A method and system for hierarchical response and module-coordinated control of cascaded battery clusters based on asymmetric performance constraints
By adopting a hierarchical response and module collaborative control method for cascaded battery clusters with asymmetric performance constraints, the problem of inconsistency in the state between modules in the cascaded battery energy storage system is solved, achieving efficient hierarchical response and collaborative control, and improving the system's auxiliary service capabilities and module lifespan.
Patent Information
- Application Number
- CN202611388051.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-08
- Publication Date
- 2026-11-17
AI Technical Summary
Existing cascaded battery energy storage systems struggle to cope with the heterogeneity and inconsistency in the state of retired power battery modules when providing auxiliary services such as peak shaving, frequency regulation, and automatic power generation control. This leads to uneven output between modules, overcharging and discharging of modules in a weak state, and exacerbated secondary degradation. There is a lack of effective hierarchical response and collaborative control mechanisms.
A hierarchical response and module collaborative control method for tiered utilization of battery clusters based on asymmetric performance constraints is adopted. By collecting battery cabinet data, multi-scale power demand analysis and asymmetric response modeling are performed to construct a dynamic game model of battery clusters, realize module-level distributed collaborative control, and generate duty cycle commands for DC/DC converters.
It improves the response accuracy and feasibility of secondary battery energy storage systems in ancillary services, reduces the risk of secondary degradation between modules, and enhances the system's state safety and lifespan friendliness.
Smart Images

Figure CN123121901A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of energy storage system scheduling, specifically relating to a hierarchical response and module collaborative control method and system for cascaded utilization of battery clusters based on asymmetric performance constraints. Background Technology
[0002] With the accelerated implementation of the dual-carbon strategy, the installed capacity of new energy sources, represented by wind power and photovoltaics, continues to rise. New energy power generation is characterized by randomness, volatility, and intermittency. Its large-scale grid connection places higher demands on the power system's power balance, frequency stability, and safety margin, necessitating the allocation of energy storage resources with rapid response and flexible adjustment capabilities. Simultaneously, with the rapid development of the electric vehicle industry, the scale of retired power batteries continues to grow, making their efficient disposal and resource utilization a crucial issue with both environmental and economic value. Applying retired power batteries to energy storage systems after testing, recombination, and tiered configuration can not only reduce energy storage construction costs but also extend the battery's life-cycle utilization value, which is of great significance for promoting the consumption of new energy sources, supporting the construction of new power systems, and achieving carbon emission reduction throughout the entire life cycle.
[0003] However, compared to energy storage systems built with brand-new batteries, cascaded battery energy storage systems face more complex operation and management challenges in large-scale applications. Retired power batteries come from diverse sources and have different service histories, resulting in significant differences in their health status, state of charge, available capacity, internal resistance characteristics, temperature rise characteristics, and aging rates. This leads to strong heterogeneity and state inconsistency among battery clusters and modules. Especially when participating in ancillary services such as peak shaving, frequency regulation, and automatic generation control, the energy storage system needs to respond quickly and accurately to external power commands while avoiding overcharging and discharging of modules in poor health states to suppress secondary degradation and safety risks. If traditional uniform modeling and unified power allocation methods are still used, modules with poor health states or insufficient safety margins are easily assigned mismatched charging and discharging tasks, thus exacerbating secondary degradation and reducing the overall available capacity and ancillary service response reliability of the energy storage system.
[0004] Existing cascaded energy storage systems, when participating in ancillary services such as peak shaving, frequency regulation, and automatic power generation control, typically treat battery cabinets, battery clusters, or battery modules as equivalent energy storage units with similar performance for unified modeling and control. This approach struggles to adequately address the complex engineering characteristics of retired power batteries, including varying sources, different aging levels, significant differences in SOC / SOH states, and asymmetric charging and discharging capabilities. Particularly in scenarios where heterogeneous battery modules are connected in parallel to the DC bus via DC / DC converters, the continued use of traditional static allocation, unified power control, or symmetrical performance constraint methods can easily lead to problems such as difficulty in reasonably decomposing cluster-level response targets, uneven power output between modules, overcharging and discharging of modules in weak health states, and accelerated secondary degradation.
[0005] Specifically, the existing technology has the following main shortcomings: 1. Lack of a hierarchical response mechanism for ancillary services across the battery cabinet-battery cluster-battery module hierarchy. Existing methods primarily focus on power point tracking at the power station or battery cabinet level, typically allocating power based on single indicators such as rated capacity, current SOC, or rated power. This makes it difficult to effectively translate complex ancillary service commands such as peak shaving, frequency regulation, and automatic generation control into cluster-level response targets and module-level executable power. Therefore, under conditions where the states of battery modules used in tiered applications differ significantly, existing methods struggle to form a continuous closed-loop control system from task analysis and target decomposition to bottom-level execution.
[0006] 2. Lack of dynamic task matching and functional clustering methods for the heterogeneous states of batteries for secondary use. Retired power battery modules exhibit significant differences in SOC, SOH, SOE, usable capacity, temperature margin, PCS regulation margin, and power response capability. Existing methods often employ average or fixed-ratio allocation, making it difficult to adaptively match power characteristics based on the differences in power characteristics of high-frequency, mid-frequency, and low-frequency auxiliary service tasks. This can easily lead to modules with weak health, low margin, or insufficient response capability undertaking unreasonable power output, thereby exacerbating state inconsistency and lifespan degradation.
[0007] 3. There is a lack of cluster-level optimal output correction methods that consider uncertain disturbances, asymmetric performance boundaries, and lifetime costs. Existing cluster-level control strategies typically rely on relatively accurate system models or fixed performance constraints, making it difficult to fully characterize the differences in safety margins, temperature rise risks, and aging costs of secondary-use batteries during charging and discharging. When ancillary service commands are subject to random fluctuations or external disturbances, traditional control methods struggle to achieve a unified optimization between power point tracking accuracy, disturbance rejection capability, and lifetime friendliness.
[0008] 4. Lack of distributed collaborative control mechanisms for DC / DC parallel modules. For parallel battery modules with DC / DC interfaces, modules with different capacities, health states, and power boundaries have differentiated output conditions. However, existing methods mostly focus on the overall power response of the battery cluster and lack control methods that further decompose cluster-level targets into module-level reference power and drive each module to perform collaboratively. It is difficult to simultaneously satisfy cluster-level power balance, module-level power constraints, state consistency constraints, and interface execution constraints.
[0009] Therefore, how to construct a hierarchical response and collaborative control method for the highly heterogeneous nature of cascaded utilization batteries, while meeting the rapid response requirements of ancillary services such as peak shaving, frequency regulation and automatic power generation control, and realizing state adaptive allocation of heterogeneous modules, cluster-level robust optimization and module-level distributed collaborative control, is a key technical problem that urgently needs to be solved in the current engineering application of cascaded utilization energy storage systems. Summary of the Invention
[0010] To address the shortcomings of existing technologies, this invention provides a hierarchical response and module collaborative control method and system for secondary-use battery clusters based on asymmetric performance constraints. This method can dynamically plan clusters based on the health status of battery modules within a secondary-use battery energy storage power station. Using battery clusters as the reference unit, it adaptively decomposes and responds to peak shaving and frequency modulation ancillary service commands, thereby achieving the technical effects of delaying the aging of battery modules within the energy storage power station and improving the economic benefits of participating in ancillary services.
[0011] The present invention adopts the following technical solution.
[0012] A hierarchical response and module-coordinated control method for cascaded utilization of battery clusters based on asymmetric performance constraints includes: Collect relevant data on the battery cabinets of the cascade battery energy storage power station, including auxiliary service instructions and multi-source status within the battery cabinets, and preprocess the relevant data of the battery cabinet controllers to obtain a unified status dataset. Multi-scale power demand analysis is performed on the unified state dataset to obtain the frequency band task demand set. Asymmetric response capability model for frequency band tasks is performed on the heterogeneous battery modules in the battery cabinet to obtain the task adaptation matrix. Based on the frequency band task demand set and the task adaptation matrix, the heterogeneous battery modules are dynamically divided into battery clusters, and an initial state vector is generated according to the battery cluster division result. A dynamic game model for battery clusters based on asymmetric performance constraints is constructed. The initial state vector is input into the dynamic game model for battery clusters, and the optimal power output of the battery cluster and the reference power of the battery module are output. A distributed collaborative control strategy for battery modules within a cluster is constructed based on asymmetric performance constraints and optimal planning. The optimal output of the battery cluster and the reference power of the battery modules are optimized based on the distributed collaborative control strategy, and the duty cycle command of the DC / DC converter corresponding to each battery module is generated.
[0013] Preferably, the collection of relevant data from the battery cabinet of the cascaded battery energy storage power station includes auxiliary service instructions and multi-source status within the battery cabinet, and preprocessing of relevant data from the battery cabinet controller to obtain a unified status dataset, specifically including: Let the battery cabinet currently participating in the response of the cascaded battery energy storage power station be numbered as follows: The battery cabinet contains A heterogeneous, cascaded-use battery module; The auxiliary service instructions include the cabinet-level auxiliary service task vector of the q-th battery cabinet obtained by collecting auxiliary service instructions through the energy management system. The multi-source state within the battery cabinet includes the PCS state matrix of the q-th battery cabinet obtained by acquiring the operating state of the DC / DC converter through the energy storage converter. By collecting the multi-source operating status of the battery modules from the battery management system, the BMS state vector of the battery module in the q-th battery cabinet is obtained. ; The collected data undergoes preprocessing, including time synchronization, data validity marking, dimensional normalization, and state coding. This preprocessing transforms the collected data into a unified state vector corresponding to each battery cabinet, yielding the unified state vector of the q-th battery cabinet. :
[0014] A unified state dataset is obtained based on the unified state vector of each battery cabinet in a tiered battery energy storage power station.
[0015] Preferably, the step of parsing auxiliary service instructions based on a unified state dataset and dividing heterogeneous battery module battery clusters, and generating an initial state vector based on the battery cluster division results, specifically includes: Cabinet-level auxiliary service task vector Multi-scale power demand analysis based on response time constraints yields a set of frequency band task requirements. ; Based on the PCS state matrix of the q-th battery cabinet Modeling the asymmetric response capability of heterogeneous battery modules for frequency band tasks yields the task adaptation matrix. ; Frequency band-based task requirements set Task Adaptability Matrix The nq heterogeneous cascaded battery modules inside the battery cabinet are dynamically divided into battery clusters, resulting in battery clusters including high-frequency fast response clusters, medium-frequency regulation and tracking clusters, and low-frequency energy support clusters. Based on the battery cluster partitioning results, the power components of each frequency band are allocated to the battery modules within the corresponding battery clusters, and the original initial power of the battery modules is projected into the power domain to generate an initial state vector.
[0016] Preferably, the PCS state matrix Modeling the asymmetric response capability of heterogeneous battery modules for frequency band tasks yields the task adaptation matrix. Specifically, it includes: Based on the BMS state vector of the i-th battery module in the q-th battery cabinet The first state margin factor is calculated using a saturation function, including the charge state factor, health state factor, energy state factor, capacity factor, and temperature margin factor. Based on the calculated factors, construct the energy response gating function for the i-th battery module of the q-th battery cabinet. and energy deficiency penalty item :
[0017]
[0018] in, This is the set of first-state margin factors; The gating slope coefficient corresponding to the first state margin factor of the nth class; Let be the normalized value of the nth type of first state margin factor for the i-th battery module in the q-th battery cabinet. For frequency band tasks The admission threshold for the z-th indicator, and satisfy:
[0019] in, , , , Frequency band tasks The following are the SOC, SOH, SOE, capacity, and temperature margin admission thresholds; Based on the PCS state matrix of the q-th battery cabinet The second state margin factors, including current margin factor, power margin factor, duty cycle up-adjustment margin factor, and down-adjustment margin factor, are calculated using saturation functions. Power response gating functions are constructed based on the calculated factors. and power bottleneck penalty item :
[0020]
[0021] in, It is the set of second-state margin factors. The gating slope coefficient corresponding to the second state margin factor of the m-th class. It is the normalized value of the m-th type second state margin factor of the i-th battery module in the q-th battery cabinet; Combined with energy response gating function and energy deficiency penalty item , thus obtaining the energy response capability of the i-th heterogeneous battery module for frequency band task r; Combined with power response gating function and power bottleneck penalty item We obtain the power response capability of the i-th heterogeneous battery module for frequency band task r; Based on the energy response capability and power response capability of the battery module for frequency band task r, the adaptability of the first battery module to frequency band task r is calculated. The task adaptation matrix is obtained by combining the adaptability of tasks in each frequency band. .
[0022] Preferably, the frequency band-based task requirement set Task Adaptability Matrix The nq heterogeneous battery modules inside the battery cabinet are dynamically divided into battery clusters, specifically including: Construct the attribution variables for the battery modules, with each battery module undertaking only one frequency band task in each control cycle; Based on the task fit matrix Construct a battery module clustering cost function, and use the battery module clustering cost function to calculate the allocation of the i-th battery module to the battery cluster. The matching cost; Construct battery cluster task capability constraints based on the available power, available energy, and available ramping capability of the battery module; Based on the attribution variables, matching costs, and task capability constraints, construct a dynamic clustering optimization problem:
[0023] in, The cluster affiliation matrix, Let q be the number of battery modules for secondary use in the q-th battery cabinet. For frequency band task set, , These are the attribution variables of the i-th battery module in the q-th battery cabinet to the frequency band task r and the allocation of the i-th battery module to the battery cluster, respectively. The matching cost; Let be the available power of the i-th battery module in the q-th battery cabinet during frequency band task r. The power component of frequency band task r; , These are the high-frequency and low-frequency clustering decision variables for the i-th battery module in the q-th battery cabinet, respectively. Let i be the available energy of the i-th battery module under low-frequency energy support tasks. The energy demand corresponding to the low-frequency power component; Let i be the available ramp-up capability of the i-th battery module under high-frequency tasks. For the ramp-up requirements corresponding to high-frequency power components; Solve the dynamic clustering optimization problem to obtain the cluster affiliation matrix, and generate a set of battery clusters based on the cluster affiliation matrix.
[0024] Preferably, based on the battery cluster partitioning results, the power components of each frequency band are allocated to the battery modules within the corresponding battery clusters, and the original initial power of the battery modules is projected into the power domain to generate an initial state vector, as follows: Calculate the power sharing coefficient based on the available power of each battery module under the frequency band task it undertakes in the current control cycle; The original initial power of each battery module is calculated based on the cluster affiliation matrix and power sharing coefficient. The PCS-side physical power boundary of the i-th battery module is constructed based on the PCS state matrix, and a continuous safety reduction function for state margin is constructed to continuously correct the PCS-side physical power boundary. Calculate the charge and discharge safety availability factor of the battery module, and determine the executable power domain of the battery module based on the corrected physical power boundary of the PCS side and the charge and discharge safety availability factor. The original initial power of the module is projected onto the executable power domain to obtain the projected initial power planning vector of the module; the initial power of each battery cluster is calculated based on the initial power planning vector of the module and the cluster affiliation matrix, and the battery module state is aggregated based on the cluster affiliation matrix to obtain the battery cluster state vector; the initial power planning vector of the module, the cluster affiliation matrix, the battery cluster state vector and the frequency band task requirement set are combined to generate the initial state vector.
[0025] Preferably, the step of projecting the original initial power of the module onto the executable power domain to obtain the projected initial state vector of the module is as follows:
[0026]
[0027]
[0028]
[0029] in, This is the initial power vector of the projected module. , Let be the original initial power vector and the original initial power of the i-th battery module projected from the q-th battery cabinet, respectively. Width of the executable power range; The power change penalty coefficient; t is the current time. The sampling period; The target power of the q-th battery cabinet; , These are the lower limit and upper limit of the discharge power of the i-th battery module, respectively; This represents the available ramp capability of the i-th battery module under task r in its respective frequency band. It is a constant.
[0030] Preferably, the construction of the battery cluster dynamic game model based on asymmetric performance constraints, which inputs the initial state vector into the battery cluster dynamic game model and outputs the optimal power output of the battery cluster and the reference power of the battery module, specifically includes: The asymmetric power response state of the battery cluster is constructed based on the initial state vector and used as the state variable of the dynamic game model of the battery cluster. Based on the aforementioned state variables, cluster-level power correction, and lumped disturbance term, a dynamic game model of battery clusters is constructed, and the optimal strategy at the HJI saddle point is analyzed. Construct an Actor network, and based on the state variables and the saddle point optimal policy, construct an HJI approximate solution network in the form of integral reinforcement learning; A fixed-time convergent network weight update law is set to update the HJI approximation solution network and the Actor network online. The optimal output of the battery cluster is output based on the updated HJI approximation solution network and Actor network, and the reference power of the battery module is obtained by further allocating the optimal output of the battery cluster.
[0031] Preferably, the construction of the asymmetric power response state of the battery cluster specifically includes: Calculate the battery cluster based on the attribution variables of the i-th battery module to the frequency band task r. Average SOC and average SOE of the internal battery module; Calculate the battery cluster power tracking error based on the initial power of the battery cluster and the frequency band power components; According to battery clusters Calculation of discharge boundary adjustment coefficient based on average SOC and average SOE of internal battery module and charging boundary adjustment coefficient ; Combined with discharge boundary adjustment coefficient and charging boundary adjustment coefficient Construct battery clusters separately for power tracking errors of battery clusters Positive asymmetric performance boundary and negative asymmetric performance boundary :
[0032]
[0033] Where t is the current time, This is the start time of the current control cycle; , Battery clusters The positive and negative initial performance boundaries; , Battery clusters The positive and negative steady-state performance boundaries; , Battery clusters The positive and negative boundary contraction rates.
[0034] Preferably, the construction of the battery cluster dynamic game model and the analysis of HJI saddle points specifically includes: With battery clusters Taking the asymmetric performance boundary as the object, and using the asymmetric performance boundary as the asymmetric performance constraint, a cluster-level performance function is constructed after transforming the power tracking error constraint of the battery cluster. Constructing Infinite Time-Domain Performance Indices Based on Cluster-Level Performance Functions The cluster-level dynamic game optimization problem can be expressed as:
[0035] in, It is the optimal value function; This is an asymmetric error transformation state; For cluster-level power correction control input Perform minimization optimization; To address lumped disturbances Perform maximization modeling; Construct a Hamiltonian function based on the cluster-level performance function, and then apply these functions to the Hamiltonian function with respect to... and Find the extreme value to obtain the optimal strategy at the saddle point.
[0036] Preferably, the step of setting a fixed-time convergent network weight update law, so that the network weights converge to the neighborhood of the optimal policy within a time upper bound independent of the initial conditions, specifically includes: The policy errors of the control Actor and the policy errors of the perturbation Actor are constructed as follows:
[0037]
[0038] in, For Actor policy error; , These are the target cluster-level power corrections output by the Critic network and the actual cluster-level power corrections output by the Actor network, respectively. The weight error of the Actor network; , These are the estimated weights for the Critic and Actor networks, respectively. Set up a fixed-time convergent network weight update law, including a normalized Bellman regression vector and a fixed-time nonlinear correction operator; This allows the Bellman residual, the policy error of the control Actor, and the policy error of the perturbation Actor to converge to the zero neighborhood within a time bound independent of the initial weights.
[0039] Preferably, the distributed collaborative control strategy for battery modules within a cluster, based on asymmetric performance constraints and optimal planning, processes the optimal output of the battery cluster and the reference power of the battery modules to generate duty cycle control commands for the DC / DC converters corresponding to each battery module, specifically including: Construct an intra-cluster distributed communication topology based on the optimal output of the battery cluster and the reference power of the battery module, and determine the predefined execution time; Based on the module reference power and actual output power, a distributed cooperative error and its asymmetric performance constraints are constructed; the distributed cooperative error is transformed into an unconstrained error state, and the corresponding module dynamic model is constructed. Based on the module dynamic model, a predefined time-distributed reference control law is set, and a module-level HJB optimal compensation problem and an ADP approximation network are constructed. The ADP approximation network is updated online using a predefined time-stable network weight update law to generate duty cycle control commands for the DC / DC converters corresponding to each battery module.
[0040] Preferably, the distributed cooperative error and asymmetric performance constraints of the constructed module specifically include: The module local power tracking error is defined based on the module's final reference power and the current module output power, and a distributed cooperative error is constructed by combining the power tracking errors of neighboring nodes within the cluster. To address the differences in charging and discharging directions, module-level positive and negative performance boundaries are constructed. Define the performance shrinkage function, the positive initial envelope, and the negative initial envelope; Construct asymmetric performance boundaries, including positive steady-state error and negative steady-state error; Based on the asymmetric performance boundary, the module cooperative error is made to satisfy the asymmetric performance constraint. A piecewise normalized error is constructed based on the asymmetric performance boundary. as follows:
[0041] in, For distributed cooperative error , These are the positive steady-state error and the negative steady-state error, respectively. Construct asymmetric error transformation variables based on piecewise normalized errors.
[0042] This invention also proposes a hierarchical response and module cooperative control system for cascaded utilization battery clusters based on asymmetric performance constraints, used to implement the hierarchical response and module cooperative control method for cascaded utilization battery clusters based on asymmetric performance constraints, comprising: The multi-source status acquisition module is used to collect relevant data of the battery cabinet in the cascade utilization battery energy storage power station, including auxiliary service instructions and multi-source status in the battery cabinet, and to preprocess the relevant data of the battery cabinet controller to obtain a unified status dataset. The battery cluster partitioning module is used to perform multi-scale power demand analysis on the unified state dataset to obtain the frequency band task demand set, to model the asymmetric response capability of heterogeneous battery modules in the battery cabinet for frequency band tasks to obtain the task adaptation matrix, to perform dynamic battery cluster partitioning on the heterogeneous battery modules based on the frequency band task demand set and the task adaptation matrix, and to generate an initial state vector based on the battery cluster partitioning result. The battery cluster game module is used to construct a dynamic game model of battery clusters based on asymmetric performance constraints. The initial state vector is input into the dynamic game model of battery clusters, and the optimal output of battery clusters and reference power of battery modules are output. The collaborative control module is used to construct a distributed collaborative control strategy for battery modules within a cluster based on asymmetric performance constraints and optimal planning. Based on the distributed collaborative control strategy, it optimizes the optimal output of the battery cluster and the reference power of the battery modules, and generates duty cycle commands for the DC / DC converters corresponding to each battery module.
[0043] The present invention also proposes a terminal, including a processor and a storage medium; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method for hierarchical response and module collaborative control of cascaded battery clusters based on asymmetric performance constraints.
[0044] The present invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for hierarchical response and module cooperative control of cascaded utilization battery clusters based on asymmetric performance constraints.
[0045] The beneficial effects of this invention are that, compared with the prior art, it proposes a hierarchical response and module collaborative control method and system for cascaded utilization battery clusters based on asymmetric performance constraints. This method constructs a hierarchical response optimization task at the battery cluster level that integrates asymmetric performance constraints, obtaining module response weights, reference power, and virtual state reference values through reinforcement learning. At the module level, it combines state filtering, dynamic adaptive planning, collaborative output planning, and distributed control to drive parallel modules with DC / DC interfaces to achieve differentiated power output. This allows the battery cluster to meet the requirements for rapid response in ancillary services while maintaining operational consistency among modules and reducing the risk of secondary degradation of cascaded utilization batteries.
[0046] Compared with existing technologies, this invention constructs a hierarchical response and collaborative control system that includes multi-source state / ancillary service instruction parsing, heterogeneous module dynamic clustering, battery cluster dynamic game optimization, and intra-cluster distributed collaborative control. This system can fully utilize the differences in SOC, SOH, SOE, available capacity, PCS regulation margin, and power response capability among secondary battery modules, adaptively decomposing battery cabinet-level ancillary service tasks into power demands of different frequency bands and dynamically matching them to battery clusters suitable for undertaking the corresponding tasks. Furthermore, this invention further utilizes asymmetric performance constraints, HJI / HJB optimal control, integral reinforcement learning, and practical predefined time-stability control to achieve continuous closed-loop control from battery cabinet-level task decomposition and battery cluster-level output optimization to module-level distributed execution. This improves the response accuracy, power executability, state safety, and lifetime friendliness of secondary energy storage systems participating in ancillary services such as peak shaving, frequency regulation, and automatic power generation control. The present invention has at least the following beneficial effects: 1. By decomposing auxiliary service instructions into high-frequency, medium-frequency and low-frequency power tasks, and dynamically clustering functions based on the module's SOC, SOH, SOE, PCS adjustment margin and power response capability, the battery modules with different state characteristics can undertake tasks that match their own capabilities, avoiding the overcharging and discharging of weak health modules caused by traditional uniform power allocation, and improving the refinement and adaptability of task allocation. 2. This invention transforms the battery cluster power correction problem into a dynamic game problem of HJI under asymmetric performance constraints, and uses integral reinforcement learning and fixed-time convergent network weight update strategy for online solution. It can obtain the optimal cluster-level output correction amount that is resistant to disturbances and has a friendly lifetime without relying on an accurate battery cluster model, thereby improving the power tracking accuracy and robustness under complex auxiliary service conditions. 3. This invention incorporates module-level power point tracking and neighbor cooperative error into a distributed control framework, and combines asymmetric performance constraints, practical predefined time-stability control, and HJB optimal compensation, enabling each battery module to quickly track the final reference power under local communication conditions, reducing the pressure of centralized control, and improving the adaptability of the DC / DC execution layer to disturbances, communication errors, and parameter changes. Attached Figure Description
[0047] Figure 1 This is a flowchart of the hierarchical response and module collaborative control method for tiered utilization of battery clusters based on asymmetric performance constraints in this invention; Figure 2 This is a flowchart of the auxiliary service instruction parsing and heterogeneous battery module dynamic clustering process in this invention; Figure 3 This is a flowchart of the process for generating the initial state vector in this invention; Figure 4 This is a flowchart of the solution and output planning strategy formulation for the dynamic game optimization problem of battery clusters based on asymmetric performance constraints in this invention; Figure 5 This is a flowchart of the distributed collaborative control strategy for intra-cluster battery modules based on asymmetric performance constraints and optimal planning in this invention. Figure 6 This is a structural diagram of the hierarchical response and module collaborative control system for cascaded utilization of battery clusters based on asymmetric performance constraints in this invention. Detailed Implementation
[0048] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, other embodiments obtained by those skilled in the art without creative effort are all within the protection scope of this invention.
[0049] like Figure 1 As shown, this invention proposes a hierarchical response and module collaborative control method for cascaded utilization of battery clusters based on asymmetric performance constraints. The method includes the following steps: Step S10: Collect relevant data from the battery cabinet controller. The collected data includes auxiliary service commands and the status of multiple sources within the battery cabinet. Preferably, step S10 is performed on the battery cabinet controller, and the relevant data collected from the battery cabinet controller specifically includes: collecting auxiliary service commands through the energy management system (EMS), collecting the DC / DC (Direct Current to Direct Current) converter operating status through the power conversion system (PCS), and collecting the multi-source operating status of the battery modules through the battery management system (BMS).
[0050] Among them, the battery cabinet number currently participating in the response of the cascaded battery energy storage power station is set as follows: The battery cabinet contains There are several battery modules for secondary use, with the i-th battery module denoted as... ,in: .
[0051] Specifically, the data collection methods for the battery cabinet controller are as follows: The data collected by EMS includes: auxiliary service response tasks assigned to the current battery cabinet based on the power plant-level multi-objective optimization results. These auxiliary service response tasks include cabinet-level target power. Target power change Power ramp-up requirements Response time requirements Allowable tracking error and the target range of cabinet-level State of Charge (SOC). This ancillary service response task is used to characterize the power response requirements that the current battery cabinet should undertake under peak shaving, frequency regulation, or automatic generation control scenarios, and constitutes the cabinet-level ancillary service task vector. .
[0052] The data collected by the PCS includes: DC / DC input voltage. Input current Output voltage Output current Output power Duty cycle Duty cycle limit Duty cycle lower limit Output current limit Output power limits Converter temperature and work mode The above data is used to characterize the executable power boundary and dynamic regulation margin when heterogeneous battery modules are connected to the DC bus in the cabinet via a DC / DC converter, and constitutes the PCS state matrix. .
[0053] Multi-source data collected through BMS includes: battery module state of charge. State of health (SOH) State of energy (SOE) Available capacity ,Voltage Current Module output power Average temperature and work mode And define the SOC margin of the battery module as... ,in This represents the SOC margin of the i-th battery module from its discharge lower limit. Let SOC be the SOC margin of the i-th battery module from its charging limit; define the SOH margin of the battery module as: ,in Let SOH be the SOH margin of the i-th battery module from the limit of its healthy state; define the thermal safety margin as: ,in This represents the maximum allowable temperature for the i-th battery module. The BMS state vector for the i-th battery module can be defined as:
[0054] The module battery state matrix can be constructed from the BMS state vectors of all modules. It characterizes the differences between each battery module in terms of state of charge, state of health, state of energy, state of capacity, thermal state, and operating mode.
[0055] Furthermore, after data acquisition, the data from EMS, BMS, and PCS undergo time synchronization, data validity marking, dimensional normalization, and status coding to form a unified state including cabinet-level auxiliary service task vectors, module battery status matrices, and PCS status matrices. This unified state dataset will then be used as input for auxiliary service instruction parsing and dynamic clustering of heterogeneous battery modules in step S20.
[0056] Through step S10, the present invention uniformly collects and encodes the cabinet-level auxiliary service tasks allocated by the power station-level EMS and the battery status, safety status, and PCS executable status of each battery module inside the battery cabinet. This enables the subsequent dynamic clustering process to not only rely on battery statuses such as SOC and SOH, but also to simultaneously consider PCS duty cycle margin, power limits, and operating modes, thereby providing reliable input for frequency band matching, dynamic clustering, and distributed collaborative control of heterogeneous battery modules.
[0057] Step S20: Auxiliary service instruction parsing and dynamic clustering of heterogeneous battery modules; Preferred, Reference Figure 2 As illustrated, step S20 can be further subdivided into four sub-steps.
[0058] Step S201: Analyze the multi-scale power demand of auxiliary service instructions based on response time constraints, and obtain high, medium, and low frequency components after analysis. , , Frequency band task requirements set ; First, from the cabinet-level auxiliary service task vector Extract the target power of the current battery cabinet Response time requirements Power ramp-up requirements and allowable tracking error .
[0059] Let the rolling control window at the current time t be: ,in, For window length, To control the cycle, the target power sequence of the battery cabinet within the current window is determined. for: .
[0060] For the target power sequence of the battery cabinet Construct a multi-scale decomposition filter H: , in, B represents the number of filters. For the b-th filter in delay The filter weights at each location can be pre-set using a finite impulse response filter bank and must satisfy the following conditions: , The function is a discrete impulse function. Based on this, the target power of the battery cabinet is decomposed into multiple scales using the above filter to obtain the power component at the b-th scale. .
[0061] For each power component Calculate its equivalent center frequency within the current scrolling window. :
[0062] in: The spectral energy of the current power. For the corresponding frequency range, To prevent small positive numbers with odd denominators, For frequency.
[0063] Secondly, based on the ancillary service response time requirements Constructing high-frequency thresholds With low frequency threshold : ,
[0064] in: , The frequency band boundary coefficient satisfies .
[0065] Based on the equivalent center frequency Power components at each scale Divided into high-frequency sets Mid-frequency collection and low frequency sets : , ,
[0066] This reconstruction yields the high-frequency fast response power component. Intermediate frequency regulation power component and low-frequency energy support power components : , ,
[0067] Specifically, for any frequency band Definition: Energy demand Climbing requirements And set response time requirements: Tolerance requirements ,in The response time required for frequency band r, To allow for tracking errors.
[0068] This allows us to form a frequency band task requirement vector. Frequency band task requirements set This set is used in step S202 to calculate the adaptability of each battery module to different frequency band tasks.
[0069] Step S202: Model the asymmetric response capability of heterogeneous battery modules for frequency band tasks to obtain the task adaptation matrix; First, based on the information obtained in step S10 The charge state factor is calculated using the saturation function. , Health status factors Energy state factor capacity factor Temperature margin factor ,by For example, the saturation function is set as follows: .
[0070] Secondly, based on the PCS state matrix of step S10 The current margin factor is calculated using the saturation function. Power margin factor Duty cycle adjustment margin factor Lower the margin factor .
[0071] Definition of the first A heterogeneous battery module for frequency band tasks The energy response capability is optimized. To avoid excessive compression of the response capability due to the direct multiplication of multiple normalization margin factors, this invention separates the module safety access conditions from the capability evaluation process and constructs an energy response gating function. and shortcomings penalty items :
[0072]
[0073] in, Frequency band tasks The SOC, SOH, SOE and temperature margin thresholds are set below.
[0074] Combined with energy response gating function and shortcomings penalty items The energy response capability of the i-th heterogeneous battery module for frequency band task r can be obtained. :
[0075] in, For the energy evaluation weights of frequency band task r, satisfying ,and , This is the energy penalty coefficient. is the normalization factor for the i-th module.
[0076] Next, define the first Each module performs frequency band tasks. The power response capability. Similar to the energy response capability, a power response gating function is constructed. and shortcomings penalty items :
[0077]
[0078] Therefore, the power response capability of the i-th heterogeneous battery module for frequency band task r can be obtained. :
[0079] in, Let be the power evaluation weight for frequency band task r, satisfying ,and , This is the power penalty coefficient.
[0080] Finally, the fit of the i-th module to the frequency band task r is defined. :
[0081] in, . This is the penalty coefficient for the imbalance between power response capability and energy response capability.
[0082] To ensure that the fit is within the effective range, amplitude limiting is applied: The effective adaptability of all modules to the three frequency band tasks is combined into a task adaptability matrix. This matrix serves as the capability input for dynamic functional clustering in step S203.
[0083] Step S203: Generate a battery cluster set based on the dynamic functional clustering of heterogeneous battery modules with optimal constraints; In step S203, based on the frequency band task requirement set in step S201... And the task effective fit matrix in step S202 For the nq heterogeneous battery modules inside the battery cabinet, a dynamic programming method is used to construct the cluster to which the battery modules belong, and the set of battery clusters in the current control cycle is obtained.
[0084] The battery clusters consist of a high-frequency fast response cluster, a mid-frequency regulation tracking cluster, and a low-frequency energy support cluster, denoted as follows: .
[0085] First, construct the attribution variables for the battery modules. Specifically, define the attribution variables for the i-th battery module in relation to frequency band task r: ,in This indicates that i battery modules belong to the battery cluster within the current control cycle. ;on the contrary, This indicates that i battery modules do not belong to the battery cluster during the current control cycle. Based on this, to ensure that the battery module undertakes only one frequency band task within the control cycle, mutual exclusion constraints are set: .
[0086] Secondly, a clustering cost function for battery modules is constructed. Specifically, based on the fitness level in step S202... Construct the i-th battery module and allocate it to the battery cluster. Matching cost :
[0087] in, This is the cluster switching penalty coefficient, used to suppress battery modules from frequently changing their functional cluster affiliation within adjacent control cycles. The degree of compatibility between the battery module and the frequency band task is characterized. The higher the degree of compatibility, the lower the matching cost. The switching penalty term is used to characterize the scheduling disturbance caused by the change of module functional cluster affiliation.
[0088] Next, task capability constraints are constructed for each battery cluster. Specifically, task capability constraints are set for each battery cluster based on the module's available power, available energy, and available ramping capability obtained in step S202.
[0089] For any frequency band task r, set a power constraint:
[0090] in, Let be the available power of the i-th battery module in frequency band task r. The frequency band power component obtained in step S201.
[0091] In particular, for low-frequency energy support clusters Set energy constraints:
[0092] in, Let i be the available energy of the i-th battery module under low-frequency energy support tasks. This represents the energy demand corresponding to the low-frequency power component.
[0093] For high-frequency fast response clusters Set climbing constraints:
[0094] in, Let i be the available ramp-up capability of the i-th battery module under high-frequency tasks. This refers to the ramp-up requirements corresponding to the high-frequency power components.
[0095] Next, a dynamic clustering optimization problem is constructed and solved. Specifically, based on the aforementioned attribution variables, matching costs, and task capability constraints, a dynamic clustering optimization problem is constructed:
[0096] in, Cluster affiliation matrix:
[0097] By solving the above optimization problem, the cluster affiliation of each battery module in the current control cycle can be obtained.
[0098] Finally, a set of battery clusters is formed. Specifically: based on the cluster affiliation matrix... Generate a set of battery clusters: The high-frequency fast response cluster, the mid-frequency regulation tracking cluster, and the low-frequency energy support cluster are respectively: , , .
[0099] The final output vector for step S203 is: This serves as the input for step S204.
[0100] Step S204: Initial power projection and state vector generation of modules for cluster-level optimization.
[0101] In step S204, based on the battery cluster division results obtained in step S203, the power components of each frequency band obtained in step S201 are allocated to the battery modules inside the corresponding battery cluster, and the initial power of the module is projected into the power domain to generate the initial state vector required in step S30.
[0102] refer to Figure 3 Step S204 specifically includes: Step S2041: Determine the power sharing weight of modules within the functional cluster; For any frequency band task r, battery cluster Based on the available power of each battery module under the operation in this frequency band, construct the power sharing coefficient of the i-th battery module. :
[0103] From the above definition, we can obtain: ,when Sometimes, That is, the i-th battery module does not participate in the initial power sharing of frequency band task r.
[0104] Step S2042: Calculate the original initial power of the module; Calculate the initial power of the i-th battery module based on the cluster affiliation matrix and power sharing coefficient:
[0105] Due to the attribution constraint, only one frequency band term in the formula is effective for the i-th battery module, meaning that each battery module only forms one original initial power in the current control cycle.
[0106] Furthermore, since the weighting within each frequency band task is normalized, the sum of the original initial power of all modules equals the sum of the power components of the three frequency bands. Also, since the high-frequency, mid-frequency, and low-frequency power components in step S201 can be reconstructed into the cabinet-level target power, therefore... .
[0107] Step S2043: Determine the physical power boundary on the PCS side of the module; Based on the PCS data collected in step S10, the power boundary of the power converter corresponding to the i-th battery module is first constructed. The upper limit of the physical power of the PCS during discharge is defined as:
[0108] in, The adjustable power limit is determined by the duty cycle adjustment margin.
[0109] Similarly, the upper limit of the physical power of the PCS during charging is defined as follows:
[0110] in, The adjustable power limit is determined by the duty cycle reduction margin.
[0111] Step S2044: Establish a continuous safety reduction function for state margin; To continuously correct the PCS physical power boundary based on the battery module's state margin, an arbitrary state margin index is defined. Sigmoid type continuous safety reduction function :
[0112] in, For discharge, To charge; This is the slope coefficient of the reduction function; This is the safe access threshold for the state margin indicator.
[0113] Step S2045: Calculate the module's charging and discharging safety availability factor; During charging, the performance is jointly determined by the charging SOC margin, SOH, SOE, temperature margin, output current margin, and output power margin. The set of charging parameters is defined as follows: Similar to the frequency band task penalty, a charging penalty is defined. Based on this, a charging safety factor can be defined. ,in , This is the charging penalty coefficient.
[0114] Similarly, define a set of discharge direction indices: Define charging penalty items Define the charging safety factor. ,in , This represents the discharge penalty coefficient.
[0115] Step S2046: Determine the module's executable power domain; Based on the PCS physical power boundary and the continuous safety availability coefficient, the upper limit of the discharge power of the i-th battery module is defined as follows: With lower power limit: This forms the executable power domain for the i-th battery module: ,in Let i be the maximum allowable discharge power of the i-th battery module. This represents the negative boundary corresponding to the maximum allowed charging absorption power of the i-th battery module. Further, the width of the executable power range for this module is defined as follows: .
[0116] Step S2047: Correct the module's initial power to the executable power domain; To ensure that the initial power satisfies the module's executable power domain, power conservation constraints, and power change rate constraints, the following projection optimization problem is constructed:
[0117] in, This is the initial power vector of the projected module. This indicates the current battery cluster type to which the i-th battery module belongs. This represents the available ramping capability of the i-th battery module under its assigned frequency band task r.
[0118] In the aforementioned optimization problem, the first term is used to keep the projected power planning value close to the original initial power, and the second term is used to suppress power abrupt changes between adjacent control cycles. The constraint terms are used to ensure the conservation of battery cabinet-level power, the satisfaction of module power boundaries, and the satisfaction of module power change rate. In addition, this optimization problem does not contain binary variables, and the nonlinear constraint terms can all be transformed into linear constraints using the Big M method, thereby transforming the optimization problem into a quadratic programming problem, which can achieve optimal planning in a very short time.
[0119] Step S2048: Calculate the initial power of the battery cluster; Based on the projected initial power planning vector of the module, the initial power of each battery cluster is calculated: , Specifically: , , ; And satisfy .
[0120] Step S2049: Generate dynamic game optimization input state for battery clusters.
[0121] For any battery cluster Construct state vector :
[0122] in, For cluster SOC, For cluster SOH, SOE, This is the highest temperature of the cluster; And the cluster SOC is defined as: ; Cluster SOH is defined as: ; Cluster SOE is defined as: ; The highest temperature of a cluster is defined as .
[0123] Based on this, the module's initial power planning vector, cluster affiliation matrix, battery cluster state vector, and frequency band task requirement set are combined into a state: Then the output of step S204 is: This is then used as input for solving the dynamic game optimization problem of battery clusters based on asymmetric performance constraints and formulating the output planning strategy in step S30.
[0124] Step S30: Solving the dynamic game optimization problem of battery clusters based on asymmetric performance constraints and formulating the output planning strategy; like Figure 4 As shown, step 3, based on the battery cluster partitioning results and initial module power planning values determined in step S20, constructs asymmetric performance boundaries for charging and discharging according to the average SOC, average SOE, and SOH distribution of each battery cluster. These asymmetric performance boundaries constrain the battery cluster power tracking error. Through error normalization and logarithmic transformation, the constrained error is converted into an unconstrained error state, which serves as the core state variable for the subsequent disturbance-resistant zero-sum differential game model. Using the battery cluster power tracking error as the core state, a disturbance-resistant zero-sum differential game model is established. This model uses the battery cluster power correction as the control input and the lumped disturbance as the adversarial input, constructing a system that includes power tracking... Performance indicators for error, control cost, disturbance suppression, and aging loss are calculated. Based on the principle of dynamic programming, the optimal value function satisfies the Hamilton-Jacobi-Bellman-Isaacs (HJI) equation. The HJI equation is solved online approximately using integral reinforcement learning and adaptive dynamic programming methods. Finally, the optimal output correction, optimal cluster-level reference power, and final module reference power are generated for each battery cluster. The optimal cluster-level reference power is obtained by superimposing the initial power of the battery cluster with the optimal output correction and then correcting for power conservation. The final module reference power is obtained by further distributing the optimal cluster-level reference power according to the power sharing coefficient within the cluster.
[0125] Step S301: Construct the asymmetric power response state of the battery cluster; First, for any battery cluster Based on the unique attribution variable obtained in step S203 Calculate battery clusters Average SOC and average SOE of the internal battery module: ,
[0126] In this step, the average SOC and average SOE are not used as tracking targets, but as adjustment factors for the battery cluster's charge and discharge capabilities, asymmetric performance boundaries, and aging costs.
[0127] Secondly, based on the initial power of the battery cluster obtained in step S204 and the frequency band power components obtained in step S201 Constructing battery cluster power tracking error .
[0128] Furthermore, according to and Constructing the discharge boundary adjustment coefficient :
[0129] and charging boundary adjustment coefficient :
[0130] in, , , , Adjust the weights for the boundary conditions, and satisfy: .
[0131] Secondly, regarding the power tracking error of the battery cluster. Construct positive and negative asymmetric performance boundaries respectively:
[0132]
[0133] in, This is the start time of the current control cycle; , These are the positive and negative initial performance boundaries, respectively; , These represent the positive and negative steady-state performance boundaries, respectively. , These represent the positive and negative boundary contraction rates, respectively.
[0134] Furthermore, the power tracking error of the battery cluster satisfies: To transform the constrained power error into an unconstrained variable, a piecewise normalized error is constructed:
[0135] Based on this, a logarithmic error transformation is constructed:
[0136] Therefore, As the core state variable in the dynamic game optimization of battery clusters, and its continuous-time dynamic representation is as follows:
[0137] in, Let be the cluster-level power correction control variable to be solved. This is the lumped disturbance term formed by ancillary service power fluctuations, PCS dynamic errors, module state estimation errors, and unmodeled decay dynamics. , , These are the equivalent drift term, the control input term, and the disturbance input term, respectively.
[0138] Step S302: Construct a dynamic game model of battery clusters and analyze the HJI saddle point; With battery clusters For the object, the cluster-level power correction problem is modeled as a continuous-time zero-sum differential game problem. The cluster-level performance function is defined as follows:
[0139] in, As the power error weight for battery clusters, The cluster-level power correction cost weight matrix, The level of disturbance suppression. The cluster-level aging loss penalty term can be defined as:
[0140] in, For frequency band tasks The aging loss weight is below. To select a vector that maps cluster-level control vectors to power corrections, The power sharing factor for modules within a cluster is defined as:
[0141] Therefore, an infinite time-domain performance index is constructed: The cluster-level dynamic game optimization problem is expressed as: Then the Hamiltonian function can be constructed:
[0142] in, express right The gradient.
[0143] The optimal value function satisfies the HJI equation: For the Hamiltonian function, respectively with respect to and Find the extreme value to obtain the optimal strategy at the saddle point:
[0144]
[0145] because Since the HJI equation is difficult to obtain precisely, an online approximate solution is obtained by using integral reinforcement learning and adaptive dynamic programming methods.
[0146] Step S303: Construct an HJI approximate solution network based on integral reinforcement learning; Selecting online data sampling intervals Integrating the HJI equations along the actual trajectory of the system yields the integral Bellman equations:
[0147] Then a critic network can be constructed to approximate the optimal value function. :
[0148] in, The Critic network basis function vectors, For Critic network weight estimates The superscript T in this application indicates transpose.
[0149] Therefore, the integral Bellman residual is defined as:
[0150] Define the Bellman regression vector:
[0151] but:
[0152] The above integral Bellman residual does not explicitly contain unknown drift dynamics. Its calculation relies on online acquired error state, control input, and disturbance estimation data, and can be directly used for iterative solution.
[0153] Furthermore, construct a network to control the actors:
[0154] Constructing a perturbation Actor network:
[0155] in, , These are the basis function vectors for controlling the Actor network and perturbing the Actor network, respectively. Control the transpose of the network weights corresponding to the Actor network and the perturbation Actor network respectively.
[0156] The target control strategy induced by the Critic network is as follows:
[0157] The target perturbation strategy induced by the Critic network is as follows:
[0158] Step S304: Design the weight update law for a fixed-time convergent network; Define the policy error for controlling the Actor network and perturbing the Actor network. and They are respectively:
[0159]
[0160] To enable the Critic network, the control Actor network, and the perturbation Actor network to complete policy approximation within a given time bound, this invention designs a fixed-time convergent network weight update law.
[0161] Define the normalized Bellman regression vector:
[0162] Define a fixed-time nonlinear correction operator:
[0163] in, , , , , , Both represent element-wise exponentiation.
[0164] The Critic network weight update law is:
[0165] in, For Critic learning rate, This is the coefficient of the Critic weight leakage term.
[0166] The law governing the weight update of the Actor network is:
[0167] in, To control the Actor's learning rate, To control the coefficient of the leaked Actor weight.
[0168] The perturbation Actor network weight update law is:
[0169] in, To perturb the Actor's learning rate, This is the coefficient of the perturbation Actor weight leakage term.
[0170] To ensure that network weights are bounded, a projection operator is introduced:
[0171] in, This represents the weight update law for the network without projection. This represents the set of allowed network weights.
[0172] Under finite-time data with sufficient excitation, the fixed-time nonlinear correction term causes the Bellman residual, the control Actor policy error, and the perturbation Actor policy error to converge to the zero neighborhood within a time upper bound independent of the initial weights. Its convergence time upper bound satisfies:
[0173] Step S305: Generate the optimal output power of the battery cluster and the final reference power of the module.
[0174] Based on the output of the control Actor network in step S304, the battery clusters are extracted. The optimal power correction amount: Thus, the preliminary optimal reference power of the battery cluster is obtained:
[0175] To ensure that all battery clusters still satisfy the cabinet-level power conservation constraint after correction, the cluster-level correction output of the Actor network is normalized to maintain this constraint. And obtain the final optimal reference power of the battery cluster: Therefore, the following conditions are met: .
[0176] Furthermore, the optimal reference power at the cluster level is allocated to the battery modules within the cluster. For the first... For each module, based on the available power of the module in step S204, the final power sharing coefficient of the module within the cluster is constructed:
[0177] Therefore, we obtain the first... The final reference power of each module: Due to the unique ownership constraint: The above formula contains only one frequency band task item for the first... Each module is active, meaning that each module ultimately receives only one reference power.
[0178] Step S40: Distributed cooperative control strategy for battery modules within a cluster based on asymmetric performance constraints and optimal planning.
[0179] Reference Figure 5As illustrated, step 4 establishes the optimal reference power of each battery cluster, the final reference power of the module, and the virtual state of the functional cluster as described in step S30. For the battery modules within the cluster, a distributed communication topology is constructed with the battery modules as nodes. Taking the module's local power tracking error and the neighbor node's cooperative error as the core states, a module-level asymmetric performance constraint oriented towards charging / discharging differences is constructed. The constrained distributed cooperative error is mapped to an unconstrained control variable through error transformation. Based on this, a practical predefined time-distributed reference control law is designed to ensure that the module power error within the cluster converges to the allowable neighborhood within a given time upper bound. Furthermore, a module-level Hamilton-Jacobi-Bellman (HJB) optimal compensation problem is constructed, and adaptive dynamic programming is employed. The ADP (Advanced Programming Process) method is used to solve the optimal compensation control law online. At the same time, a practical predefined time-converged network weight update law is designed to enhance the ability of the underlying controller to suppress DC / DC parameter perturbations, sampling errors, communication disturbances and unmodeled power dynamics. Finally, the duty cycle control command of the DC / DC converter corresponding to each battery module is generated to realize the distributed cooperative tracking of the optimal reference power in step S30 by the battery modules within the cluster.
[0180] Preferred, Reference Figure 5 Step S40 includes the following sub-steps: Step S401: Construct the intra-cluster distributed communication topology and predefined execution time; First, for any battery cluster: The battery modules and their DC / DC converters belonging to the battery cluster are regarded as distributed control nodes, and an intra-cluster communication graph is constructed. The node set is as follows: edge set This represents the information exchange relationship between modules within the same battery cluster. Adjacency coefficient is defined. ,in, This represents the communication connection between the i-th module and the j-th module; This indicates that there is no direct communication connection between the two. Therefore, we obtain the neighbor set: Define the Laplace matrix within the cluster:
[0181] in, It is an adjacency matrix. It is a degree matrix.
[0182] Finally, based on the frequency band task response time in step S201... The time consumed for data acquisition, clustering, optimization, and planning in steps S10 to S30. Determine the predefined execution time of the underlying distributed controller:
[0183] in, To allow for the minimum physical adjustment time for low-level control, Maximum execution time.
[0184] For the i-th module, its cluster type is: The corresponding underlying predefined execution time is: .
[0185] Step S402: Construct distributed cooperative error and asymmetric performance constraints for the modules; Based on the final reference power of the i-th module in step S30 and the current module output power collected by the PCS. Define the module's local power tracking error: Furthermore, by combining the power tracking errors of neighboring nodes within the cluster, the distributed cooperative error of the i-th module is constructed:
[0186] in, Let be the pinning weight of the i-th module relative to the local reference power; This is the intra-cluster communication weight. This error includes both local power tracking and neighbor coordination terms, which can drive the power errors of intra-cluster modules to tend to be consistent and track the optimal reference power issued in step S30 under distributed conditions.
[0187] To achieve coordinated control of different safety boundaries under charging and discharging directions, positive and negative performance boundaries are constructed for the i-th module. Let the start time of the current control cycle be... Define the performance shrinkage function:
[0188] in: This represents the steady-state tolerance.
[0189] To ensure that the initial cooperative error is enveloped by the performance boundary, positive and negative initial envelope quantities are defined:
[0190]
[0191] in, , This represents positive and negative safety redundancy.
[0192] Constructing asymmetric performance boundaries:
[0193]
[0194] in, , These represent the positive and negative steady-state allowable errors, respectively.
[0195] Therefore, the module coordination error satisfies the following asymmetric performance constraint:
[0196] To transform constrained error into unconstrained error, a piecewise normalized error is constructed:
[0197] Further construct the asymmetric error transformation variables:
[0198] in, This represents the core error state of the distributed cooperative control in step S40. As can be seen from the above transformation, Bounded correspondence It always remains within the asymmetric performance boundary.
[0199] Step S403: Construct the module dynamic model after asymmetric error transformation; right By taking the derivative and substituting the converter's power regulation dynamics, PCS measurement dynamics, and neighbor error information, we obtain the local affine dynamic model after the asymmetric error transformation:
[0200] in, The equivalent control input for the module level to be designed is ultimately used to generate DC / DC duty cycle commands. The equivalent nonlinear drift term is formed by the error transformation, PCS output power dynamics, and neighbor interaction terms. To control the input gain term; It is a lumped disturbance consisting of DC / DC parameter perturbation, sampling error, communication disturbance, and unmodeled power dynamics.
[0201] The module-level equivalent control input is decomposed into a practical predefined time-base control term and an HJB optimal compensation term:
[0202] in, Used to achieve practical predefined time convergence for asymmetric error transformation variables. Used to compensate for unknown nonlinear residuals and optimize control costs.
[0203] Step S404: Design a practical predefined time-distributed reference control law; Design practical predefined time base control items:
[0204] in, for The generalized inverse, , , , ,and: Substitute Afterwards, the ideal closed-loop dominant term of the asymmetric error transformation variable satisfies:
[0205] The baseline control term provides a practical, predefined time-converged dominant structure, while the HJB optimal compensation term is used to further reduce input energy consumption and suppress residual disturbances.
[0206] Step S405: Construct the module-level HJB optimal compensation problem and ADP approximation network; Construct the HJB optimal compensation problem for the i-th module. Define the local performance function:
[0207] in, For error weights, The optimal compensation input weight matrix is... This is a module power boundary barrier penalty term, which is based on the module's executable power determined in step S204. Constructing obstacle penalty items :
[0208] in, The obstacle penalty coefficient, To prevent real numbers with a denominator of 0 from being used, it can be set to 0.01. , These are the maximum and minimum power of the battery module, respectively. It is the absolute value of the equivalent control input at the module level.
[0209] Then, an infinite time-domain performance index can be defined: The corresponding optimal value function is: Based on this, the HJB equation is constructed:
[0210] The optimal compensation control law is obtained from the extreme value conditions:
[0211] For online approximate solution Constructing the Critic network:
[0212] in, The Critic network basis function vectors, For Critic network weights, the superscript T indicates the transpose operation.
[0213] Constructing an Actor network:
[0214] in, Let Actor be the basis function vector. These are the weights of the Actor network.
[0215] The target compensation strategy induced by the Critic network is as follows:
[0216] Define Actor policy error Construct the integral Bellman residual:
[0217] Define the Bellman regression vector:
[0218] but:
[0219] Step S406: Design a practical, predefined, time-stable network weight update law; To ensure that the Critic network and Actor network converge in a practically predefined time context, a normalized Bellman regression vector is defined:
[0220] Define a practical predefined time nonlinearity correction operator: ,in, , and This is an element-wise exponentiation operation.
[0221] The Critic network weight update law is designed as follows:
[0222] in, For Critic learning rate, Predefined time adjustment parameters for Critic This is the coefficient of the Critic weight leakage term.
[0223] The Actor network weight update law is designed as follows:
[0224] in, For Actor learning rate, Set time adjustment parameters for the Actor. This represents the coefficient of the Actor weight leakage term.
[0225] To ensure that network weights are bounded, a projection operator is used:
[0226] in, To correspond to the weight update law for the unprojected network, This represents the allowed set of network weights.
[0227] Let the Lyapunov function corresponding to the closed-loop system in step S40 be:
[0228] in, , These represent the weight estimation errors for the Critic and Actor networks, respectively. , It is a positive definite matrix.
[0229] Under the conditions of bounded approximation error, bounded perturbation, and sufficient excitation of finite-time data, combining the practical predefined time base control law in step S404 and the network weight update law in step S406, we can obtain:
[0230] in, It is a bounded constant determined by the approximation error of the neural network, the upper bound of the perturbation, and the projection residual.
[0231] According to the practically predefined time stability criterion, the closed-loop system has an upper time bound. Convergence to the residual set Therefore, the distributed cooperative error of the module. within no more than Within a certain time, it enters the practical neighborhood corresponding to the asymmetric performance boundary, and due to the error transformation variable Bounded, the original cooperative error always satisfies .
[0232] like Figure 6As shown, this invention also provides a hierarchical response and module collaborative control system for cascaded battery clusters based on asymmetric performance constraints. This system is applied in the battery cabinet controller of a cascaded battery energy storage power station to execute the aforementioned hierarchical response and module collaborative control method for cascaded battery clusters based on asymmetric performance constraints. The system includes: a multi-source state acquisition module, a battery cluster partitioning module, a battery cluster game theory module, and a collaborative control module.
[0233] The multi-source status acquisition module is used to execute auxiliary service commands and acquire multi-source status data within the battery cabinet.
[0234] Specifically, the multi-source status acquisition module is communicatively connected to the Energy Management System (EMS), the Energy Storage Converter (PCS), and the Battery Management System (BMS) to acquire the current battery cabinet's cabinet-level ancillary service response tasks, the operating status of the DC / DC converter on the PCS side, and the multi-source operating status of the battery modules on the BMS side. Specifically, the multi-source status acquisition module acquires, through the EMS, the ancillary service response tasks assigned to the current battery cabinet based on the power plant-level multi-objective optimization results. These ancillary service response tasks include cabinet-level target power, target power change, power ramp-up requirements, response time requirements, allowable tracking error, and cabinet-level SOC target range, forming a cabinet-level ancillary service task vector.
[0235] The multi-source status acquisition module obtains the input voltage, input current, output voltage, output current, output power, duty cycle, upper duty cycle limit, lower duty cycle limit, output current limit, output power limit, converter temperature, and operating mode of the corresponding DC / DC converter of each battery module through the PCS, and constructs the executable state matrix of the PCS.
[0236] The multi-source state acquisition module obtains the SOC, SOH, SOE, available capacity, voltage, current, module output power, average temperature and operating mode of each battery module through the BMS, and further calculates the SOC margin, SOH margin and thermal safety margin, thereby forming the module battery state matrix.
[0237] The multi-source status acquisition module is also used to perform time synchronization, data validity marking, dimension normalization and status coding on data from EMS, PCS and BMS, forming a unified status dataset including cabinet-level auxiliary service task vectors, module battery status matrix and PCS executable status matrix, and sending the unified status dataset to the battery cluster partitioning module.
[0238] The battery cluster partitioning module is used to execute step S20: auxiliary service instruction parsing and dynamic clustering of heterogeneous battery modules.
[0239] Specifically, the battery cluster partitioning module receives a unified state dataset output by the multi-source state acquisition module. First, based on the cabinet-level target power, response time requirements, power ramp-up requirements, and allowable tracking error, it performs multi-scale power requirement analysis on the auxiliary service instructions, decomposing the cabinet-level target power into high-frequency fast response power components, mid-frequency regulation power components, and low-frequency energy support power components, and forming a set of frequency band task requirements.
[0240] Furthermore, the battery cluster partitioning module calculates the asymmetric response capability of each battery module for different frequency band tasks based on the module battery state matrix and the PCS executable state matrix, including energy response capability, power response capability and effective adaptability, and forms a module-task adaptability matrix.
[0241] The battery cluster partitioning module is also used to construct a unique ownership variable for the battery module, a clustering cost function, and constraints on power capability, energy capability, and ramp capability based on the frequency band task requirement set and task adaptation matrix, to solve the dynamic functional clustering problem of heterogeneous battery modules, and form a high-frequency fast response cluster, a mid-frequency regulation and tracking cluster, and a low-frequency energy support cluster.
[0242] In addition, the battery cluster partitioning module is used to determine the module power sharing weight within each battery cluster based on the dynamic clustering results, and generate the original initial power of the module. At the same time, based on the physical power boundary on the PCS side, the state margin continuous safety reduction function, and the charge and discharge safety availability coefficient, the executable power domain of each battery module is determined, and the original initial power of the module is projected onto the executable power domain to form the module initial power planning vector, the battery cluster initial power, and the input state required for step S30.
[0243] The battery cluster game module is used to execute step S30: solving the dynamic game optimization problem of battery clusters based on asymmetric performance constraints and formulating the output planning strategy.
[0244] Specifically, the battery cluster game module receives the module initial power planning vector, battery cluster affiliation matrix, battery cluster state vector and frequency band task requirement set output by the battery cluster partitioning module, and, based on the battery cluster partitioning result and module initial power planning value determined in step S20, constructs dynamic power response optimization models for the high-frequency fast response cluster, the mid-frequency regulation tracking cluster and the low-frequency energy support cluster respectively.
[0245] The battery cluster game module first calculates the average SOC and average SOE of the battery modules within each battery cluster, and uses the average SOC and average SOE as adjustment parameters for the battery cluster's charge and discharge capacity, asymmetric performance boundary, and aging cost. Then, it constructs the battery cluster power tracking error based on the initial power of the battery cluster and the corresponding frequency band power components, and further constructs the discharge boundary adjustment coefficient, charging boundary adjustment coefficient, positive asymmetric performance boundary, and negative asymmetric performance boundary to map the constrained power error into an unconstrained error state.
[0246] The battery cluster game module further constructs an anti-disturbance zero-sum differential game model with the battery cluster power tracking error as the core state, and constructs a performance function that includes a power error term, a power correction cost term, a disturbance suppression term, and an aging loss penalty term, and derives the HJI equation and its saddle point optimal strategy.
[0247] The battery cluster game module is also used to construct an HJI approximate solution network based on integral reinforcement learning, including a Critic network, a control Actor network, and a perturbation Actor network. A model-free online learning mechanism is constructed through integral Bellman residuals, and a fixed-time convergent network weight update law is adopted to make the network weights converge to the neighborhood of the optimal policy within the time upper bound independent of the initial conditions.
[0248] Finally, the battery cluster game module generates the optimal output correction amount, the optimal cluster-level reference power, and the final reference power of each battery cluster based on the output of the control Actor network, and sends the final reference power to the cooperative control module.
[0249] The collaborative control module is used to execute step S40: a distributed collaborative control strategy for battery modules within a cluster based on asymmetric performance constraints and optimal planning.
[0250] Specifically, the collaborative control module receives the optimal reference power of each battery cluster, the final reference power of the module, and the virtual state reference of the functional cluster from the battery cluster game module, and constructs a distributed communication topology for the battery modules and their DC / DC converters within the cluster.
[0251] The collaborative control module uses battery modules and their DC / DC converters belonging to the same battery cluster as distributed control nodes, constructs an intra-cluster communication graph, adjacency matrix, degree matrix and Laplace matrix, and determines the predefined execution time of the underlying distributed controller based on the task response time of the corresponding frequency band and the execution time of steps S10 to S30.
[0252] Furthermore, the cooperative control module constructs the module's local power tracking error based on the module's final reference power and the current module output power collected by the PCS, and constructs a distributed cooperative error by combining the power tracking errors of neighboring nodes; in response to the difference in adjustment capabilities in the charging and discharging directions, it constructs module-level positive performance boundaries and negative performance boundaries respectively, and maps the constrained cooperative error to an unconstrained error state through error transformation.
[0253] The collaborative control module further constructs the module dynamic model after asymmetric error transformation and decomposes the module-level equivalent control input into a practical predefined time base control term and an HJB optimal compensation term. The practical predefined time base control term ensures that the intra-cluster module collaborative error converges to the allowable neighborhood within a given time upper bound; the HJB optimal compensation term compensates for unknown nonlinear residuals, DC / DC parameter perturbations, sampling errors, communication disturbances, and unmodeled power dynamics.
[0254] The collaborative control module is also used to construct the module-level HJB optimal compensation problem and ADP approximation network, design a practical predefined time-stable network weight update law, and map the composite control law to the duty cycle command of the DC / DC converter corresponding to each battery module, ultimately realizing the distributed collaborative tracking of the optimal reference power by the battery modules within the cluster.
[0255] The beneficial effects of this invention are that, compared with the prior art, this invention constructs a hierarchical response optimization task that integrates asymmetric performance constraints at the battery cluster level, and combines state filtering, dynamic adaptive planning, collaborative output planning and distributed control at the module level, so that the battery cluster can meet the requirements of rapid response of auxiliary services, maintain the consistency of operation between modules and reduce the risk of secondary degradation of batteries used in tiered applications.
[0256] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.
[0257] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.
[0258] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.
[0259] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.
[0260] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.
Claims
1. A method for hierarchical response and module-coordinated control of cascaded battery clusters based on asymmetric performance constraints, characterized in that, include: Collect relevant data on the battery cabinets of the cascade battery energy storage power station, including auxiliary service instructions and multi-source status within the battery cabinets, and preprocess the relevant data of the battery cabinet controllers to obtain a unified status dataset. Multi-scale power demand analysis is performed on the unified state dataset to obtain the frequency band task demand set. Asymmetric response capability model for frequency band tasks is performed on the heterogeneous battery modules in the battery cabinet to obtain the task adaptation matrix. Based on the frequency band task demand set and the task adaptation matrix, the heterogeneous battery modules are dynamically divided into battery clusters, and an initial state vector is generated according to the battery cluster division result. A dynamic game model for battery clusters based on asymmetric performance constraints is constructed. The initial state vector is input into the dynamic game model for battery clusters, and the optimal power output of the battery cluster and the reference power of the battery module are output. A distributed collaborative control strategy for battery modules within a cluster is constructed based on asymmetric performance constraints and optimal planning. The optimal output of the battery cluster and the reference power of the battery modules are optimized based on the distributed collaborative control strategy, and the duty cycle command of the DC / DC converter corresponding to each battery module is generated.
2. The method for hierarchical response and module coordinated control of battery clusters based on asymmetric performance constraints according to claim 1, characterized in that, The data collected includes relevant data from the battery cabinets of the tiered battery energy storage power station, such as auxiliary service commands and multi-source status within the battery cabinets. The data from the battery cabinet controllers is preprocessed to obtain a unified status dataset, specifically including: Let the battery cabinet currently participating in the response of the cascaded battery energy storage power station be numbered as follows: The battery cabinet contains A heterogeneous, cascaded-use battery module; The auxiliary service instructions include the cabinet-level auxiliary service task vector of the q-th battery cabinet obtained by collecting auxiliary service instructions through the energy management system. The multi-source state within the battery cabinet includes the PCS state matrix of the q-th battery cabinet obtained by acquiring the operating state of the DC / DC converter through the energy storage converter. By collecting the multi-source operating status of the battery modules from the battery management system, the BMS state vector of the battery module in the q-th battery cabinet is obtained. ; The collected data undergoes preprocessing, including time synchronization, data validity marking, dimensional normalization, and state coding. This preprocessing transforms the collected data into a unified state vector corresponding to each battery cabinet, yielding the unified state vector of the q-th battery cabinet. : A unified state dataset is obtained based on the unified state vector of each battery cabinet in a tiered battery energy storage power station.
3. The method for hierarchical response and module collaborative control of battery clusters based on asymmetric performance constraints according to claim 2, characterized in that, The process of parsing auxiliary service instructions based on a unified state dataset and dividing heterogeneous battery module battery clusters, and generating an initial state vector based on the battery cluster division results, specifically includes: Cabinet-level auxiliary service task vector Multi-scale power demand analysis based on response time constraints yields a set of frequency band task requirements. ; Based on the PCS state matrix of the q-th battery cabinet Modeling the asymmetric response capability of heterogeneous battery modules for frequency band tasks yields the task adaptation matrix. ; Frequency band-based task requirements set Task Adaptability Matrix The nq heterogeneous cascaded battery modules inside the battery cabinet are dynamically divided into battery clusters, resulting in battery clusters including high-frequency fast response clusters, medium-frequency regulation and tracking clusters, and low-frequency energy support clusters. Based on the battery cluster partitioning results, the power components of each frequency band are allocated to the battery modules within the corresponding battery clusters, and the original initial power of the battery modules is projected into the power domain to generate an initial state vector.
4. The method for hierarchical response and module coordinated control of battery clusters based on asymmetric performance constraints according to claim 3, characterized in that, The PCS state matrix Modeling the asymmetric response capability of heterogeneous battery modules for frequency band tasks yields the task adaptation matrix. Specifically, it includes: Based on the BMS state vector of the i-th battery module in the q-th battery cabinet The first state margin factor is calculated using a saturation function, including the charge state factor, health state factor, energy state factor, capacity factor, and temperature margin factor. Based on the calculated factors, construct the energy response gating function for the i-th battery module of the q-th battery cabinet. and energy deficiency penalty item : in, This is the set of first-state margin factors; The gating slope coefficient corresponding to the first state margin factor of the nth class; Let be the normalized value of the nth type of first state margin factor for the i-th battery module in the q-th battery cabinet. For frequency band tasks The admission threshold for the z-th indicator, and satisfy: in, , , , Frequency band tasks The following are the SOC, SOH, SOE, capacity, and temperature margin admission thresholds; Based on the PCS state matrix of the q-th battery cabinet The second state margin factors, including current margin factor, power margin factor, duty cycle up-adjustment margin factor, and down-adjustment margin factor, are calculated using saturation functions. Power response gating functions are constructed based on the calculated factors. and power bottleneck penalty item : in, The set of second-state margin factors. The gating slope coefficient corresponding to the second state margin factor of the m-th class. It is the normalized value of the m-th type second state margin factor of the i-th battery module in the q-th battery cabinet; Combined with energy response gating function and energy deficiency penalty item , thus obtaining the energy response capability of the i-th heterogeneous battery module for frequency band task r; Combined with power response gating function and power bottleneck penalty item We obtain the power response capability of the i-th heterogeneous battery module for frequency band task r; Based on the energy response capability and power response capability of the battery module for frequency band task r, the adaptability of the first battery module to frequency band task r is calculated. The task adaptation matrix is obtained by combining the adaptability of tasks in each frequency band. .
5. The method for hierarchical response and module coordinated control of battery clusters based on asymmetric performance constraints according to claim 3, characterized in that, The frequency band-based task requirement set Task Adaptability Matrix The nq heterogeneous battery modules inside the battery cabinet are dynamically divided into battery clusters, specifically including: Construct the attribution variables for the battery modules, with each battery module undertaking only one frequency band task in each control cycle; Based on the task fit matrix Construct a battery module clustering cost function, and use the battery module clustering cost function to calculate the allocation of the i-th battery module to the battery cluster. The matching cost; Construct battery cluster task capability constraints based on the available power, available energy, and available ramping capability of the battery module; Based on the attribution variables, matching costs, and task capability constraints, construct a dynamic clustering optimization problem: in, The cluster affiliation matrix, Let q be the number of battery modules for secondary use in the q-th battery cabinet. For frequency band task set, , These are the attribution variables of the i-th battery module in the q-th battery cabinet to the frequency band task r and the allocation of the i-th battery module to the battery cluster, respectively. The matching cost; Let be the available power of the i-th battery module in the q-th battery cabinet under frequency band task r. The power component of frequency band task r; , These are the high-frequency and low-frequency clustering decision variables for the i-th battery module in the q-th battery cabinet, respectively. Let i be the available energy of the i-th battery module under low-frequency energy support tasks. The energy demand corresponding to the low-frequency power component; Let i be the available ramp-up capability of the i-th battery module under high-frequency tasks. For the ramp-up requirements corresponding to high-frequency power components; Solve the dynamic clustering optimization problem to obtain the cluster affiliation matrix, and generate a set of battery clusters based on the cluster affiliation matrix.
6. The method for hierarchical response and module coordinated control of battery clusters based on asymmetric performance constraints according to claim 5, characterized in that, Based on the battery cluster partitioning results, the power components of each frequency band are allocated to the battery modules within the corresponding battery clusters, and the original initial power of the battery modules is projected into the power domain to generate an initial state vector, as detailed below: Calculate the power sharing coefficient based on the available power of each battery module under the frequency band task it undertakes in the current control cycle; The original initial power of each battery module is calculated based on the cluster affiliation matrix and power sharing coefficient. The PCS-side physical power boundary of the i-th battery module is constructed based on the PCS state matrix, and a continuous safety reduction function for state margin is constructed to continuously correct the PCS-side physical power boundary. Calculate the charge and discharge safety availability factor of the battery module, and determine the executable power domain of the battery module based on the corrected physical power boundary of the PCS side and the charge and discharge safety availability factor. The original initial power of the module is projected onto the executable power domain to obtain the projected initial power planning vector of the module. The initial power of each battery cluster is calculated based on the initial power planning vector of the module and the cluster affiliation matrix. The battery cluster state vector is obtained by aggregating the battery module state based on the cluster affiliation matrix. The initial power planning vector of the module, the cluster affiliation matrix, the battery cluster state vector and the frequency band task requirement set are combined to generate the initial state vector.
7. The method for hierarchical response and module coordinated control of battery clusters based on asymmetric performance constraints according to claim 6, characterized in that, The process of projecting the original initial power of the module onto the executable power domain to obtain the projected initial state vector of the module is as follows: in, This is the initial power vector of the projected module. , Let be the original initial power vector and the original initial power of the i-th battery module projected from the q-th battery cabinet, respectively. Width of the executable power range; The power change penalty coefficient; t is the current time. The sampling period; Let q be the target power of the battery cabinet; , These are the lower limit and upper limit of the discharge power of the i-th battery module, respectively; This represents the available ramp capability of the i-th battery module under task r in its respective frequency band. It is a constant.
8. The method for hierarchical response and module coordinated control of battery clusters based on asymmetric performance constraints according to claim 1, characterized in that, The construction of a dynamic game model for battery clusters based on asymmetric performance constraints involves inputting the initial state vector into the model and outputting the optimal power output of the battery cluster and the reference power of the battery module. Specifically, this includes: The asymmetric power response state of the battery cluster is constructed based on the initial state vector and used as the state variable of the dynamic game model of the battery cluster. Based on the aforementioned state variables, cluster-level power correction, and lumped disturbance term, a dynamic game model of battery clusters is constructed, and the optimal strategy at the HJI saddle point is analyzed. Construct an Actor network, and based on the state variables and the saddle point optimal policy, construct an HJI approximate solution network in the form of integral reinforcement learning; A fixed-time convergent network weight update law is set to update the HJI approximation solution network and the Actor network online. The optimal output of the battery cluster is output based on the updated HJI approximation solution network and Actor network, and the reference power of the battery module is obtained by further allocating the optimal output of the battery cluster.
9. The method for hierarchical response and module coordinated control of cascaded utilization battery clusters based on asymmetric performance constraints according to claim 8, characterized in that, The constructed asymmetric power response state of the battery cluster specifically includes: Calculate the battery cluster based on the attribution variables of the i-th battery module to the frequency band task r. Average SOC and average SOE of the internal battery module; Calculate the battery cluster power tracking error based on the initial power of the battery cluster and the frequency band power components; According to battery clusters Calculation of discharge boundary adjustment coefficient based on average SOC and average SOE of internal battery module and charging boundary adjustment coefficient ; Combined with discharge boundary adjustment coefficient and charging boundary adjustment coefficient Construct battery clusters separately for power tracking errors of battery clusters Positive asymmetric performance boundary and negative asymmetric performance boundary : Where t is the current time, This is the start time of the current control cycle; , Battery clusters The positive and negative initial performance boundaries; , Battery clusters The positive and negative steady-state performance boundaries; , Battery clusters The positive and negative boundary contraction rates.
10. The method for hierarchical response and module coordinated control of cascaded utilization battery clusters based on asymmetric performance constraints according to claim 8, characterized in that, The construction of the dynamic game model for battery clusters and the analysis of HJI saddle points specifically include: With battery clusters Taking the asymmetric performance boundary as the object, and using the asymmetric performance boundary as the asymmetric performance constraint, a cluster-level performance function is constructed after transforming the power tracking error constraint of the battery cluster. Constructing Infinite-Time-Domain Performance Indices Based on Cluster-Level Performance Functions The cluster-level dynamic game optimization problem can be expressed as: in, It is the optimal value function; This is an asymmetric error transformation state; For cluster-level power correction control input Perform minimization optimization; To address lumped disturbances Perform maximization modeling; Construct a Hamiltonian function based on the cluster-level performance function, and then apply these functions to the Hamiltonian function with respect to... and Find the extreme value to obtain the optimal strategy at the saddle point.
11. The method for hierarchical response and module coordinated control of cascaded utilization battery clusters based on asymmetric performance constraints according to claim 8, characterized in that, The setting of a fixed-time convergent network weight update law, which enables the network weights to converge to the neighborhood of the optimal policy within a time upper bound independent of the initial conditions, specifically includes: The policy errors of the control Actor and the policy errors of the perturbation Actor are constructed as follows: in, For Actor policy error; , These are the target cluster-level power corrections output by the Critic network and the actual cluster-level power corrections output by the Actor network, respectively. The weight error of the Actor network; , These are the estimated weights for the Critic and Actor networks, respectively. Set up a fixed-time convergent network weight update law, including a normalized Bellman regression vector and a fixed-time nonlinear correction operator; This allows the Bellman residual, the policy error of the control Actor, and the policy error of the perturbation Actor to converge to the zero neighborhood within a time bound independent of the initial weights.
12. The method for hierarchical response and module collaborative control of battery clusters based on asymmetric performance constraints according to claim 1, characterized in that, The constructed cluster-based distributed collaborative control strategy for battery modules, based on asymmetric performance constraints and optimal planning, processes the optimal output of the battery cluster and the reference power of the battery modules to generate duty cycle control commands for the DC / DC converters corresponding to each battery module. Specifically, this includes: Construct an intra-cluster distributed communication topology based on the optimal output of the battery cluster and the reference power of the battery module, and determine the predefined execution time; Based on the module reference power and actual output power, a distributed cooperative error and its asymmetric performance constraints are constructed; the distributed cooperative error is transformed into an unconstrained error state, and the corresponding module dynamic model is constructed. Based on the module dynamic model, a predefined time-distributed reference control law is set, and a module-level HJB optimal compensation problem and an ADP approximation network are constructed. The ADP approximation network is updated online using a predefined time-stable network weight update law to generate duty cycle control commands for the DC / DC converters corresponding to each battery module.
13. The method for hierarchical response and module collaborative control of battery clusters based on asymmetric performance constraints according to claim 12, characterized in that, The distributed cooperative error and asymmetric performance constraints of the constructed module specifically include: The module local power tracking error is defined based on the module's final reference power and the current module output power, and a distributed cooperative error is constructed by combining the power tracking errors of neighboring nodes within the cluster. To address the differences in charging and discharging directions, module-level positive and negative performance boundaries are constructed. Define the performance shrinkage function, the positive initial envelope, and the negative initial envelope; Construct asymmetric performance boundaries, including positive steady-state error and negative steady-state error; Based on the asymmetric performance boundary, the module cooperative error is made to satisfy the asymmetric performance constraint. A piecewise normalized error is constructed based on the asymmetric performance boundary. as follows: in, For distributed cooperative error , These are the positive steady-state error and the negative steady-state error, respectively. Construct asymmetric error transformation variables based on piecewise normalized errors.
14. A hierarchical response and module collaborative control system for cascaded utilization battery clusters based on asymmetric performance constraints, used to implement the hierarchical response and module collaborative control method for cascaded utilization battery clusters based on asymmetric performance constraints as described in any one of claims 1-13, characterized in that, include: The multi-source status acquisition module is used to collect relevant data of the battery cabinet in the cascade utilization battery energy storage power station, including auxiliary service instructions and multi-source status in the battery cabinet, and to preprocess the relevant data of the battery cabinet controller to obtain a unified status dataset. The battery cluster partitioning module is used to perform multi-scale power demand analysis on the unified state dataset to obtain the frequency band task demand set, to model the asymmetric response capability of heterogeneous battery modules in the battery cabinet for frequency band tasks to obtain the task adaptation matrix, to perform dynamic battery cluster partitioning on the heterogeneous battery modules based on the frequency band task demand set and the task adaptation matrix, and to generate an initial state vector based on the battery cluster partitioning result. The battery cluster game module is used to construct a dynamic game model of battery clusters based on asymmetric performance constraints. The initial state vector is input into the dynamic game model of battery clusters, and the optimal output of battery clusters and reference power of battery modules are output. The collaborative control module is used to construct a distributed collaborative control strategy for battery modules within a cluster based on asymmetric performance constraints and optimal planning. Based on the distributed collaborative control strategy, it optimizes the optimal output of the battery cluster and the reference power of the battery modules, and generates duty cycle commands for the DC / DC converters corresponding to each battery module.
15. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-13.
16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-13.
