A multi-objective coordinated control method and system for an energy storage system based on a hierarchical architecture

Abstract

The application provides a multi-target coordinated control method and system of an energy storage system based on a hierarchical architecture, comprising: establishing a cross-level heterogeneous data space-time mapping mechanism, aligning and fusing a top discrete scheduling sequence and a bottom continuous state sequence to generate a power state response baseline; introducing a rhythm invasion degree index and a residual error divergence feature to identify a multi-target conflict surge critical singular point and form a continuous time domain surge zone; actively injecting a controlled micro offset into the top scheduling sequence within the surge zone for time sequence decoupling and reconstructing a decoupled response baseline; and for a transient power step, performing dynamic variable slope limiting in the time dimension, combining cluster level health differences in the space dimension for power oriented distribution, and generating an updated response baseline. The application realizes pre-diagnosis and active decoupling of multi-target conflicts, fundamentally cuts off the positive feedback coupling channel, space-time dual dimensionally relieves the impact current, effectively suppresses the rate overrun and local thermal runaway, significantly prolongs the battery life and guarantees the smooth operation of the system.

Description

Technical Field

[0001] This invention relates to the field of energy storage system control technology, specifically to a multi-objective coordinated control method and system for energy storage systems based on a hierarchical architecture. Background Technology

[0002] With the large-scale grid connection of new energy sources and the advancement of new power system construction, electrochemical energy storage power stations have become core equipment for undertaking multiple tasks such as grid peak shaving, frequency regulation, and new energy source fluctuation mitigation. To ensure the safe and efficient operation of energy storage systems, existing energy storage control architectures typically adopt a layered distributed structure: the top-level energy management system (EMS) is responsible for long-scale economic dispatch planning, the intermediate coordination control layer is responsible for medium-scale power allocation and inter-cluster equalization, and the bottom-level local execution layer (PCS / BMS) is responsible for short-scale power point tracking and physical protection.

[0003] However, under complex real-world operating conditions, this layered architecture faces severe multi-objective coupling conflicts, and existing technologies have the following drawbacks:

[0004] The inherent decoupling across different spatiotemporal scales leads to a lag in conflict perception: top-level scheduling commands are typically discrete events generated based on minute-level macroeconomic factors, while bottom-level battery states (SOC, SOH, temperature rise, etc.) are analog quantities that evolve continuously on the order of seconds or even milliseconds. Traditional control methods lack a spatiotemporal fusion mechanism for heterogeneous data across different levels, resulting in a "blind alignment" between top-level commands and bottom-level states on the time axis. When high-frequency disturbances in the power grid (such as AGC frequency regulation commands) forcibly intervene and disrupt the slow rhythm of the top-level system, the system cannot detect the brewing of multi-objective conflicts in advance, often only responding passively when deviations surge or equipment exceeds limits.

[0005] Traditional power limiting is a passive amplitude truncation that fails to sever the root of coupling: when faced with multiple conflicting objectives (such as needing to respond quickly to high-frequency power commands while simultaneously balancing inter-cluster SOC equalization and rate limiting), existing methods often employ rigid power amplitude limiting or fixed-slope limiting. This approach not only sacrifices grid response performance but also fails to sever the positive feedback coupling channels between different control objectives from a timing logic perspective, easily leading the system into a vicious cycle of "command oscillation - equalization deterioration - protection tripping".

[0006] The underlying converter lacks a spatiotemporally coordinated adaptive guidance mechanism: Under transient power step impact, existing underlying control often adopts a single time-dimensional limiting or spatial-dimensional average distribution strategy. On the one hand, fixed slope limiting cannot adapt to the dynamic tolerance of batteries in different SOC ranges; on the other hand, ignoring the differences in cluster-level health (SOH) and averaging distribution can easily trigger the "weakest link effect," leading to rate exceeding limits and local thermal runaway of aging battery clusters, severely reducing the lifespan of the energy storage system.

[0007] The lack of a long-term rhythmic phase maintenance mechanism: Even if temporary peak-shifting decoupling is achieved through certain means, under continuous strong disturbances in the power grid, the timing phase of the top-level dispatch and the bottom-level response is very likely to drift again and recouple. Existing technologies lack dynamic tracking and feedforward prediction mechanisms similar to phase-locked loops, and cannot maintain the stable decoupled state of the system during long-term operation.

[0008] Therefore, there is an urgent need for a coordinated control method for energy storage systems that can perceive spatiotemporal coupling characteristics across levels, actively decouple multi-objective conflicts from the temporal source, and realize spatiotemporal dual-dimensional energy diversion at the bottom layer. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a multi-objective coordinated control method and system for energy storage systems based on a hierarchical architecture, thereby solving the problems mentioned in the background section.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a multi-objective coordinated control method for an energy storage system based on a hierarchical architecture, comprising the following steps:

[0011] S1: Establish a cross-level heterogeneous data spatiotemporal mapping mechanism to perform event-level alignment and feature fusion of the top-level discrete baseline scheduling sequence and the bottom-level continuous battery cluster state sequence on the same time axis to generate a power state response baseline.

[0012] S2: Based on the response baseline, the evolution segment where the top-level scheduling cycle is forced to shorten due to high-frequency disturbances in the power grid is identified by introducing a rhythmic intrusion index to determine that the system has entered a multi-objective conflict incubation period. Combined with the residual divergence characteristics of the middle-level power allocation, the critical singular point of the deviation surge is captured. The singular point is connected in series in the time domain and extended in both directions to form a continuous time domain surge band.

[0013] S3: Within the time window of the surge zone, a controlled micro-offset is actively injected into the top-level baseline scheduling sequence to break the time synchronization between the top-level extreme value and the sensitive interval of the middle-level power response, and to reconstruct the decoupled response baseline in order to cut off the multi-objective positive feedback coupling channel.

[0014] S4: Based on the decoupled response baseline, for transient power step, the energy release cycle is extended in the time dimension by dynamically varying the slope, and power-guided allocation is performed in the spatial dimension by combining the differences in cluster-level health status to generate an updated response baseline.

[0015] S5: Based on the updated response baseline, the phase difference between the top-level scheduling and the bottom-level response is finely adjusted using a phase-locked loop-like approach. Combined with feedforward prediction and feedback correction, the system locks into a peak-shifting and decoupled state.

[0016] As a preferred embodiment, the specific process for generating the power state response baseline includes:

[0017] A multi-dimensional high-frequency synchronous sensing array is deployed on the cross-layer data interaction link. The top-level communication bus bypass access message parsing probe captures and schedules the message distribution. Sensors are deployed on the bottom-level DC bus section to obtain cluster-level status data, and the timestamp is aligned using the network time protocol or IEEE1588 protocol.

[0018] By introducing a dynamic time warping algorithm combined with sliding window cross-correlation analysis to perform time series elastic alignment, the step front of the top-level scheduling instruction is used as the driving anchor point, and the corresponding response inflection point in the bottom-level state sequence is found as the response anchor point. The elastic delay compensation factor is calculated for pre- or post-compensation, and a continuous spatiotemporal time chain is constructed.

[0019] In the continuous spatiotemporal time chain, the extreme points, step change points and rate of change of the top-level reference power curve are extracted as the driving event set, and the extreme points of the first derivative of the bottom-level battery cluster SOC curve and the extreme points of inter-cluster temperature difference are identified as the response event set. The causal correlation determination algorithm is used to pair the driving events with the next nearest and most correlated response events, and connect them sequentially to form the power state response baseline.

[0020] As a preferred embodiment, the specific process for determining whether the system has entered the multi-target conflict incubation period in step S2 includes:

[0021] The evolution of the top-level rhythm under the intrusion of high-frequency grid rhythms is scanned along the response baseline, and the actual time interval sequence of adjacent top-level scheduling extrema is calculated. Fluctuation cycle sequence of high-frequency commands from the power grid Define rhythmic invasiveness index ;

[0022] when Shrinking to the second level and When the value continues to decrease and converges to within a preset threshold, the system is determined to have entered the multi-objective conflict incubation period.

[0023] As a preferred embodiment, the specific process for forming the continuous temporal surge band in step S2 includes:

[0024] The variance coefficient of variation method is used to calculate the instantaneous residual sequence of the actual power distribution in the middle layer and the ideal smooth trajectory along the time axis, and the local variance of the residual sequence is calculated.

[0025] When the local variance changes from steady-state small-amplitude fluctuations to exponential divergence over multiple consecutive control periods, and the sign of the variance derivative changes from positive to negative, the first mutation moment is extracted as the starting node of the conflict surge.

[0026] The starting nodes are sequentially connected along the timeline to form the main trunk, and the rhythmic invasiveness is extracted forward. The segment within the second preset threshold range is designated as the warning zone, and the segment whose deviation variance converges to within twice the steady state is designated as the decay zone, forming a continuous time-domain surge band with an energy reservoir effect.

[0027] As a preferred embodiment, the specific process for reconstructing the decoupled response baseline includes:

[0028] Within the surge time window, compare the timing of the top-level benchmark extreme values. When the mid-level deviation surge occurs If the absolute time difference between the two If the duration is less than one-third of a single scheduling cycle, it is determined to be a synchronous coupling node;

[0029] Extract the current surge energy storage tank area S and the current SOC range ΔSOC of the underlying battery cluster, and substitute them into the adaptive elastic micro-time difference calculation model to calculate the time series offset. ,in and The coefficient is set according to the system inertia;

[0030] For the aforementioned synchronous coupling node, if the top-level extreme value is earlier than the surge node and the time difference is too small, then an application is made to the instruction issuance time corresponding to that extreme value. The delay buffer; if the top-level extreme value lags behind the surge node, its distribution time is slightly advanced. In subsequent non-surge periods, a power compensation wavelet with halved amplitude and doubled time is introduced to ensure absolute conservation of energy integral before and after translation, generating a decoupled response baseline.

[0031] As a preferred embodiment, the specific process for generating the updated response baseline includes:

[0032] The differential value of the DC-side power of the underlying PCS is calculated in real time along the decoupling response baseline. When the interval between adjacent step jumps is much lower than the system's average electromechanical time constant, and the local temperature rise rate of the battery cluster in that section is... When the thermal runaway warning threshold is exceeded, it is determined to be in the power concentration step impact zone;

[0033] Perform dynamic slope shaping over time within the impact zone;

[0034] In the spatial dimension, energy flooding allocation is performed based on cluster-level health. Differentiated guidance is given according to the SOH and real-time internal resistance of each battery cluster. A smaller power share is allocated to aged battery clusters with low SOH and high internal resistance to reduce their C-rate, while the impact power share borne by healthy battery clusters is increased, generating an updated response baseline after spatiotemporal two-dimensional shaping.

[0035] As a preferred embodiment, the specific process of performing dynamic slope shaping in the time dimension within the impact zone includes:

[0036] Dynamic variable slope shaping is performed using a nonlinear slope limiter based on SOC feedback within the power-concentrated step impact region.

[0037] When the SOC is in the preset two-end range, the slope limit is automatically tightened; when the SOC is in the preset middle range, the slope limit is increased.

[0038] During the power rise period, the original instantaneous large step command is broken down into multiple S-shaped ramp commands with smoothing factors to reduce the current peak.

[0039] During the power reduction period, control the unloading rate of the converter and start the liquid cooling system in advance to avoid adiabatic temperature rise.

[0040] As a preferred embodiment, the specific process of locking the peak-shifting decoupling state includes:

[0041] The time difference sequence between the top-level extreme value and the bottom-level response extreme value is extracted in real time. When the absolute value of the time difference sequence decreases to a preset first time difference threshold, a negative adjustment signal is output to apply delay compensation to the top-level instruction of the next cycle.

[0042] When the absolute value of the time difference sequence increases to a preset second time difference threshold, a positive adjustment signal is output to apply advance to the top-level command, forcing the time difference back to the stable decoupling zone;

[0043] By using long short-term memory networks or autoregressive models, based on the grid load fluctuation patterns and electricity price change trends of the previous N periods, the timing and duration of the next top-level dispatch extreme value can be predicted.

[0044] If the predicted extreme value of the next cycle falls into the high-frequency instruction dense area, a reverse energy pulse constrained by the safety threshold of the battery management system is injected in advance at the end of the current cycle to change the initial SOC state of the underlying battery, so as to induce the top-level scheduling logic to actively adjust the delivery time in the next cycle.

[0045] As a preferred embodiment, a multi-objective coordinated control system for an energy storage system based on a hierarchical architecture, used to implement the above-mentioned multi-objective coordinated control method for an energy storage system based on a hierarchical architecture, includes:

[0046] The baseline generation module is used to establish a cross-level heterogeneous data spatiotemporal mapping mechanism. It performs event-level alignment and feature fusion on the same time axis between the top-level discrete baseline scheduling sequence and the bottom-level continuous battery cluster state sequence to generate a power state response baseline.

[0047] The surge band definition module is used to identify the evolution segment where the top-level scheduling cycle is forced to shorten due to high-frequency disturbances in the power grid, based on the response baseline, by introducing a rhythmic intrusion index. It also captures the critical singular point of the deviation surge by combining the residual divergence characteristics of the middle-level power allocation, and connects the singular point in the time domain and extends it bidirectionally to form a continuous time-domain surge band.

[0048] The timing decoupling module is used to actively inject controlled micro-offsets into the top-level reference scheduling sequence within the time window of the surge zone, break the timing synchronization between the top-level extreme value and the middle-level power response sensitive interval, reconstruct the decoupling response baseline, and cut off the multi-objective positive feedback coupling channel.

[0049] The shaping and flood discharge control module is used to extend the energy release cycle in the time dimension by dynamically varying the slope for transient power steps, and to perform power-guided allocation in the spatial dimension by combining the differences in cluster-level health status, based on the decoupled response baseline, to generate an updated response baseline.

[0050] The phase locking and maintenance module is used to fine-tune the phase difference between the top-level scheduling and the bottom-level response based on the updated response baseline, using a phase-locked loop-like concept, and combining feedforward prediction and feedback correction to lock the system in a peak-shifting decoupling state.

[0051] This invention provides a multi-objective coordinated control method and system for energy storage systems based on a hierarchical architecture, which has the following beneficial effects:

[0052] 1. This invention deploys a multi-dimensional high-frequency synchronous sensing array and introduces a dynamic time warping algorithm to align the top-level discrete scheduling instructions with the bottom-level continuous battery state at the event level, generating a power state response baseline that integrates timing phase difference and thermodynamic characteristics. By combining the rhythmic intrusion index and residual divergence characteristics, discrete conflict singularities are modeled as continuous time-domain surge bands with an "energy reservoir" effect, providing an accurate spatiotemporal multi-dimensional reference system for the early perception and quantitative assessment of multi-target conflicts, completely changing the limitations of traditional post-event passive protection.

[0053] 2. This invention actively injects controlled micro-offsets into the top-level baseline scheduling sequence within the surge time window, breaking the time synchronization between the top-level extreme value and the sensitive interval of the middle-level power response. At the same time, by introducing power compensation wavelets with halved amplitude and doubled time during non-surge periods, the multi-objective positive feedback coupling channel is fundamentally cut off while ensuring the absolute conservation of energy integral before and after the shift. This allows the system to regain dynamic stability and adjustment margin, taking into account both grid response performance and intraday power dispatching plan.

[0054] 3. In the time dimension, this invention employs a nonlinear dynamic slope limiting based on SOC feedback to transform sharp step waves into gentle trapezoidal waves; in the spatial dimension, it combines cluster-level SOH and internal resistance differences to conduct "energy flooding" guided distribution. This mechanism transforms concentrated current surges into long-period, multi-branch dispersed releases, completely eliminating the local barrel effect and suppressing lithium plating or local overheating caused by lithium-ion concentration polarization from the electrochemical mechanism level, thus greatly delaying the cycle aging process of the energy storage battery.

[0055] 4. This invention introduces a phase tracker similar to a phase-locked loop at the system level, combined with feedforward prediction compensation based on LSTM or autoregressive models, to dynamically fine-tune the phase difference between the top-level scheduling and the bottom-level response. By injecting reverse energy pulses in advance at the physical level to induce scheduling logic adjustments, the system locks into a stable peak-shaving and decoupling state during long-term operation. Even under strong grid disturbances, it can ensure stable and continuous energy throughput, significantly reducing operational risks and providing a highly forward-looking underlying control guarantee for energy storage power stations to participate in frequency regulation and peak shaving ancillary services and energy arbitrage for a long time. Attached Figure Description

[0056] Figure 1 This is a flowchart of the multi-objective coordinated control method for energy storage systems based on a hierarchical architecture, as described in this invention.

[0057] Figure 2 This is a schematic diagram of the multi-objective coordinated control method for energy storage systems based on a hierarchical architecture, as described in this invention.

[0058] Figure 3 This is a block diagram of the multi-objective coordinated control system for the energy storage system based on a hierarchical architecture, as presented in this invention. Detailed Implementation

[0059] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0060] like Figure 1 and 2 As shown, this embodiment of the invention provides a multi-objective coordinated control method for an energy storage system based on a hierarchical architecture, comprising the following steps:

[0061] Step 1: Construct a cross-level power-state spatiotemporal coupled response baseline: Establish a cross-level heterogeneous data spatiotemporal mapping mechanism, and perform event-level alignment and feature fusion on the same time axis to generate a power state response baseline;

[0062] Specifically, taking advantage of the inherent decoupling of the top-level energy management layer, the intermediate coordination and control layer, and the bottom-level local execution layer on the time scale, a cross-level heterogeneous data spatiotemporal mapping mechanism is established. The discretely distributed baseline scheduling sequence from the top level and the continuously evolving battery cluster state sequence from the bottom level are aligned at the event level and fused with features on the same time axis to generate a power state response baseline for characterizing the dynamic traction relationship between the "source and load" across scales, providing a high-fidelity spatiotemporal reference system for early detection of multi-objective conflicts.

[0063] In this embodiment, a cross-level power-state spatiotemporal coupling response baseline is constructed, and the specific steps are as follows:

[0064] A multi-dimensional high-frequency synchronous sensing array is deployed on the cross-layer data interaction link. A message parsing probe is bypassed between the core communication bus and the scheduling logic core of the top-level EMS (Energy Management System) to capture the active / reactive planned values ​​and execution time in the scheduling messages with microsecond-level timestamps. At the bottom local execution layer, high-precision Hall sensors and a distributed temperature acquisition matrix are deployed on the DC bus section between the battery cluster combiner cabinet and the DC / DC converter. This not only collects DC voltage and current but also simultaneously acquires the cluster-level inlet and outlet temperature difference and the extreme voltage of individual cells to directly reflect the microscopic evolution of the battery's true state of charge (SOC) and state of health (SOH). The two sensing nodes are hard-aligned using a unified Network Time Protocol (NTP) or the IEEE 1588 precision clock synchronization protocol to eliminate timing distortions caused by communication network jitter.

[0065] After completing multidimensional data acquisition, time series elastic alignment of heterogeneous signals is performed. Since top-level instructions are discrete events while bottom-level states are continuous analog quantities, this step introduces a Dynamic Time Warping (DTW) algorithm combined with sliding window cross-correlation analysis. Using the initial sampling time as the time origin, the step leading edge of the top-level scheduling instruction sequence is used as the driving anchor point, and the corresponding response anchor point is found in the bottom-level state sequence. To address the inherent random delay of the communication link, an elastic delay compensation factor is calculated between the two, and the driving anchor point and response anchor point are pre- or post-compensated on the time axis, thereby constructing a seamless, continuous spatiotemporal time chain.

[0066] It should be noted that the alignment process is executed in the mid-level coordinating controller using a low-frequency parallel computing architecture, which does not occupy the core computing resources of the underlying converter, thus ensuring real-time performance and practicality in industrial settings.

[0067] Furthermore, based on this, deep extraction of rhythmic coupling features and baseline generation are performed. In the spatiotemporal time chain, not only are the extreme points and step abrupt change points of the top-level baseline power curve extracted, but their rate of change is also extracted as a set of driving events. Simultaneously, the extreme points of the first derivative of the bottom-level battery cluster SOC curve and the extreme points of inter-cluster temperature difference are identified as a set of response events. Following chronological order, a causal correlation determination algorithm is used to pair each driving event with the next nearest response event with the highest correlation coefficient, constructing multi-dimensional feature-related node pairs. Connecting the node pairs sequentially across the entire time domain forms the power state response baseline. This baseline not only includes traditional power-state two-dimensional information but also integrates temporal phase difference and thermodynamic features. Its vertical gradient characterizes the pulling strength of scheduling commands on the evolution of the bottom-level state, and its horizontal span characterizes the hysteresis characteristics of the system response. As the system runs, this baseline is updated with the time window, providing a dynamic spatiotemporal multi-dimensional reference system for subsequent conflict identification.

[0068] Step 2: Define the time-domain surge zone of multi-objective control conflict and the boundary of the energy reservoir: Based on the response baseline, the evolution segment where the top-level scheduling cycle is forced to shorten due to high-frequency disturbances in the power grid is identified by introducing a rhythmic intrusion index to determine that the system has entered the multi-objective conflict incubation period. Combined with the residual divergence characteristics of the mid-level power allocation, the critical singular point of the deviation surge is captured. The singular point is connected in series in the time domain and extended bidirectionally to form a continuous time-domain surge zone.

[0069] Specifically, based on the aforementioned multidimensional response baseline, by introducing a metric for rhythmic intrusion, the evolutionary segments where the top-level scheduling cycle is forced to shorten due to high-frequency grid disturbances (such as AGC frequency regulation commands and new energy slope control) are accurately identified. Combining the residual divergence characteristics of the power allocation objective function within the intermediate coordination control layer (such as fluctuation mitigation, inter-cluster equilibrium, and extreme value limitation), the critical singular points of deviation surges are captured. These discrete singular points are then connected in series in the time domain and extended bidirectionally to form a continuous time-domain surge band with an energy reservoir effect.

[0070] In this embodiment, the temporal surge zone of multi-objective control conflict and the boundary of the energy storage pool are defined. The specific process includes:

[0071] The evolution of the top-level rhythm under the intrusion of high-frequency grid rhythms is scanned along the response baseline, and the actual time interval sequence of adjacent top-level scheduling extrema is calculated in real time. Simultaneously, the fluctuation cycle sequence of high-frequency commands such as power grid AGC is extracted. Define rhythmic invasiveness index .when When the frequency rapidly shrinks from the minute level (such as the planning and tracking stage) to the second level, and the intrusion degree η continues to decrease and converges to within a preset threshold (such as 0.2), it is determined that the top-level slow rhythm has been deeply intruded by the high-frequency rhythm, and the system enters the multi-target conflict incubation period.

[0072] For example, taking a 10MW / 20MWh energy storage power station as an example, the top-level planned extreme value interval is about 300 seconds in the first 100 seconds (pure economic dispatch); affected by grid fluctuations, the interval between 200 and 300 seconds is shortened to 120 seconds (electricity price follows); at 400 seconds, due to the forced intervention of AGC commands, it drops sharply to 15 seconds. During the same period, the grid AGC response cycle fluctuates around 12 seconds. When the top-level interval approaches 15 seconds, the intrusion degree ηη is less than 0.25. At this time, the originally decoupled "power smoothing" and "inter-cluster SOC balance" objectives within the middle-level coordination and control layer become severely coupled. The peak-valley deviation of the power command increases dramatically from 200kW to 800kW, and the inter-cluster voltage difference expands from 20mV to 150mV, marking the beginning of the multi-objective conflict.

[0073] Subsequently, the coefficient of variation method was used to capture the critical starting point of the deviation surge. The instantaneous residual sequence of the actual power distribution in the middle layer and the ideal smooth trajectory was calculated along the time axis, and the local variance of the residual sequence was calculated. When the local variance suddenly changes from steady-state small-amplitude fluctuations to exponential divergence over multiple consecutive control cycles, and the sign of the variance derivative changes from positive to negative (i.e., reaching the divergence extreme point and beginning to erode the system's stability margin), the first abrupt change moment is extracted as the starting node of the conflict surge. For example, under the condition that the steady-state residual variance is 0.15, the variance suddenly jumps to 0.82 at 252.4 seconds and continues to amplify; this point is identified as the surge singularity.

[0074] Finally, the discrete starting nodes are sequentially concatenated and extended along the energy dimension to form a surge zone. These starting nodes are then connected along the time axis to form the backbone of the conflict surge zone. To enable this surge zone to provide both early warning and late-stage buffering, a pre-warning zone is defined as the segment where the rhythmic intrusion degree η drops to 0.5 to 1.0, and a post-warning zone is defined as the segment where the deviation variance converges to within twice the steady-state value. Through this bidirectional extension, the surge zone is not only a time interval but also modeled as an energy reservoir. Its area integral represents the amount of conflict energy accumulated in the system, providing a quantitative basis for calculating the intensity of subsequent intervention measures.

[0075] Step 3: Perform cross-scale scheduling instruction timing decoupling intervention based on elastic time window: Within the time window of the surge zone, actively inject controlled micro-offsets into the top-level baseline scheduling sequence to break the timing synchronization between the top-level extreme value and the middle-level power response sensitive interval, reconstruct the decoupling response baseline, and cut off the multi-objective positive feedback coupling channel.

[0076] Specifically, within the time window of the time-domain surge zone, the limitations of traditional rigid power tracking are overcome by actively injecting controlled micro-offsets into the top-level reference scheduling sequence, thus breaking the time synchronization between the top-level extreme values ​​and the sensitive intervals of the mid-level power response. Combining energy conservation constraints with an elastic micro-time difference injection strategy, a conflict-free decoupled response baseline is reconstructed, fundamentally severing the multi-objective positive feedback coupling channel.

[0077] In this embodiment, cross-scale scheduling instruction timing decoupling intervention based on elastic time windows is implemented. The specific process includes:

[0078] Within the surge time window, high-risk coupled node pairs with overlapping timelines are first identified. The timing of the top-level baseline extreme values ​​is then compared one by one. When the mid-level deviation surge occurs If the absolute time difference between the two Nodes with a duration less than one-third of a single scheduling cycle are identified as synchronously coupled nodes. For these nodes, the area S of the surge-driven energy storage tank at the current moment and the current SOC range ΔSOC of the underlying battery cluster are extracted as input parameters for calculating the time-series offset.

[0079] To address this issue, an adaptive elastic micro-time difference calculation model is proposed for peak-shifting intervention. The model is as follows:

[0080] ,in and This is a coefficient set based on system inertia. If the top-level extreme value is earlier than the surge node and the time difference is too small, then the instruction issuance time corresponding to that extreme value is affected. The delay buffer; if the top-level extreme value lags behind the surge node, its distribution time is slightly advanced. .

[0081] For example, taking a 10MW system as an example, it was detected that the original extreme value of the top-level plan lagged behind the surge node by 20 seconds and overlapped by 45 seconds. The system calculated the elastic micro-time difference. The timeframe is 30 seconds. The system advances the issuance time of this extreme value by 30 seconds (from 320 seconds to 290 seconds) to completely remove it from the influence radius of the surge node. At the same time, in order not to disrupt the overall power balance constraint, after the command shift, a power compensation wavelet with halved amplitude and doubled time is introduced in the subsequent non-surge period to ensure that the energy integral (i.e., throughput power) before and after the shift is absolutely conserved and does not disrupt the intraday energy scheduling plan.

[0082] Specifically, after the peak shifting operation is completed, top-level commands and bottom-level state data are re-acquired to generate a decoupled response baseline. On the new baseline, the top-level power peak and the extreme values ​​of the mid-level power deviation exhibit an alternating peak shifting arrangement, eliminating temporal overlap areas. This cross-scale active peak shifting based on an elastic time window cuts off the positive feedback channels between different control objectives in the time dimension, preventing the intermediate coordination layer from facing the instantaneous deadlock dilemma of simultaneously responding quickly to high-frequency commands and maintaining inter-cluster balance when handling power allocation. The system regains a dynamic and stable adjustment margin.

[0083] Step 4, Spatiotemporal dual-dimensional adaptive shaping and energy flood discharge control of the bottom-level converter power slope: Based on the decoupled response baseline, for transient power step, the energy release cycle is extended in the time dimension by dynamically varying the slope, and power-oriented allocation is performed in the spatial dimension by combining the differences in cluster-level health status to generate an updated response baseline;

[0084] Specifically, based on the decoupled baseline, a spatiotemporal adaptive shaping strategy is proposed to address the transient power step surges faced by the underlying PCS and DC / DC converter. In the time dimension, the energy release cycle is extended by dynamically varying the slope; in the spatial dimension, power-guided allocation is performed by combining cluster-level SOH differences, transforming concentrated current surges into long-cycle, multi-branch distributed releases, thus completely suppressing battery rate exceedances and local thermal runaway.

[0085] In this embodiment, the spatiotemporal two-dimensional adaptive shaping of the power slope and energy flood discharge control of the underlying converter specifically includes the following process:

[0086] High-risk boundaries of the power step impact region are identified along the decoupling baseline, and the differential value of the DC-side power of the underlying PCS is calculated in real time. When the interval between adjacent step jumps is much lower than the system's average electromechanical time constant, and the local temperature rise rate of the battery cluster in that section is... When the thermal runaway warning threshold is exceeded, it is identified as a power concentration step impact zone. This region is usually accompanied by drastic switching of charging and discharging currents, which can easily trigger uneven lithium-ion concentration polarization inside the battery.

[0087] Specifically, dynamic variable slope shaping is performed in the time dimension within the impact zone, using the decoupling baseline as a reference trajectory to correct the duty cycle limiting and inner loop current loop PI parameters of the DC / DC converter online. Instead of traditional fixed slope limiting, a nonlinear slope limiter based on SOC feedback is adopted. When the SOC is at both ends (<10% or >90%), the upper limit of the slope is automatically tightened, significantly extending the ramp-up time; when the SOC is in the middle range (20%-80%), the slope is appropriately relaxed. During the power rise phase, the originally planned instantaneous large step command is decomposed into multiple S-shaped ramp commands with smoothing factors, reducing the current peak by extending the ramp-up time. During the power fall phase, the unloading rate of the PCS is controlled, and the liquid cooling system is activated early to avoid adiabatic temperature rise.

[0088] Furthermore, energy flooding allocation based on cluster-level health is implemented in the spatial dimension: under the premise of total power limitation, the intermediate coordination control layer no longer distributes power evenly, but differentiates it according to the SOH and real-time internal resistance of each battery cluster. For aging battery clusters with low SOH and high internal resistance, a smaller power share is allocated, reducing their C-rate; for healthy battery clusters, the share of impact power they bear is appropriately increased. Through this spatial dimension of "energy flooding," the impact current that was originally concentrated in a certain battery cluster is evenly distributed throughout the entire array, avoiding rate exceeding the limit caused by the local bottleneck effect.

[0089] Understandably, through the aforementioned spatiotemporal dual-dimensional shaping control, the underlying power response pattern is reconstructed. In the time domain, sharp step waves are transformed into smooth trapezoidal waves; spatially, current stress no longer concentrates on a single branch. The charge / discharge rate of the underlying battery is strictly controlled within a safe envelope, and lithium-ion diffusion within the cell tends to be unidirectional and uniform. This completely eliminates lithium plating or localized overheating conditions caused by high-rate stretching at the electrochemical mechanism level, greatly extending the battery's cycle life.

[0090] Step 5: Establish a long-term, full-time-domain rhythm phase locking and balance maintenance mechanism: Based on the updated response baseline, use the concept of a phase-locked loop to fine-tune the phase difference between the top-level scheduling and the bottom-level response, and combine feedforward prediction and feedback correction to lock the system in the peak-shifting and decoupling state.

[0091] Specifically, based on the baseline after shaping control, the phase difference between the top-level scheduling and the bottom-level response is continuously fine-tuned at the system level using the concept of phase-locked loops. By combining feedforward prediction and feedback correction, the system can lock into a stable peak-shifting and decoupling state during long-term operation, ensuring smooth and continuous energy throughput even under strong grid disturbances.

[0092] In this embodiment, a long-term, full-time-domain rhythm phase locking and balance maintenance mechanism is established. The specific process includes:

[0093] During long-term system operation, a phase tracker similar to a phase-locked loop (PLL) is introduced to monitor the phase drift between the top-level extrema and the bottom-level response extrema, and to extract the time difference sequence between the two in real time. If the difference remains constant within the dead zone, it indicates that the system is in the coordination interval; if the difference shows a convergent or divergent trend, it indicates that the rhythm has drifted. For example, when the time difference between the top-level extreme value and the bottom-level response extreme value narrows from the baseline of 0.25s to 0.13s, the system determines that the top-level value has once again invaded the sensitive zone. At this time, the phase tracker outputs a negative adjustment signal to apply a 0.12s delay compensation to the top-level command in the next cycle; conversely, if the time difference widens to 0.42s, a positive adjustment signal is output to apply an advance to the top-level command, forcibly pulling the time difference back to the stable decoupling band of 0.25s.

[0094] To improve response speed, a feedforward prediction compensation based on historical data is introduced. The system uses a Long Short-Term Memory (LSTM) network or an autoregressive model to predict the possible timing and duration of the next top-level dispatch extreme value based on the grid load fluctuation patterns and electricity price trends of the previous N cycles. If the prediction results indicate that the extreme value of the next cycle will fall within the dense area of ​​high-frequency AGC commands, the system not only performs feedback fine-tuning but also injects a small reverse energy pulse at the end of the current cycle in advance (strictly constrained by the BMS safety threshold) to change the initial SOC state of the underlying battery. This physically induces the top-level dispatch logic to proactively adjust the timing of its issuance in the next cycle, achieving feedforward decoupling to "prevent problems before they occur."

[0095] By employing continuous periodic feedforward prediction and feedback fine-tuning, the system is locked into a fixed, complementary temporal state. In this state, the power allocation cycle of the middle layer and the scheduling cycle of the top layer maintain peak-shaving coordination, and the temporal conditions for multi-objective conflicts are permanently destroyed. Thus, under the coupled conditions of multiple objectives such as economic benefits, lifespan support, and grid response, the energy storage system can maintain unidirectional and stable energy flow, avoid command oscillation overshoot, and achieve safe and efficient operation throughout its entire life cycle from the top level of the control architecture.

[0096] This invention constructs a multi-dimensional spatiotemporal coupling response baseline across multiple levels and innovatively introduces an "energy reservoir" boundary model to quantify multi-objective conflict surge zones, achieving pre-diagnosis and precise location of multi-timescale coupling conflicts in energy storage systems. Unlike traditional methods that only passively limit power amplitude, this invention utilizes the synergy of time-series decoupling intervention with an elastic time window and spatiotemporal dual-dimensional adaptive shaping control of the underlying converter to transform drastic power jumps at the underlying level into long-period, multi-branch distributed releases. This mechanism not only cuts off the positive feedback coupling channel in the time dimension but also fundamentally weakens the current superposition effect by channeling the impact current in the spatial dimension. It effectively suppresses the risk of rate exceeding limits and local thermal runaway in battery clusters, providing comprehensive and deep physical protection for the underlying cells and converter power devices, significantly delaying the cyclic aging process of energy storage batteries.

[0097] This invention constructs a long-term phase-locking mechanism based on a phase-locked loop (PLL) and predictive feedforward at the system level, completely blocking the timing conditions that cause repeated triggering of multi-objective conflicts. Through dynamic guidance of the baseline phase and energy integral conservation constraints, the energy flow within the energy storage system remains highly continuous and stable even when facing strong grid disturbances. The depth and rate of battery charging and discharging responses are more balanced, fundamentally eliminating local overcharging or deep discharging phenomena. This method enables the energy storage system to maintain a stable energy throughput state under complex multi-objective coupling conditions, significantly reducing the operational risks caused by grid fluctuations, and providing a highly forward-looking underlying control guarantee for energy storage power stations to participate in grid frequency regulation and peak shaving ancillary services and energy arbitrage in the long term.

[0098] like Figure 3 As shown, this embodiment also provides a multi-objective coordinated control system for energy storage systems based on a hierarchical architecture, used to implement the above-mentioned multi-objective coordinated control method for energy storage systems based on a hierarchical architecture, including:

[0099] The baseline generation module is used to establish a cross-level heterogeneous data spatiotemporal mapping mechanism. It performs event-level alignment and feature fusion on the same time axis between the top-level discrete baseline scheduling sequence and the bottom-level continuous battery cluster state sequence to generate a power state response baseline.

[0100] The surge band definition module is used to identify the evolution segment where the top-level scheduling cycle is forced to shorten due to high-frequency disturbances in the power grid, based on the response baseline, by introducing a rhythmic intrusion index. It also captures the critical singular point of the deviation surge by combining the residual divergence characteristics of the middle-level power allocation, and connects the singular point in the time domain and extends it bidirectionally to form a continuous time-domain surge band.

[0101] The timing decoupling module is used to actively inject controlled micro-offsets into the top-level reference scheduling sequence within the time window of the surge zone, break the timing synchronization between the top-level extreme value and the middle-level power response sensitive interval, reconstruct the decoupling response baseline, and cut off the multi-objective positive feedback coupling channel.

[0102] The shaping and flood discharge control module is used to extend the energy release cycle in the time dimension by dynamically varying the slope for transient power steps, and to perform power-guided allocation in the spatial dimension by combining the differences in cluster-level health status, based on the decoupled response baseline, to generate an updated response baseline.

[0103] The phase locking and maintenance module is used to fine-tune the phase difference between the top-level scheduling and the bottom-level response based on the updated response baseline, using a phase-locked loop-like concept, and combining feedforward prediction and feedback correction to lock the system in a peak-shifting decoupling state.

[0104] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-objective coordinated control method for an energy storage system based on a hierarchical architecture, characterized in that, Includes the following steps: S1: Establish a cross-level heterogeneous data spatiotemporal mapping mechanism to perform event-level alignment and feature fusion of the top-level discrete baseline scheduling sequence and the bottom-level continuous battery cluster state sequence on the same time axis to generate a power state response baseline. S2: Based on the response baseline, the evolution segment where the top-level scheduling cycle is forced to shorten due to high-frequency disturbances in the power grid is identified by introducing a rhythmic intrusion index to determine that the system has entered a multi-objective conflict incubation period. Combined with the residual divergence characteristics of the middle-level power allocation, the critical singular point of the deviation surge is captured. The singular point is connected in series in the time domain and extended in both directions to form a continuous time domain surge band. S3: Within the time window of the surge zone, a controlled micro-offset is actively injected into the top-level baseline scheduling sequence to break the time synchronization between the top-level extreme value and the sensitive interval of the middle-level power response, and to reconstruct the decoupled response baseline in order to cut off the multi-objective positive feedback coupling channel. S4: Based on the decoupled response baseline, for transient power step, the energy release cycle is extended in the time dimension by dynamically varying the slope, and power-guided allocation is performed in the spatial dimension by combining the differences in cluster-level health status to generate an updated response baseline. S5: Based on the updated response baseline, the phase difference between the top-level scheduling and the bottom-level response is finely adjusted using a phase-locked loop-like approach. Combined with feedforward prediction and feedback correction, the system locks into a peak-shifting and decoupled state.

2. The multi-objective coordinated control method for an energy storage system based on a hierarchical architecture according to claim 1, characterized in that, The specific process of generating a power state response baseline includes: A multi-dimensional high-frequency synchronous sensing array is deployed on the cross-layer data interaction link. The top-level communication bus bypass access message parsing probe captures and schedules the message distribution. Sensors are deployed on the bottom-level DC bus section to obtain cluster-level status data, and the timestamp is aligned using the network time protocol or IEEE1588 protocol. By introducing a dynamic time warping algorithm combined with sliding window cross-correlation analysis to perform time series elastic alignment, the step front of the top-level scheduling instruction is used as the driving anchor point, and the corresponding response inflection point in the bottom-level state sequence is found as the response anchor point. The elastic delay compensation factor is calculated for pre- or post-compensation, and a continuous spatiotemporal time chain is constructed. In the continuous spatiotemporal time chain, the extreme points, step change points and rate of change of the top-level reference power curve are extracted as the driving event set, and the extreme points of the first derivative of the bottom-level battery cluster SOC curve and the extreme points of inter-cluster temperature difference are identified as the response event set. The causal correlation determination algorithm is used to pair the driving events with the next nearest and most correlated response events, and connect them sequentially to form the power state response baseline.

3. The multi-objective coordinated control method for an energy storage system based on a hierarchical architecture according to claim 2, characterized in that, The specific process for determining whether the system has entered the multi-objective conflict incubation period in step S2 includes: The evolution of the top-level rhythm under the intrusion of high-frequency grid rhythms is scanned along the response baseline, and the actual time interval sequence of adjacent top-level scheduling extrema is calculated. Fluctuation cycle sequence of high-frequency commands from the power grid Define rhythmic invasiveness index ; when Shrinking to the second level and When the value continues to decrease and converges to within a preset threshold, the system is determined to have entered the multi-objective conflict incubation period.

4. The multi-objective coordinated control method for an energy storage system based on a hierarchical architecture according to claim 3, characterized in that, The specific process for forming the continuous temporal burst band in step S2 includes: The variance coefficient of variation method is used to calculate the instantaneous residual sequence of the actual power distribution in the middle layer and the ideal smooth trajectory along the time axis, and the local variance of the residual sequence is calculated. When the local variance changes from steady-state small-amplitude fluctuations to exponential divergence over multiple consecutive control periods, and the sign of the variance derivative changes from positive to negative, the first mutation moment is extracted as the starting node of the conflict surge. The starting nodes are sequentially connected along the timeline to form the main trunk, and the rhythmic invasiveness is extracted forward. The segment within the second preset threshold range is designated as the warning zone, and the segment whose deviation variance converges to within twice the steady state is designated as the decay zone, forming a continuous time-domain surge band with an energy reservoir effect.

5. The multi-objective coordinated control method for an energy storage system based on a hierarchical architecture according to claim 4, characterized in that, The specific process for reconstructing the decoupled response baseline includes: Within the surge time window, compare the timing of the top-level benchmark extreme values. When the mid-level deviation surge occurs If the absolute time difference between the two If the duration is less than one-third of a single scheduling cycle, it is determined to be a synchronous coupling node; Extract the current surge energy storage tank area S and the current SOC range ΔSOC of the underlying battery cluster, and substitute them into the adaptive elastic micro-time difference calculation model to calculate the time series offset. ,in and The coefficient is set according to the system inertia; For the aforementioned synchronous coupling node, if the top-level extreme value is earlier than the surge node and the time difference is too small, then an application is made to the instruction issuance time corresponding to that extreme value. The delay buffer; if the top-level extreme value lags behind the surge node, its distribution time is slightly advanced. In subsequent non-surge periods, a power compensation wavelet with halved amplitude and doubled time is introduced to ensure absolute conservation of energy integral before and after translation, generating a decoupled response baseline.

6. The multi-objective coordinated control method for an energy storage system based on a hierarchical architecture according to claim 5, characterized in that, The specific process for generating the updated response baseline includes: The differential value of the DC-side power of the underlying PCS is calculated in real time along the decoupling response baseline. When the interval between adjacent step jumps is much lower than the system's average electromechanical time constant, and the local temperature rise rate of the battery cluster in that section is... When the thermal runaway warning threshold is exceeded, it is determined to be in the power concentration step impact zone; Perform dynamic slope shaping over time within the impact zone; In the spatial dimension, energy flooding allocation is performed based on cluster-level health. Differentiated guidance is given according to the SOH and real-time internal resistance of each battery cluster. A smaller power share is allocated to aged battery clusters with low SOH and high internal resistance to reduce their C-rate, while the impact power share borne by healthy battery clusters is increased, generating an updated response baseline after spatiotemporal two-dimensional shaping.

7. The multi-objective coordinated control method for an energy storage system based on a hierarchical architecture according to claim 6, characterized in that, The specific process of performing dynamic slope shaping in the time dimension within the impact zone includes: Dynamic variable slope shaping is performed using a nonlinear slope limiter based on SOC feedback within the power-concentrated step impact region. When the SOC is in the preset two-end range, the slope limit is automatically tightened; when the SOC is in the preset middle range, the slope limit is increased. During the power rise period, the original instantaneous large step command is broken down into multiple S-shaped ramp commands with smoothing factors to reduce the current peak. During the power reduction period, control the unloading rate of the converter and start the liquid cooling system in advance to avoid adiabatic temperature rise.

8. The multi-objective coordinated control method for an energy storage system based on a hierarchical architecture according to claim 1, characterized in that, The specific process for locking the peak-shifting decoupling state includes: The time difference sequence between the top-level extreme value and the bottom-level response extreme value is extracted in real time. When the absolute value of the time difference sequence decreases to a preset first time difference threshold, a negative adjustment signal is output to apply delay compensation to the top-level instruction of the next cycle. When the absolute value of the time difference sequence increases to a preset second time difference threshold, a positive adjustment signal is output to apply advance to the top-level command, forcing the time difference back to the stable decoupling zone; By using long short-term memory networks or autoregressive models, based on the grid load fluctuation patterns and electricity price change trends of the previous N periods, the timing and duration of the next top-level dispatch extreme value can be predicted. If the predicted extreme value of the next cycle falls into the high-frequency instruction dense area, a reverse energy pulse constrained by the safety threshold of the battery management system is injected in advance at the end of the current cycle to change the initial SOC state of the underlying battery, so as to induce the top-level scheduling logic to actively adjust the delivery time in the next cycle.

9. A multi-objective coordinated control system for an energy storage system based on a hierarchical architecture, used to implement the multi-objective coordinated control method for an energy storage system based on a hierarchical architecture as described in any one of claims 1-8, characterized in that, include: The baseline generation module is used to establish a cross-level heterogeneous data spatiotemporal mapping mechanism. It performs event-level alignment and feature fusion on the same time axis between the top-level discrete baseline scheduling sequence and the bottom-level continuous battery cluster state sequence to generate a power state response baseline. The surge band definition module is used to identify the evolution segment where the top-level scheduling cycle is forced to shorten due to high-frequency disturbances in the power grid, based on the response baseline, by introducing a rhythmic intrusion index. It also captures the critical singular point of the deviation surge by combining the residual divergence characteristics of the middle-level power allocation, and connects the singular point in the time domain and extends it bidirectionally to form a continuous time-domain surge band. The timing decoupling module is used to actively inject controlled micro-offsets into the top-level reference scheduling sequence within the time window of the surge zone, break the timing synchronization between the top-level extreme value and the middle-level power response sensitive interval, reconstruct the decoupling response baseline, and cut off the multi-objective positive feedback coupling channel. The shaping and flood discharge control module is used to extend the energy release cycle in the time dimension by dynamically varying the slope for transient power steps, and to perform power-guided allocation in the spatial dimension by combining the differences in cluster-level health status, based on the decoupled response baseline, to generate an updated response baseline. The phase locking and maintenance module is used to fine-tune the phase difference between the top-level scheduling and the bottom-level response based on the updated response baseline, using a phase-locked loop-like concept, and combining feedforward prediction and feedback correction to lock the system in a peak-shifting decoupling state.

Images