specifically a hierarchical collaborative dispatch optimization method and system for distributed energy storage. The method first acquires the operating status data of the energy storage system and external environmental information; based on this, it constructs a collaborative framework comprising a day-ahead optimization layer, an intraday rolling optimization layer, and a real-time control layer. In the day-ahead optimization layer, with the objective of minimizing the total system operating cost, an improved model predictive control algorithm is used to solve for the planned charge and discharge power curves over the next several days. In the intraday rolling optimization layer, based on updated ultra-short-term forecast information, the same algorithm is invoked to perform rolling corrections on the planned curves, generating a rolling optimization command sequence. In the real-time control layer, based on the deviation between the command sequence and the actual operating status, an adaptive droop control strategy is used to generate the final real-time power command for each energy storage unit and issue it for execution.

Description

Technical Field

[0001] This invention relates to the field of power system energy storage dispatch and control technology, and in particular to a hierarchical collaborative dispatch optimization method and system for distributed energy storage. Background Technology

[0002] With the increasing penetration of new energy sources, distributed energy storage has become an important regulatory resource for maintaining the stable operation of the power grid. Achieving coordinated and optimized operation of a large number of distributed energy storage units is a current research hotspot. Existing scheduling methods are mainly divided into two categories: centralized and decentralized. Centralized scheduling treats all energy storage units as an aggregate, with the upper-level controller uniformly calculating and issuing commands. This method theoretically has good optimization effects, but it relies on highly reliable communication and accurate models, resulting in a concentrated computational burden, slow response to changes in the state of local units and ultra-short-term fluctuations, and difficulty in handling abnormal operating conditions such as communication interruptions. Decentralized control relies on local information, with each unit autonomously responding to grid frequency or voltage signals, resulting in fast response speed and high reliability. However, the lack of coordination between units can easily lead to action conflicts or over-regulation, failing to achieve system-level economic optimization goals and resulting in higher overall operating costs.

[0003] Existing research attempts to combine the advantages of both approaches to design a hierarchical control architecture. A typical framework consists of two layers: day-ahead planning and local control. The day-ahead layer formulates economic plans based on forecast data, while the local layer executes the plans or performs autonomous adjustments. However, this architecture still has significant shortcomings. On the one hand, day-ahead planning is based on short-term forecasts with limited accuracy, making it difficult to adapt to the actual fluctuations in renewable energy and load during the day. Deviations between the plan and real-time operating conditions are directly handled autonomously by the lower-level units, potentially leading to deviations from the optimal trajectory. On the other hand, lower-level control often employs droop control with fixed parameters, whose regulation characteristics cannot adapt to changes in the real-time status of units and grid demand. When there are significant differences in state of charge, health, etc., it can easily lead to uneven output among units, accelerating the aging of some units.

[0004] Therefore, existing technologies have failed to effectively address the coordination issues between long-term economic optimization and short-term fluctuation mitigation, as well as between global planning and local autonomous response. A more refined time scale and more flexible control strategies need to be introduced into the scheduling framework to simultaneously ensure economy, tracking accuracy, and operational reliability. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a hierarchical collaborative scheduling optimization method and system for distributed energy storage.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a hierarchical collaborative scheduling optimization method for distributed energy storage, comprising: The system acquires operational status data and external environment information of the distributed energy storage system. The operational status data includes the state of charge, charge and discharge power boundaries, and health indicators of each energy storage unit. The external environment information includes predicted load curves, new energy power generation curves, and grid dispatch command curves. Based on the operational status data and the external environment information, a collaborative framework is constructed that includes a day-ahead optimization layer, an intraday rolling optimization layer, and a real-time control layer. In the aforementioned day-ahead optimization layer, an improved model predictive control algorithm is used to solve for the planned charge and discharge power curves of each energy storage unit over the next few days, with the goal of minimizing the total system operating cost. In the intraday rolling optimization layer, based on the updated ultra-short-term forecast information, the improved model predictive control algorithm is invoked to perform rolling correction on the planned charge and discharge power curve, generating a rolling optimization instruction sequence; In the real-time control layer, based on the deviation between the rolling optimization command sequence and the actual operating state, an adaptive droop control strategy is adopted to generate the final real-time power command for each energy storage unit. The final real-time power command is sent to the corresponding energy storage unit controller for execution.

[0007] As a further aspect of the present invention, the working principle of the improved model predictive control algorithm includes: At the beginning of each optimization cycle of the day-ahead optimization layer, a rolling optimization problem containing an objective function and multiple constraints is constructed. The objective function is the total system operating cost function that includes electricity purchase and sale costs, energy storage loss costs, and power fluctuation penalty terms. The control time domain and prediction time domain of the improved model predictive control algorithm are initialized, wherein the control time domain includes multiple discrete control step sizes, and the prediction time domain is longer than the control time domain; Within each discrete control step, taking the current state of the energy storage system as the initial point, the rolling optimization problem is solved to obtain the optimal control sequence in the control time domain; The first control quantity of the optimal control sequence is used as the output command at the current moment and applied to the controlled energy storage system. The actual operating state of the energy storage system is fed back to the improved model predictive control algorithm as the initial state for the start of the next optimization cycle; Update some parameters in the objective function, and roll forward one control step, repeating the steps of solving the rolling optimization problem, outputting instructions, and providing feedback on the status.

[0008] As a further aspect of the present invention, the construction of the rolling optimization problem including the objective function and multiple constraints includes: The total operating cost function of the system is constructed using a mathematical expression that includes the cost of electricity purchase, revenue from electricity sales, aging costs of energy storage unit charge-discharge cycles, and a penalty term for the total power ramp-up rate of the energy storage cluster. The first type of constraint is set, which includes upper and lower limits of the state of charge of each energy storage unit, upper and lower limits of charging and discharging power, and balance constraints of the state of charge at the beginning and end of the scheduling cycle. Set a second type of constraint, which includes upper and lower limits of the total power of the energy storage cluster and the power exchanged with the grid, as well as the power change rate constraint. A third type of constraint is set, which is based on the health index and sets differentiated power allocation weight coefficients for energy storage units with different health levels. The rolling optimization problem is formed by combining the total system operating cost function, the first type of constraint, the second type of constraint, and the third type of constraint.

[0009] As a further aspect of the present invention, in the intraday rolling optimization layer, based on updated ultra-short-term forecast information, the improved model predictive control algorithm is invoked to perform rolling correction on the planned charge-discharge power curve, generating a rolling optimization instruction sequence, including: Receive updated ultra-short-term load forecast curves, ultra-short-term renewable energy generation forecast curves, and the latest real-time grid dispatch instructions; Based on the planned charge and discharge power curve, the improved model predictive control algorithm is re-called for optimization calculation within the shortened prediction time domain and the shortened control time domain. In the optimization calculation, a deviation penalty term for the planned charge and discharge power curve is introduced to ensure that the deviation between the rolling optimization instruction sequence and the planned charge and discharge power curve is within the allowable range. The optimization results obtained are time-scaled smaller than the current day's optimization step size, forming a rolling optimization instruction sequence for the next few hours.

[0010] As a further aspect of the present invention, the step of re-invoking the improved model predictive control algorithm for optimization calculation includes: Within the shortened prediction time domain, higher time resolution is used to model the uncertainties of load, new energy generation, and grid commands; Within the shortened control time domain, based on the real-time available power and state of charge of the energy storage unit, the power allocation weight coefficients set based on the health index in the rolling optimization problem of the improved model predictive control algorithm are adjusted. Add a power fluctuation quadratic penalty term to the objective function to smooth the rolling optimization instruction sequence; An optimization solver with warm start is adopted, using the solution of the previous optimization cycle as the starting point for the current solution, thereby accelerating the solution process of the improved model predictive control algorithm.

[0011] As a further aspect of the present invention, in the real-time control layer, based on the deviation between the rolling optimization command sequence and the actual operating state, an adaptive droop control strategy is adopted to generate the final real-time power command for each energy storage unit, including: Real-time acquisition of total output power of energy storage clusters, power of grid interconnection lines, and voltage and frequency of key busbars; Calculate the power deviation between the total output power of the energy storage cluster and the power value of the corresponding instruction value in the rolling optimization instruction sequence; Based on the magnitude and direction of the power deviation, the droop coefficient in the adaptive droop control strategy is dynamically adjusted, and the droop coefficient is positively correlated with the absolute value of the power deviation. Based on the adjusted droop coefficient, the real-time state of charge of each energy storage unit, and the health index, the power deviation is allocated proportionally to form the power regulation component of each energy storage unit. The power regulation component is superimposed with the instruction reference value at the corresponding moment in the rolling optimization instruction sequence to obtain the final real-time power instruction for each energy storage unit.

[0012] As a further aspect of the present invention, the droop coefficient in the adaptive droop control strategy is dynamically adjusted according to the magnitude and direction of the power deviation, including: A preset adjustment range for the droop coefficient, wherein the adjustment range includes the minimum droop coefficient and the maximum droop coefficient; Establish a piecewise linear mapping relationship between the droop coefficient and the absolute value of the power deviation. When the absolute value of the power deviation is less than the first threshold, the minimum droop coefficient is used. When the absolute value of the power deviation is between the first threshold and the second threshold, the droop coefficient increases linearly with the absolute value of the power deviation. When the absolute value of the power deviation is greater than the second threshold, the maximum droop coefficient is used; When the power deviation direction is positive, a droop coefficient adjustment strategy is adopted to address the power deficit. When the power deviation is negative, a droop coefficient adjustment strategy is adopted to address power excess.

[0013] As a further aspect of the present invention, the setting of a third type of constraint condition, based on the health index, sets differentiated power allocation weighting coefficients for energy storage units with different health levels, including: A mapping relationship is established between the health index and the health level of the energy storage unit. The health index is calculated based on the number of historical cycles, the capacity decay rate, and the internal resistance growth rate. The health level includes healthy, slightly decayed, moderately decayed, and severely decayed. Different power allocation weight coefficients are set for different health levels, with the highest power allocation weight coefficient for energy storage units in the healthy level and the lowest power allocation weight coefficient for energy storage units in the severely degraded level. A power allocation term is added to the objective function of the rolling optimization problem. The power allocation term represents the sum of squares of the deviations between the actual charging and discharging power of each energy storage unit and the product of its rated power and the power allocation weight coefficient. During the optimization process, by minimizing the power allocation term, energy storage units with higher health levels can bear a larger proportion of the charging and discharging power, while power protection is provided for energy storage units with lower health levels.

[0014] As a further aspect of the present invention, the step of establishing a piecewise linear mapping relationship between the droop coefficient and the absolute value of the power deviation, wherein when the absolute value of the power deviation is less than a first threshold, the minimum droop coefficient is used, includes: Set segmented intervals for the absolute value of power deviation, wherein the segmented intervals include at least a first interval, a second interval, and a third interval, wherein the first interval is defined as the absolute value of power deviation being less than a first threshold. A corresponding droop coefficient calculation rule is defined for each segmented interval. For the first interval, the corresponding droop coefficient calculation rule is: regardless of how the absolute value of the power deviation changes, a preset fixed value is used as the minimum droop coefficient. During system operation, the absolute value of the power deviation is calculated in real time; The calculated absolute value of the power deviation is compared with the first threshold. If the absolute value of the power deviation is determined to be less than the first threshold, then the droop coefficient is set to the minimum droop coefficient according to the droop coefficient calculation rule defined for the first interval.

[0015] As a further aspect of the present invention, the present invention also includes a hierarchical collaborative scheduling optimization system for distributed energy storage, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the hierarchical collaborative scheduling optimization method for distributed energy storage as described above.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: A three-layer collaborative framework, comprising day-ahead optimization, intraday rolling optimization, and real-time control, is adopted, with improved advanced optimization algorithms and adaptive control strategies adapted at different levels. The day-ahead optimization layer, oriented towards a multi-day timescale, aims to minimize total operating cost and generates an initial planning curve, laying the foundation for global economic optimization. The intraday rolling optimization layer introduces shorter time windows and updated ultra-short-term forecast data to continuously revise the day-ahead plan. This design, while maintaining the stability of the long-term optimization framework, significantly reduces the impact of forecast errors caused by fluctuations in renewable energy power and load through high-frequency local re-optimization. This allows dispatch instructions to closely track changes in the actual operating status of the system, improving the executability of the plan and its ability to mitigate fluctuations.

[0017] An adaptive droop control strategy is employed at the real-time control layer. Based on the deviation between the received rolling optimization commands and the actual operating state of the units, control parameters are dynamically adjusted to generate the final power command. This strategy ensures that control parameters are no longer fixed values ​​but can be adaptively adjusted according to the real-time state of charge, health, and grid demand of the units. In response to command deviations or grid disturbances, each unit can rationally allocate and regulate power according to its own state, preventing some units from being overloaded or deeply discharged for extended periods, thus achieving autonomous coordination among units and reasonable load sharing. This enhances robustness to sudden situations and communication anomalies, effectively extending the overall lifespan of the energy storage system while ensuring tracking of upper-level optimization objectives. Attached Figure Description

[0018] Figure 1 This is a state diagram of the hierarchical collaborative scheduling optimization method for distributed energy storage described in this invention. Figure 2 To construct a flowchart for the rolling optimization problem; Figure 3 The flowchart shows the process of re-invoking the improved model predictive control algorithm for optimization calculations in the inner layer. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 This invention provides a hierarchical collaborative scheduling optimization method for distributed energy storage. The method includes: The system acquires operational status data and external environmental information for the distributed energy storage system. Operational status data includes the real-time state of charge (SOC) of each energy storage unit, the currently available charge / discharge power boundaries, and health indicators characterizing its aging. External environmental information includes predicted future load demand curves, predicted power generation curves from new energy sources (such as photovoltaics and wind power), and dispatch command curves issued by the power grid. Based on this data and information, a top-down collaborative optimization framework with three time scales is constructed: a day-ahead optimization layer, an intraday rolling optimization layer, and a real-time control layer. In the day-ahead optimization layer, an improved model predictive control algorithm is used to minimize the total system operating cost over the next few days, calculating the planned charge / discharge power curves for each energy storage unit in the coming days. In the intraday rolling optimization layer, based on the latest updated ultra-short-term forecast information, the aforementioned improved model predictive control algorithm is invoked again to continuously correct and refine the planned charge / discharge power curves issued by the day-ahead layer, generating a rolling optimization command sequence for the next few hours. At the real-time control layer, based on the instantaneous deviation between the rolling optimization command sequence and the actual operating state of the energy storage cluster, an adaptive droop control strategy is adopted to dynamically and quickly distribute the total power deviation to each energy storage unit, thereby generating the final real-time power command for each energy storage unit. These real-time power commands are then sent to the corresponding local controller of the energy storage unit, which drives the power converter to perform specific charging and discharging actions.

[0022] In one embodiment of the present invention, at the beginning of each optimization cycle of the day-ahead optimization layer, such as 0:00 each day, a rolling optimization problem containing a specific objective function and multiple constraints needs to be constructed. The objective function is defined as the total system operating cost function, which includes electricity purchase cost, electricity sales revenue, energy storage device loss cost, and system power fluctuation penalty term. During algorithm initialization, the lengths of the control time domain and the prediction time domain need to be set. The control time domain includes multiple discrete control steps (e.g., one step every 15 minutes), while the prediction time domain needs to cover a longer time range, usually longer than the control time domain, such as covering the next 24 to 72 hours. At the calculation time of each discrete control step, the actual state of each unit of the energy storage system at the current moment (e.g., state of charge) is used as the initial state point of the optimization problem. The constructed rolling optimization problem is solved to obtain a series of optimal control quantities in the entire control time domain, forming an optimal control sequence. From this optimal control sequence, the first control quantity, i.e., the charging and discharging power command corresponding to the next control step, is selected as the command to be output at the current moment and issued to the controlled energy storage system for execution. After receiving the instruction and executing one control step, the energy storage system's actual operating state (such as the new state of charge and actual power) is measured and fed back to the improved model predictive control algorithm. This feedback state information serves as the initial state for the next optimization cycle (i.e., at the start of the next control step). Simultaneously, the algorithm updates some parameters in the objective function or constraints (such as updated electricity prices and more accurate prediction data), then rolls the entire optimization window (including the control and prediction time domains) forward by one control step, repeatedly performing the steps of solving the rolling optimization problem, outputting instructions, and receiving state feedback, thus achieving closed-loop optimization and rolling advancement.

[0023] In practical implementation, the working principle of the improved model predictive control algorithm is realized in the day-ahead optimization layer. At the beginning of each optimization cycle in the day-ahead optimization layer, such as midnight each day, a rolling optimization problem is constructed, which includes an objective function and multiple constraints. The objective function is the total system operating cost function, which includes the cost of purchasing and selling electricity, the cost of energy storage losses, and a power fluctuation penalty term. The control time domain and prediction time domain of the improved model predictive control algorithm are initialized. The control time domain contains multiple discrete control steps, and the length of the prediction time domain exceeds the length of the control time domain. Within each discrete control step, the current real-time state of the distributed energy storage system is used as the initial point for optimization calculation, and the rolling optimization problem is solved to obtain the optimal control sequence within the control time domain. The first control quantity is extracted from the optimal control sequence and used as the instruction to be output at the current moment, which is then issued to the controlled energy storage system for execution. The actual operating state of the distributed energy storage system, including the updated state of charge and measured power of each energy storage unit, is fed back to the improved model predictive control algorithm, and the actual operating state will serve as the initial state for the next optimization cycle. Update some parameters in the objective function, such as updated electricity price information and new energy forecast data, and roll the entire optimization window forward by one control step. Repeat the steps of solving the rolling optimization problem, outputting instructions, and receiving status feedback.

[0024] In some embodiments, the improved model predictive control algorithm constructs a rolling optimization problem at the beginning of each optimization cycle with a definite form. The mathematical expression of the rolling optimization problem aims to minimize the total operating cost function of the system. A specific expression for the total operating cost function of the system is as follows: Where: characters Represents the total operating cost of the system, character Represents the total number of steps in the prediction time domain, character Indicates the time step index, character Indicates at time step Electricity price, character Indicates at time step The exchange power between the energy storage cluster and the power grid, characters Indicates the duration of a single control step, character Indicates the power fluctuation penalty coefficient, character Indicates at time step Total power of the energy storage cluster, character Indicates the total number of energy storage units, character Indicates the first The loss cost coefficient of each energy storage unit, character Indicates the first Energy storage in time step The charging and discharging power. Solving the rolling optimization problem involves finding a set of control variables that, under all preset constraints, makes the total system cost function... Minimum.

[0025] In practical implementation, the initial settings of the control time domain and prediction time domain are fundamental to the operation of the improved model predictive control algorithm. The length of the control time domain determines the number of future control steps required for each optimization calculation. The control time domain contains multiple discrete control steps, each corresponding to a fixed time interval, such as fifteen minutes. The prediction time domain is longer than the control time domain, covering a range from the current moment to a longer future time, such as the next twenty-four hours. Within each discrete control step, the optimization calculation only solves for the optimal control sequence within the control time domain, but the evaluation of the objective function and the consideration of constraints cover the entire prediction time domain. The improved model predictive control algorithm sets the lengths of the control and prediction time domains during initialization, and each time the optimization window rolls, both the control and prediction time domains move forward synchronously by one control step.

[0026] In some embodiments, the process of solving and applying the optimal control sequence constitutes a closed-loop feedback. Within each discrete control step, the current state of the energy storage system, including the state of charge of all energy storage units, is used as the initial state of the optimization problem. After solving the rolling optimization problem, an optimal control sequence covering the entire control time domain from the current moment is obtained. The optimal control sequence contains a series of charging and discharging power command values ​​corresponding to future control steps. The first control variable in the optimal control sequence, i.e., the command corresponding to the next immediately executed control step, is used as the output command at the current moment and sent to the controllers of each unit of the distributed energy storage system. After the energy storage system executes the output command for one control step, its actual operating state is measured and fed back by sensors. The actual operating state is used as a new initial state for the optimization calculation of the starting point of the next control step, thereby realizing closed-loop rolling optimization based on state feedback.

[0027] Optionally, the method for updating some parameters in the objective function can be based on changes in external information. Before rolling the optimization window forward at each control step, the time-related parameters in the objective function need to be updated. For example, the time-of-use electricity price parameter on which the electricity purchase and sale cost term in the total system operating cost function depends needs to be updated to the latest predicted or published electricity price curve. The baseline power value in the power fluctuation penalty term can be adjusted based on the latest ultra-short-term load forecast. After updating some parameters in the objective function, the improved model predictive control algorithm reconstructs and solves the rolling optimization problem with the new initial state, new parameters, and the rolled optimization time window. This parameter update mechanism enables the improved model predictive control algorithm to adapt to dynamic changes in the external environment and market information.

[0028] It is understandable that the rolling advancement mechanism is a core feature of the improved model predictive control algorithm. After completing the optimization calculation, command output, and state feedback at the current moment, the entire optimization time window, including both the control and prediction time domains, will move forward by one discrete control step. The starting points of the control and prediction time domains are updated synchronously to the next moment. In the new time window, although the mathematical form of the optimization problem remains the same, the initial state, some parameters, and the predicted data sequence have all been updated. Repeating the steps of solving the rolling optimization problem, outputting commands, and providing state feedback allows the optimization decision to be continuously updated based on the latest information, forming a rolling forward optimization control process.

[0029] In practical implementation, the improved model predictive control algorithm can employ a mathematical programming solver to solve the problem. The rolling optimization problem is typically constructed as a constrained mathematical programming problem, such as quadratic programming or mixed-integer linear programming. Within each control step, the improved model predictive control algorithm invokes a dedicated optimization solver, such as an interior-point solver, to numerically solve the constructed rolling optimization problem. The solver receives the initial state parameters, the predicted data sequence, and all coefficient matrices of the objective function and constraints, and outputs the optimal control sequence through iterative calculation. The solver's computation speed must meet the requirements of the control step time interval to ensure that the optimization results can be applied to system control in a timely manner.

[0030] Optionally, the accuracy of the state feedback directly affects the performance of the improved model predictive control algorithm. The actual operating state is fed back to the improved model predictive control algorithm, primarily including the actual state of charge and actual charging / discharging power of each energy storage unit. These state variables need to be measured and estimated in real time using high-precision sensors, such as voltage and current sensors and fuel gauges. The deviation between the measured actual operating state and the state predicted by the improved model predictive control algorithm in the previous cycle reflects model errors and external disturbances. Using the actual operating state as the initial state for the next optimization cycle can effectively correct these deviations, giving the rolling optimization process of the improved model predictive control algorithm closed-loop correction capability, thereby improving the tracking accuracy and robustness of medium- and long-term scheduling plans.

[0031] In one embodiment of the present invention, see [reference] Figure 2The construction process uses a mathematical expression that includes electricity purchase cost, electricity sales revenue, energy storage unit charge / discharge cycle aging cost, and a penalty term for the total power ramp-up rate of the energy storage cluster to form the total system operating cost function. First-type constraints need to be set, primarily targeting individual energy storage units. These include upper and lower limits for the state of charge (SOC) of each unit to ensure safe operation; upper and lower limits for the charge / discharge power of each unit to reflect its instantaneous power capacity; and SOC balance constraints at the beginning and end of the scheduling cycle to ensure periodic energy recovery. Second-type constraints need to be set, addressing the interaction between the entire energy storage cluster and the grid. These include upper and lower limits for the total power of the energy storage cluster and the power exchanged with the grid to meet transformer or line capacity limitations; and a total power change rate constraint to smooth power fluctuations. Third-type constraints need to be set, based on health indicators, assigning differentiated power allocation weight coefficients to energy storage units with different health levels. A mapping relationship between the health index and health level of energy storage units is established. The health index is calculated by analyzing the historical cycle count, capacity decay rate, and internal resistance growth rate of the energy storage units. The health level can be divided into several levels, including healthy, slightly decayed, moderately decayed, and severely decayed. Differentiated power allocation weight coefficients are set for different health levels, with the highest value set for the power allocation weight coefficient of the healthy level energy storage units and the lowest value set for the power allocation weight coefficient of the severely decayed level energy storage units. A power allocation term is added to the objective function of the rolling optimization problem. This term represents the sum of squared deviations of the actual charging and discharging power of each energy storage unit from its rated power and the power allocation weight coefficient it is assigned. During the optimization process, by minimizing this power allocation term, energy storage units with higher health levels tend to undertake a larger proportion of charging and discharging power tasks, while limiting the charging and discharging power of energy storage units with lower health levels, thus achieving power protection. Finally, the total system operating cost function, the first type of constraints, the second type of constraints, and the third type of constraints are combined to form a complete rolling optimization problem to be solved.

[0032] In practical implementation, constructing a rolling optimization problem containing an objective function and multiple constraints is the core step in the operation of the improved model predictive control algorithm at the day-ahead optimization layer. The construction process begins with defining the total system operating cost function, a mathematical expression that includes electricity purchase cost, electricity sales revenue, energy storage unit charge / discharge cycle aging cost, and a penalty term for the total power ramp-up rate of the energy storage cluster. First-type constraints are set, including upper and lower limits for the state of charge (SOC) of each energy storage unit, upper and lower limits for the charge / discharge power of each energy storage unit, and SOC balance constraints at the beginning and end of the scheduling cycle. Second-type constraints are set, including upper and lower limits for the power exchanged between the total power of the energy storage cluster and the grid, and a rate of change constraint for the power exchanged between the total power of the energy storage cluster and the grid. Third-type constraints are set, based on health indicators, assigning differentiated power allocation weight coefficients to energy storage units with different health levels. Combining the total system operating cost function, the first type of constraints, the second type of constraints, and the third type of constraints constitutes a complete constrained mathematical optimization problem, namely the rolling optimization problem.

[0033] In some embodiments, the specific composition of the total system operating cost function needs to be clearly defined. The electricity purchase cost is calculated as the product of the time-of-use price and the power purchased from the grid. The electricity sales revenue is calculated as the product of the electricity sales price and the power delivered to the grid, and is typically included in the mathematical expression as a negative cost. The energy storage unit charge / discharge cycle aging cost is quantified as a function of the charge / discharge power of each energy storage unit and its corresponding cycle aging coefficient. The energy storage cluster total power ramp-up penalty is calculated as the product of the square of the change in the total power of the energy storage cluster over adjacent time periods and a penalty coefficient. This term is used to smooth the total power output curve of the energy storage cluster and suppress drastic power fluctuations. These cost components are combined linearly or nonlinearly to form the complete total system operating cost function, which serves as the objective to be minimized in the rolling optimization problem.

[0034] In practical implementation, the first and second types of constraints define the physical boundaries and operating rules of the system. The first type of constraints directly applies to each individual energy storage unit. The upper and lower limits of the state of charge (SOC) ensure that the SOC of each unit remains above its set lower limit and upper limit during optimization. The upper and lower limits of charging and discharging power ensure that the charging and discharging power command value of each unit at any given time is within its currently allowed maximum charging and discharging power range. The SOC balance constraint at the beginning and end of the scheduling cycle requires that the SOC of each unit at the start of the optimization cycle be equal to or maintain within a preset difference from the SOC at the end of the optimization cycle, thus achieving periodic scheduling of energy storage. The second type of constraints applies to the interaction interface between the entire energy storage cluster and the power grid. The upper and lower limits of the total power of the energy storage cluster and the power exchanged with the grid are determined by transformer capacity, line current carrying capacity, or contractually agreed power limits. The power change rate constraint limits the maximum variation in the total power of the energy storage cluster within adjacent time intervals to meet the grid's power regulation rate requirements.

[0035] In some embodiments, the implementation of the third type of constraint is based on the assessment and mapping of the health status of the energy storage unit. Establishing a mapping relationship between the energy storage unit's health index and its health level is fundamental to the implementation. The health index is calculated by analyzing the historical operating data of the energy storage unit, involving multiple dimensions of parameters such as the number of historical cycles, the rate of decay of the current capacity relative to the initial capacity, and the rate of increase of the current internal resistance relative to the initial internal resistance. The health level can be divided into several discrete levels, such as healthy, slightly decayed, moderately decayed, and severely decayed, with each level corresponding to a numerical range of the health index. Differentiated power allocation weight coefficients are set for different health levels, with the power allocation weight coefficient for the healthy level energy storage unit set to the highest value, the power allocation weight coefficient for the severely decayed level energy storage unit set to the lowest value, and the power allocation weight coefficients for the slightly decayed and moderately decayed levels set to an intermediate value between the highest and lowest values. The level of the power allocation weight coefficient directly determines the priority of the energy storage unit in the power allocation task.

[0036] In practical implementation, a power allocation term is introduced as part of the objective function to achieve differentiated power scheduling. Adding a power allocation term to the objective function of the rolling optimization problem, the mathematical expression of which aims to quantify the deviation between the actual charging and discharging power undertaken by each energy storage unit and a desired baseline value, is crucial. The desired baseline value is proportional to the rated power of the energy storage unit and its corresponding power allocation weighting coefficient. A specific expression for the power allocation term is as follows: Where: characters Indicates power allocation item, character Indicates the total number of time periods to be optimized, character Indicates the total number of energy storage units, character Represents a time index, character Indicates the energy storage unit index, character Indicates the first Power deviation penalty coefficient for each energy storage unit, character Indicates the first The energy storage in the first Planned charge / discharge power for each time period, characters Indicates the first Each energy storage unit has a power allocation weighting coefficient based on its health level. Indicates the first The rated power of each energy storage unit. In the optimization process, minimizing the objective function means simultaneously minimizing the power allocation term. The value of .

[0037] Optionally, the health index calculation can incorporate multiple characteristic parameters. Historical cycle count is obtained by accumulating the number of complete charge-discharge cycles recorded by the battery management system. Capacity decay rate is calculated by comparing the actual usable capacity under full charge with the factory-specified capacity. Internal resistance growth rate is calculated by performing periodic diagnostic tests on the energy storage unit, obtaining the current AC internal resistance value, and comparing it with the initial internal resistance value. The health index can be designed as a weighted comprehensive score of these characteristic parameters; the lower the score, the worse the health status of the energy storage unit. Based on the calculated health index score, a preset mapping table is consulted to determine the current health level of the energy storage unit, and then the corresponding power allocation weight coefficient is assigned to it.

[0038] It is understandable that minimizing the power allocation term can guide the optimization results to conform to the power allocation strategy. During the optimization process, by minimizing the power allocation term in the objective function, the algorithm will tend to allocate the actual charging and discharging power of the energy storage units. Approaching its expected benchmark value Because energy storage units with higher health ratings have a higher power allocation weighting factor. Their expected baseline value is also higher, so in the optimization results, energy storage units with higher health levels will naturally bear a larger proportion of charging and discharging power. Meanwhile, for energy storage units with lower health levels, due to their lower power allocation weighting coefficient... The actual charging and discharging power is relatively low, and the optimization algorithm will limit it. The size of the power storage unit is adjusted to achieve power protection and delay further aging of low-health energy storage units. This mechanism achieves optimized power scheduling and collaborative management of equipment health by optimizing the constraints and objectives of the problem, rather than simply allocating rules.

[0039] In practice, multiple constraints and the objective function together constitute a mathematical optimization problem. The relevant definitions in the first, second, and third types of constraints, such as the power allocation weight coefficients, are expressed in the form of linear equations or inequalities. The total system operating cost function and the power allocation term together constitute the overall objective function to be minimized. All these equations and inequalities, along with the objective function, are input into the mathematical programming solver. Under the premise of satisfying all constraints, the solver finds a set of optimal decision variable values. The decision variables include the charging and discharging power of each energy storage unit in each time period, thereby achieving the dual objectives of minimizing total cost and optimizing power allocation. The resulting sequence of decision variables is part of the optimal control sequence obtained by the improved model predictive control algorithm within the current optimization cycle.

[0040] In one embodiment of the invention, the intraday rolling optimization layer needs to receive updated ultra-short-term load forecast curves, ultra-short-term renewable energy generation forecast curves, and the latest real-time grid dispatch instructions. Using the planned charge / discharge power curve issued by the day-ahead optimization layer as a reference trajectory, within a shortened prediction time domain and a shortened control time domain (e.g., predicting the next 4 hours and controlling the next 1 hour), the improved model predictive control algorithm is re-invoked for optimization calculations. In the optimization calculation, the algorithm introduces a deviation penalty term for the planned charge / discharge power curve. This term aims to penalize excessive deviations of the optimization result from the day-ahead plan, thereby ensuring that the deviation between the generated rolling optimization instruction sequence and the original planned charge / discharge power curve is controlled within an allowable range. See also... Figure 3 Within a shorter optimization time domain, a higher time resolution, such as 5-minute or 1-minute step sizes, is used to refine the modeling of uncertainties in load, renewable energy generation, and grid commands. Based on the real-time available power and state of charge of the energy storage unit during the day, the power allocation weight coefficients set based on health indicators in the rolling optimization problem solved by the improved model predictive control algorithm are adjusted to better reflect the current reality. A power fluctuation quadratic penalty term is added to the objective function to further smooth the generated rolling optimization command sequence and reduce drastic changes in power commands. An optimization solver with hot-start capability is used for solving the problem, using the optimal solution calculated in the previous optimization cycle as the starting point for the current iteration, thereby accelerating the convergence process of the improved model predictive control algorithm. Through solving, optimization results with finer time scales and shorter cycles are obtained, forming a rolling optimization command sequence for the next few hours.

[0041] In practical implementation, the intraday rolling optimization layer, based on updated ultra-short-term forecast information, invokes an improved model predictive control algorithm to perform rolling corrections on the planned charge-discharge power curves issued by the day-ahead optimization layer, and generates a rolling optimization command sequence. The intraday rolling optimization layer receives updated ultra-short-term load forecast curves, ultra-short-term renewable energy generation forecast curves, and the latest real-time grid dispatch commands. Using the planned charge-discharge power curves generated by the day-ahead optimization layer as a reference trajectory, the intraday rolling optimization layer re-invokes the improved model predictive control algorithm for optimization calculations within a significantly shortened forecast time domain and a shortened control time domain. In the optimization calculations, the intraday rolling optimization layer introduces a deviation penalty term for the planned charge-discharge power curves. This deviation penalty term aims to ensure that the deviation between the generated rolling optimization command sequence and the original planned charge-discharge power curves is constrained within an allowable range. The intraday rolling optimization layer obtains optimization results with finer time scales and shorter cycles through solving, forming a rolling optimization command sequence for the next few hours.

[0042] In some embodiments, the shortened prediction and control time domains need to be explicitly defined. The prediction time domain is significantly shortened compared to the prediction time domain of the day-ahead optimization layer, which covers multiple days, for example, shortened to the next 4 hours. The control time domain is also shortened accordingly, for example, shortened to the next 1 hour. The optimization start cycle of the intraday rolling optimization layer is much shorter than that of the day-ahead optimization layer, for example, rolling optimization is triggered every 15 minutes or every 5 minutes. In a specific operational example, assume that the day-ahead optimization layer generates planned charge and discharge power curves for 96 time points (one point every 15 minutes) throughout the day at 0:00. At 10:00 on the same day, the intraday rolling optimization layer is started, with its prediction time domain covering 10:00 to 14:00 and its control time domain covering 10:00 to 11:00. The intraday rolling optimization layer will re-optimize the plan for the 1 hour from 10:00 to 11:00 based on the latest ultra-short-term forecast information obtained at 10:00, and output a fine-tuning instruction for the first control step size (e.g., 10:00-10:05). Five minutes later, at 10:05, the optimization window scrolls forward, the prediction time domain is updated to 10:05 to 14:05, the control time domain is updated to 10:05 to 11:05, and optimization is performed again based on the latest information at 10:05, and this process is repeated.

[0043] In practical implementation, adjustments are made to the internal model and parameters of the algorithm. Within the shortened prediction time domain, a higher time resolution is used to model the uncertainties of load, renewable energy generation, and grid commands. For example, the modeling step size is increased from 15 minutes in the day-ahead optimization layer to 5 minutes or 1 minute, thus more accurately describing the details of near-future power fluctuations. Within the shortened control time domain, based on the real-time available power and real-time state of charge of the energy storage unit, the power allocation weight coefficients set based on the health index in the rolling optimization problem of the improved model predictive control algorithm are dynamically adjusted. For example, if the state of charge of an energy storage unit is close to the lower limit during real-time operation, even if its health level is high, its power allocation weight coefficient may be temporarily reduced in the rolling optimization to prevent over-discharge. A power fluctuation quadratic penalty term is added to the objective function to smooth the rolling optimization command sequence and avoid drastic jumps in power commands between adjacent time points. The intraday rolling optimization layer employs a hot-start optimization solver, using the optimal solution obtained from the previous optimization cycle as the initial point for the current optimization calculation iteration. This significantly accelerates the solution convergence process of the improved model predictive control algorithm, meeting the shorter computation time requirement of the intraday rolling optimization layer.

[0044] In some embodiments, introducing a deviation penalty term for the planned charge / discharge power curve is crucial to ensuring day-ahead and intraday coordination. The deviation penalty term is added as an additional cost component to the objective function of rolling optimization. A mathematical expression for the deviation penalty term is: Where: characters Represents the total cost of the deviation penalty term, character Indicates the total number of steps in the current control time domain of the rolling optimization layer, character Indicates the time step index of the scroll optimization layer, character Indicates at time step Deviation penalty coefficient, character This indicates that the rolling optimization layer needs to be optimized at time step [number]. Total power command for energy storage cluster, character This indicates the value extracted from the current-day planning curve at the corresponding time step. The planned total power of the energy storage cluster. This is achieved by setting an appropriate deviation penalty factor. It can control scroll optimization commands Plans before tracking The trade-off between responding to the latest ultra-short-term forecasts and the current situation. The existence of a deviation penalty term ensures that intraday rolling optimization, while making full use of the latest information for local adjustments, does not completely deviate from the globally economically optimal plan established at the previous day.

[0045] Optionally, the updated ultra-short-term forecast information in the rolling optimization process is compared with the day-ahead forecast information. Refer to Table 1, which shows a comparison of load and PV power data between the day-ahead and ultra-short-term forecasts for the same future period at a specific time in an example scenario. Intraday rolling optimization will make decisions based on more accurate ultra-short-term forecast data.

[0046] Table 1: Comparison of Day-ahead Forecast and Ultra-short-term Forecast Data Ultra-short-term load forecasts are generally higher than day-ahead forecasts, while ultra-short-term photovoltaic (PV) forecasts are significantly lower than day-ahead forecasts at T+5 min and T+10 min. This forecast discrepancy directly leads to changes in the net load (load minus PV generation) forecast. When the intraday rolling optimization layer calls the improved model predictive control algorithm, the load and renewable energy forecast data in its model will use the data from the "Ultra-short-term load forecast" and "Ultra-short-term PV forecast" columns in the table above, instead of day-ahead forecast data. Therefore, the resulting rolling optimization instruction sequence will differ from the planned charge and discharge power curve based on day-ahead forecasts. The deviation penalty term manages the magnitude of this difference.

[0047] It is understandable that using a hot-start optimization solver can effectively improve computational efficiency. In the intraday rolling optimization layer, optimization calculations are triggered frequently (e.g., every 5 minutes), thus placing strict requirements on the solution speed of each optimization problem. The hot-start optimization solver leverages the high similarity between two consecutive rolling optimization problems. Specifically, the solver uses the optimal solution calculated in the previous optimization cycle (e.g., at time t), including the power values ​​of each energy storage unit in each future time period, as the initial iteration point for the current optimization cycle (at time t+Δt). Since the Δt time interval is very short, the changes in system state and external conditions are limited, and the solution of the previous cycle is usually very close to the optimal solution of the current cycle. Starting the iteration from this initial point close to the optimal solution can significantly reduce the number of iterations required for the solver to reach the optimal solution, thereby obtaining the optimization result while meeting the time requirements and ensuring the real-time performance of intraday rolling optimization.

[0048] In practice, generating the rolling optimization instruction sequence is a continuously rolling process. Each time the improved model predictive control algorithm is re-invoked for optimization calculation, the resulting solution is the optimal control sequence covering the entire shortened control time domain. The intraday rolling optimization layer only sends the instruction for the first control step in the sequence—the instruction to be executed immediately—as the final output at the current moment to the real-time control layer. Instructions for subsequent control steps in the sequence are only used as intermediate calculation results for internal optimization iterations and are not directly sent out. For example, in the optimization at 10:00, the solution yields power instructions for 12 time periods: 10:00-10:05, 10:05-10:10, ..., 10:55-11:00. The intraday rolling optimization layer uses the instruction for the 10:00-10:05 time period as the output of the rolling optimization instruction sequence at the current moment. At 10:05, the optimization window rolls, the solution is re-evaluated based on new information, and the instruction for the 10:05-10:10 time period is output again. In this way, the rolling optimization instruction sequence can be dynamically generated and continuously updated, forming a refined short-term power instruction stream that adapts to the latest system state.

[0049] In one embodiment of the present invention, at the real-time control layer, the total output power of the energy storage cluster, the power of the tie line connected to the grid, and the voltage and frequency signals of the key bus are collected in real time. The power deviation between the measured value of the total output power of the energy storage cluster and the corresponding command value in the rolling optimization command sequence issued by the rolling optimization layer within the day is calculated. Based on the magnitude and direction of the calculated power deviation, the droop coefficient used in the adaptive droop control strategy is dynamically adjusted, and the value of the droop coefficient is positively correlated with the absolute value of the power deviation. Based on the adjusted droop coefficient, the real-time state of charge and health indicators of each energy storage unit, and according to a preset allocation principle, the total power deviation is proportionally decomposed to each energy storage unit, forming the power regulation component that each energy storage unit needs to undertake. The calculated power regulation component of each energy storage unit is algebraically superimposed with the command reference value at the corresponding time obtained from the rolling optimization command sequence to obtain the final real-time power command for each energy storage unit.

[0050] In practical implementation, the real-time control layer employs an adaptive droop control strategy based on the deviation between the rolling optimization command sequence and the actual operating state to generate the final real-time power command for each energy storage unit. The real-time control layer continuously collects data on the total output power of the energy storage cluster, the power of the grid interconnection lines, and the voltage and frequency of key buses. These collected data form the basis for calculating power deviation and executing control. The real-time control layer calculates the power deviation between the measured total output power of the energy storage cluster and the corresponding command value in the rolling optimization command sequence issued by the rolling optimization layer within the day. The positive or negative direction of the power deviation indicates whether the actual output power of the energy storage cluster is too large or too small relative to the planned command. Based on the magnitude and direction of the calculated power deviation, the droop coefficient used in the adaptive droop control strategy is dynamically adjusted. The specific value of the droop coefficient is positively correlated with the absolute value of the power deviation. Based on the adjusted droop coefficient, the real-time state of charge and health indicators of each energy storage unit, the total power deviation is decomposed to each energy storage unit according to a preset allocation ratio, forming the power regulation component that each energy storage unit needs to undertake. The calculated power regulation components of each energy storage unit are algebraically superimposed with the corresponding instruction baseline values ​​obtained from the rolling optimization instruction sequence to obtain the final real-time power instruction for each energy storage unit and issue it for execution.

[0051] In some embodiments, the calculation of power deviation relies on high-frequency sampling and matching. The real-time control layer samples the total output power of the energy storage cluster at a frequency much higher than that of the day-ahead optimization layer and the intraday rolling optimization layer (e.g., once per second). Simultaneously, the real-time control layer searches for the power command value at the timestamp strictly corresponding to the current moment from the rolling optimization command sequence received from the intraday rolling optimization layer. Power Deviation By formula: Where: characters Indicates the total power deviation of the energy storage cluster, character This represents the real-time measured value of the total output power of the energy storage cluster. This represents the current power command baseline value obtained from the rolling optimization command sequence. When When the actual output power of the energy storage cluster exceeds the planned command, there is a power surplus; when When the power deviation is low, it indicates that the actual output power of the energy storage cluster is insufficient, resulting in a power deficit. The magnitude and direction of the power deviation are key inputs for subsequent adaptive droop control strategies.

[0052] In practical implementation, dynamically adjusting the droop coefficient in the adaptive droop control strategy is the core element in responding to power imbalance. Droop coefficient The adjustment follows the absolute value of the power deviation. The principle of positive correlation means that the larger the absolute value of the power deviation, the greater the droop coefficient. The larger the value, the stronger the control effect. The adjustment process can be described as follows: , where the function A mapping relationship from the absolute value of power deviation to the droop coefficient is defined. This mapping relationship can be a piecewise linear function or other nonlinear functions. When the power deviation is positive, i.e., there is excess power, a droop coefficient adjustment strategy specifically for the excess power scenario is adopted. This strategy may focus on rapidly increasing the charging power to absorb the excess power. When the power deviation is negative, i.e., there is a power deficit, another droop coefficient adjustment strategy specifically for the power deficit scenario is adopted. This strategy may focus on rapidly increasing the discharging power to compensate for the power deficit. By dynamically adjusting the droop coefficient, the adaptive droop control strategy can adaptively change the control strength according to the severity of the power imbalance.

[0053] Optionally, proportionally distributing power deviations requires comprehensive consideration of various real-time factors. This is based on the adjusted droop coefficient. Real-time state of charge of each energy storage unit and health indicators of each energy storage unit , No. Power regulation components of each energy storage unit You can use the formula: Where: characters Indicates assignment to the first The power regulation component of each energy storage unit (positive values ​​indicate increased discharge or decreased charging, negative values ​​indicate increased charging or decreased discharge), character Indicates the state of charge adjustment factor, character Indicates the first Real-time state of charge of each energy storage unit, character Indicates the adjustment factor for the health index, characters Indicates the first The health index of each energy storage unit (a higher value indicates a better health status), characters Indicates the total number of energy storage units participating in the allocation, character This represents the calculated total power deviation, represented by the character. This represents the adjusted droop coefficient. In the formula, the denominator is the sum of the weights of all energy storage units, and the numerator is the weight of the first energy storage unit. The weight of each energy storage unit is determined by the reciprocal of its real-time state of charge (encouraging units with moderate state of charge to undertake more regulation tasks) and its health index. The sum of the power regulation components of each energy storage unit equals... That is, the total adjustment.

[0054] Table 2: Real-time Control Layer Power Allocation Calculation Table Refer to Table 2, which demonstrates a scenario with a power deficit ( The allocation process of the adaptive droop control strategy at a specific moment. The calculation assumes... Total adjustment demand is This means that the total output of the energy storage cluster needs to increase by 200kW. The "computational weight" of each unit is determined by... The "power regulation component" ΔPi of each unit is calculated based on the allocation ratio, and then superimposed with the "command reference value" to obtain the "final real-time command". It can be seen that unit 3, with a lower SOC (50%) and better health (0.8), receives the largest regulation component, while unit 4, with a very high SOC (90%) or poor health (0.5), receives the smallest regulation component. This reflects the principle of optimized allocation based on state of charge and health. The total output command of the cluster is 300kW, which is exactly 200kW more than the reference value of 100kW, making up for the 100kW shortfall and leaving a 100kW droop margin.

[0055] In some embodiments, the generation of real-time power commands involves superimposing an adjustment component with a reference value. The reference value for the power command for each energy storage unit at the current moment is obtained from the rolling optimization command sequence issued by the intraday rolling optimization layer. The power regulation component for each energy storage unit will be calculated from the adaptive droop control strategy. With the corresponding power command reference value Performing algebraic addition, i.e. ,character Indicates the first The final real-time power command for each energy storage unit. It also needs to undergo final verification of the power upper and lower limits and state-of-charge protection constraints of each energy storage unit. After the verification is passed, it is sent to the local controller of the corresponding energy storage unit for execution. Power regulation component The introduction of this feature enables the final instruction to quickly compensate for rapid power deviations that occur in real time and for which the rolling optimization layer has not yet responded, based on the rolling optimization instruction.

[0056] Understandably, the adaptive droop control strategy achieves a smooth transition between control at different time scales. The rolling optimization command sequence provides a power planning benchmark based on the latest forecasts, with time scales ranging from several minutes to tens of minutes. The adaptive droop control strategy handles real-time power deviations at the second to minute level. By superimposing the power regulation components calculated based on the droop coefficient, real-time state of charge, and health indicators onto the benchmark command, the real-time power command inherits the economy and planning of the upper-level optimization scheduling while also possessing the speed and autonomy to cope with real-time fluctuations. This superposition mechanism allows the actions of the real-time control layer and the commands of the intraday rolling optimization layer to smoothly connect in the dynamic process, avoiding abrupt changes in control commands and ensuring the stability of the energy storage system operation.

[0057] In one embodiment of the present invention, a preset adjustment range for the droop coefficient is defined, which specifies the lower and upper limits of the possible values ​​of the droop coefficient, namely the minimum droop coefficient and the maximum droop coefficient. A piecewise linear mapping relationship between the droop coefficient and the absolute value of the power deviation is established. Segmented intervals for the absolute value of the power deviation are set, including at least a first interval, a second interval, and a third interval, wherein the first interval is defined as the absolute value of the power deviation being less than a first threshold. A corresponding droop coefficient calculation rule is defined for each segmented interval. For the first interval, the corresponding droop coefficient calculation rule is: regardless of how the absolute value of the power deviation changes within the interval, a preset fixed value is used as the droop coefficient, namely the minimum droop coefficient. During system operation, the absolute value of the power deviation is calculated in real time. The calculated absolute value of the power deviation is compared with the preset first threshold. If it is determined that the absolute value of the power deviation is less than the first threshold, then according to the droop coefficient calculation rule defined for the first interval, the currently used droop coefficient is set to the minimum droop coefficient. When the absolute value of the power deviation is between the first threshold and a larger second threshold, i.e., it enters the second interval, the droop coefficient needs to increase linearly with the increase of the absolute value of the power deviation. When the absolute value of the power deviation exceeds the second threshold, i.e., it enters the third interval, and the maximum droop coefficient is applied. Furthermore, when the power deviation is positive, indicating a power deficit in the system, a droop coefficient adjustment strategy is employed for power deficit scenarios. When the power deviation is negative, indicating a power surplus in the system, a different droop coefficient adjustment strategy is employed for power surplus scenarios.

[0058] In practical implementation, a piecewise linear mapping relationship between the droop coefficient and the absolute value of the power deviation is established to achieve dynamic adjustment. The preset adjustment range of the droop coefficient defines the upper and lower limits of the possible values ​​for the droop coefficient, and includes a minimum droop coefficient and a maximum droop coefficient. Establishing the piecewise linear mapping relationship between the droop coefficient and the absolute value of the power deviation requires defining segmented intervals and calculation rules. The segmented intervals for the absolute value of the power deviation include at least a first interval, a second interval, and a third interval, where the first interval is defined as the absolute value of the power deviation being less than a first threshold. A corresponding droop coefficient calculation rule is defined for each segmented interval. For the first interval, the corresponding droop coefficient calculation rule is to use a preset fixed value as the minimum droop coefficient regardless of how the absolute value of the power deviation changes. During system operation, the real-time control layer calculates the absolute value of the power deviation in real time and compares the calculated absolute value of the power deviation with the preset first threshold. If it is determined that the absolute value of the power deviation is less than the first threshold, the droop coefficient is set to the minimum droop coefficient according to the droop coefficient calculation rule defined for the first interval. When the absolute value of the power deviation is between a first threshold and a larger second threshold, the droop coefficient increases linearly with the absolute value of the power deviation. When the absolute value of the power deviation is greater than the second threshold, the maximum droop coefficient is used. When the power deviation is positive, a droop coefficient adjustment strategy for power deficit is adopted; when the power deviation is negative, a droop coefficient adjustment strategy for power surplus is adopted.

[0059] In some embodiments, the adjustment range of the preset droop coefficient needs to be determined according to the specific characteristics of the control system. The minimum droop coefficient corresponds to the system's response mode to small power deviations. Setting the minimum droop coefficient to a small positive number makes the droop control characteristics smoother to avoid over-regulation of small fluctuations. The maximum droop coefficient corresponds to the system's mode requiring a fast and strong response to large power deviations. Setting the maximum droop coefficient to a large positive number provides stronger power regulation capability. The specific values ​​of the adjustment range are obtained by tuning the rated power of the energy storage cluster, the stability requirements of the grid frequency or voltage, and the tracking accuracy requirements of the power command. For example, in an energy storage cluster with a rated power of 1 MW, the minimum droop coefficient may be tuned to 0.01, and the maximum droop coefficient may be tuned to 0.05. The piecewise linear mapping relationship between the droop coefficient and the absolute value of the power deviation is formally described by a piecewise function. One mathematical expression of the piecewise function is: Where: characters This represents the calculated droop coefficient, character. This represents the preset minimum droop coefficient, character. Indicates the preset maximum droop coefficient, character Represents the absolute value of the power deviation calculated in real time, character Indicates the first threshold, character This represents the second threshold. The piecewise function explicitly specifies the method for determining the droop coefficient within different absolute power deviation ranges.

[0060] In practical implementation, defining the droop coefficient calculation rule for each segmented interval of the absolute power deviation is a crucial step. The segmented intervals of the absolute power deviation include at least a first interval, a second interval, and a third interval. The first interval is defined as when the absolute power deviation is less than a first threshold. For the first interval, the calculation rule for the corresponding droop coefficient is to use a fixed value, that is, regardless of how the absolute value of the power deviation changes within this interval, the droop coefficient is set to the minimum droop coefficient. The second interval is defined as the absolute value of the power deviation falling within the first threshold. With the second threshold Between. For the second interval, the calculation rule for its corresponding droop coefficient is the droop coefficient. absolute value of power deviation linearly increasing, from linear growth to The third interval is defined as the absolute value of the power deviation being greater than the second threshold. For the third interval, the calculation rule for its corresponding droop coefficient is to use another fixed value, that is, regardless of the absolute value of the power deviation, the droop coefficient is set to the maximum droop coefficient. These calculation rules are implemented in the real-time control layer through program logic or table lookup methods.

[0061] Optionally, the first and second thresholds can be set based on system operating data. First threshold This can be set to a small percentage of the energy storage cluster's rated power, such as 2% of the rated power, to define a range of negligible or minimal power deviations that do not require drastic adjustments. Second threshold This can be set to a large percentage of the rated power, such as 10% of the rated power, to define the range of severe power deviations requiring the activation of maximum regulation capacity. (Assuming the energy storage cluster's rated power...) Then you can set (2%) (10%). The absolute value of the power deviation calculated in real time. Since 15kW < 20kW, it falls within the first interval. According to the calculation rules, the droop coefficient is set to the minimum droop coefficient. For example, 0.01. When When 20kW≤60kW≤100kW falls within the second interval, the droop coefficient is calculated using the linear formula. ,but .when Since 150kW > 100kW, falling within the third range, the droop coefficient is directly set to the maximum droop coefficient. .

[0062] In practical implementation, different adjustment strategies are adopted for different power deviation directions to enhance the adaptability of control. When the power deviation is positive, it indicates that the actual total output power of the energy storage cluster is greater than the rolling optimization command value, indicating power excess. In this case, a droop coefficient adjustment strategy for power deficit is adopted. The droop coefficient adjustment strategy for power deficit may use a set of independent segmented interval thresholds and extreme values ​​of the droop coefficient, such as setting a small minimum droop coefficient and a large maximum droop coefficient, aiming to make a more sensitive and powerful response to power deficit, prompting energy storage units to increase discharge or reduce charging to make up for the deficit. When the power deviation is negative, it indicates that the actual total output power of the energy storage cluster is less than the rolling optimization command value, indicating power excess. In this case, a droop coefficient adjustment strategy for power excess is adopted. The droop coefficient adjustment strategy for power excess may use another set of parameters, such as setting a relatively large minimum droop coefficient to suppress overreaction to small power excesses and avoid unnecessary charging actions. The direction of power deviation is determined by comparing the real-time measured total power value with the command reference value. The direction information and the magnitude of the absolute value together determine which set of parameter mapping relationships is used to calculate the droop coefficient.

[0063] In some embodiments, the rule of minimum droop coefficient is used to implement the power deviation when the absolute value is less than a first threshold, implemented through comparison logic. During system operation, the real-time control layer calculates the absolute value of the power deviation in each control cycle. The calculated absolute value of the power deviation is compared with a pre-stored first threshold. Perform a numerical comparison. If the logical judgment result is that the absolute value of the power deviation is less than the first threshold, then... Then, the control logic switches to the branch for calculating the droop coefficient defined for the first interval. In this branch, according to the calculation rules for the droop coefficient defined for the first interval, the droop coefficient is directly assigned a preset fixed value, i.e., the minimum droop coefficient. This minimum droop coefficient It is used for subsequent calculation of the power regulation component. This mechanism ensures that when the power tracking deviation is very small, the droop control action is very gentle, mainly relying on upper-level optimization commands for adjustment, thus avoiding oscillations caused by the control system overreacting to measurement noise or small disturbances.

[0064] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A hierarchical collaborative scheduling optimization method for distributed energy storage, characterized in that, include: The system acquires operational status data and external environment information of the distributed energy storage system. The operational status data includes the state of charge, charge and discharge power boundaries, and health indicators of each energy storage unit. The external environment information includes predicted load curves, new energy power generation curves, and grid dispatch command curves. Based on the operational status data and the external environment information, a collaborative framework is constructed that includes a day-ahead optimization layer, an intraday rolling optimization layer, and a real-time control layer. In the aforementioned day-ahead optimization layer, an improved model predictive control algorithm is used to solve for the planned charge and discharge power curves of each energy storage unit over the next few days, with the goal of minimizing the total system operating cost. In the intraday rolling optimization layer, based on the updated ultra-short-term forecast information, the improved model predictive control algorithm is invoked to perform rolling correction on the planned charge and discharge power curve, generating a rolling optimization instruction sequence; In the real-time control layer, based on the deviation between the rolling optimization command sequence and the actual operating state, an adaptive droop control strategy is adopted to generate the final real-time power command for each energy storage unit. The final real-time power command is sent to the corresponding energy storage unit controller for execution.

2. The hierarchical collaborative scheduling optimization method for distributed energy storage according to claim 1, characterized in that, The working principle of the improved model predictive control algorithm includes: At the beginning of each optimization cycle of the day-ahead optimization layer, a rolling optimization problem containing an objective function and multiple constraints is constructed. The objective function is the total system operating cost function that includes electricity purchase and sale costs, energy storage loss costs, and power fluctuation penalty terms. The control time domain and prediction time domain of the improved model predictive control algorithm are initialized, wherein the control time domain includes multiple discrete control step sizes, and the prediction time domain is longer than the control time domain; Within each discrete control step, taking the current state of the energy storage system as the initial point, the rolling optimization problem is solved to obtain the optimal control sequence in the control time domain; The first control quantity of the optimal control sequence is used as the output command at the current moment and applied to the controlled energy storage system. The actual operating state of the energy storage system is fed back to the improved model predictive control algorithm as the initial state for the start of the next optimization cycle; Update some parameters in the objective function, and roll forward one control step, repeating the steps of solving the rolling optimization problem, outputting instructions, and providing feedback on the status.

3. The hierarchical collaborative scheduling optimization method for distributed energy storage according to claim 2, characterized in that, The construction of the rolling optimization problem, which includes an objective function and multiple constraints, includes: The total operating cost function of the system is constructed using a mathematical expression that includes the cost of electricity purchase, revenue from electricity sales, aging costs of energy storage unit charge-discharge cycles, and a penalty term for the total power ramp-up rate of the energy storage cluster. The first type of constraint is set, which includes upper and lower limits of the state of charge of each energy storage unit, upper and lower limits of charging and discharging power, and balance constraints of the state of charge at the beginning and end of the scheduling cycle. Set a second type of constraint, which includes upper and lower limits of the total power of the energy storage cluster and the power exchanged with the grid, as well as the power change rate constraint. A third type of constraint is set, which is based on the health index and sets differentiated power allocation weight coefficients for energy storage units with different health levels. The rolling optimization problem is formed by combining the total system operating cost function, the first type of constraint, the second type of constraint, and the third type of constraint.

4. The hierarchical collaborative scheduling optimization method for distributed energy storage according to claim 1, characterized in that, In the intraday rolling optimization layer, based on updated ultra-short-term forecast information, the improved model predictive control algorithm is invoked to perform rolling correction on the planned charge-discharge power curve, generating a rolling optimization instruction sequence, including: Receive updated ultra-short-term load forecast curves, ultra-short-term renewable energy generation forecast curves, and the latest real-time grid dispatch instructions; Based on the planned charge and discharge power curve, the improved model predictive control algorithm is re-called for optimization calculation within the shortened prediction time domain and the shortened control time domain. In the optimization calculation, a deviation penalty term for the planned charge and discharge power curve is introduced to ensure that the deviation between the rolling optimization instruction sequence and the planned charge and discharge power curve is within the allowable range. The optimization results obtained are time-scaled smaller than the current day's optimization step size, forming a rolling optimization instruction sequence for the next few hours.

5. The hierarchical collaborative scheduling optimization method for distributed energy storage according to claim 4, characterized in that, The step of re-invoking the improved model predictive control algorithm for optimization calculation includes: Within the shortened prediction time domain, higher time resolution is used to model the uncertainties of load, new energy generation, and grid commands; Within the shortened control time domain, based on the real-time available power and state of charge of the energy storage unit, the power allocation weight coefficients set based on the health index in the rolling optimization problem of the improved model predictive control algorithm are adjusted. Add a power fluctuation quadratic penalty term to the objective function to smooth the rolling optimization instruction sequence; An optimization solver with warm start is adopted, using the solution of the previous optimization cycle as the starting point for the current solution, thereby accelerating the solution process of the improved model predictive control algorithm.

6. The hierarchical collaborative scheduling optimization method for distributed energy storage according to claim 1, characterized in that, In the real-time control layer, based on the deviation between the rolling optimization command sequence and the actual operating state, an adaptive droop control strategy is adopted to generate the final real-time power command for each energy storage unit, including: Real-time acquisition of total output power of energy storage clusters, power of grid interconnection lines, and voltage and frequency of key busbars; Calculate the power deviation between the total output power of the energy storage cluster and the power value of the corresponding instruction value in the rolling optimization instruction sequence; Based on the magnitude and direction of the power deviation, the droop coefficient in the adaptive droop control strategy is dynamically adjusted, and the droop coefficient is positively correlated with the absolute value of the power deviation. Based on the adjusted droop coefficient, the real-time state of charge of each energy storage unit, and the health index, the power deviation is allocated proportionally to form the power regulation component of each energy storage unit. The power regulation component is superimposed with the instruction reference value at the corresponding moment in the rolling optimization instruction sequence to obtain the final real-time power instruction for each energy storage unit.

7. A hierarchical collaborative scheduling optimization method for distributed energy storage according to claim 6, characterized in that, Based on the magnitude and direction of the power deviation, the droop coefficient in the adaptive droop control strategy is dynamically adjusted, including: A preset adjustment range for the droop coefficient, wherein the adjustment range includes the minimum droop coefficient and the maximum droop coefficient; Establish a piecewise linear mapping relationship between the droop coefficient and the absolute value of the power deviation. When the absolute value of the power deviation is less than the first threshold, the minimum droop coefficient is used. When the absolute value of the power deviation is between the first threshold and the second threshold, the droop coefficient increases linearly with the absolute value of the power deviation. When the absolute value of the power deviation is greater than the second threshold, the maximum droop coefficient is used; When the power deviation direction is positive, a droop coefficient adjustment strategy is adopted to address the power deficit. When the power deviation is negative, a droop coefficient adjustment strategy is adopted to address power excess.

8. The hierarchical collaborative scheduling optimization method for distributed energy storage according to claim 3, characterized in that, The third type of constraint, based on the health index, sets differentiated power allocation weighting coefficients for energy storage units with different health levels, including: A mapping relationship is established between the health index and the health level of the energy storage unit. The health index is calculated based on the number of historical cycles, the capacity decay rate, and the internal resistance growth rate. The health level includes healthy, slightly decayed, moderately decayed, and severely decayed. Different power allocation weight coefficients are set for different health levels, with the highest power allocation weight coefficient for energy storage units in the healthy level and the lowest power allocation weight coefficient for energy storage units in the severely degraded level. A power allocation term is added to the objective function of the rolling optimization problem. The power allocation term represents the sum of squares of the deviations between the actual charging and discharging power of each energy storage unit and the product of its rated power and the power allocation weight coefficient. During the optimization process, by minimizing the power allocation term, energy storage units with higher health levels can bear a larger proportion of the charging and discharging power, while power protection is provided for energy storage units with lower health levels.

9. A hierarchical collaborative scheduling optimization method for distributed energy storage according to claim 7, characterized in that, The establishment of a piecewise linear mapping relationship between the droop coefficient and the absolute value of the power deviation, wherein when the absolute value of the power deviation is less than a first threshold, the minimum droop coefficient is used, includes: Set segmented intervals for the absolute value of power deviation, wherein the segmented intervals include at least a first interval, a second interval, and a third interval, wherein the first interval is defined as the absolute value of power deviation being less than a first threshold. A corresponding droop coefficient calculation rule is defined for each segmented interval. For the first interval, the corresponding droop coefficient calculation rule is: regardless of how the absolute value of the power deviation changes, a preset fixed value is used as the minimum droop coefficient. During system operation, the absolute value of the power deviation is calculated in real time; The calculated absolute value of the power deviation is compared with the first threshold. If the absolute value of the power deviation is determined to be less than the first threshold, then the droop coefficient is set to the minimum droop coefficient according to the droop coefficient calculation rule defined for the first interval.

10. A hierarchical collaborative scheduling optimization system for distributed energy storage, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the hierarchical collaborative scheduling optimization method for distributed energy storage as described in any one of claims 1 to 9.

Images