A multi-time scale energy management method for energy storage power stations

Abstract

This application relates to a multi-timescale energy management method for energy storage power stations, comprising the following steps: injecting wideband small-amplitude excitation pulses into the battery clusters during the dead time of charge-discharge switching of the energy storage converter and simultaneously acquiring voltage response signals; performing a fast Fourier transform on the voltage response signals to extract a wideband electrochemical impedance spectrum; adaptively selecting an equivalent circuit model based on the current SOC value to perform online fitting of the electrochemical impedance spectrum to obtain key aging parameters; dynamically generating a safety scheduling boundary based on the real-time change rate of the key aging parameters, the safety scheduling boundary including the maximum allowable charge-discharge power and the available SOC range; embedding the safety scheduling boundary as a constraint condition into a multi-timescale optimization scheduling model to generate scheduling instructions for each battery cluster. This application achieves synergistic optimization of battery life protection and power station economic benefits, improving the full life-cycle economic efficiency and operational safety of energy storage power stations.

Description

Technical Field

[0001] This application relates to the field of energy storage power station management, and in particular to a multi-timescale energy management method for energy storage power stations. Background Technology

[0002] As a key infrastructure supporting a high proportion of renewable energy integration, energy storage power stations utilize an Energy Management System (EMS) to coordinate the charging and discharging behavior of battery clusters to respond to grid dispatch demands and maximize economic benefits. Current mainstream energy management methods typically employ a multi-timescale optimization framework. For example, during the day-ahead phase, charging and discharging strategies for the following day are planned based on forecast data; during the intraday phase, rolling adjustments are made based on real-time electricity prices and load fluctuations; and power point tracking is executed at the second or millisecond level through a local controller. This hierarchical dispatching model balances economic efficiency and response speed to a certain extent and is widely used in grid-side energy storage, renewable energy distribution and storage, and shared energy storage power stations.

[0003] However, existing energy management methods have fundamental flaws in sensing and utilizing battery health status. Scheduling optimization models typically focus only on instantaneous power balance and short-term economic benefits, with battery lifespan considerations remaining at a statistical level. For example, they estimate health status based on cumulative throughput or set fixed SOC operating windows and power limits as safety constraints. This "statistical posterior" approach fails to perceive real-time electrochemical changes within the battery, such as the risk of lithium plating indicated by a sudden increase in ohmic internal resistance, or the accelerated degradation of active materials reflected by a continuous rise in charge transfer impedance. When scheduling commands are executed according to fixed boundaries even when the battery is in a deterioration range, it not only accelerates lifespan degradation but may also lead to safety accidents due to the inability to identify high-risk states. Furthermore, the decoupling of scheduling boundaries from the actual battery state in existing methods forces the system to trade off between "overly conservative" and "underprotective" approaches, making it difficult to achieve a balance between economic efficiency and safety throughout the entire battery lifecycle.

[0004] To address the aforementioned issues, this application proposes a multi-timescale energy management method for energy storage power stations. Summary of the Invention

[0005] To address the aforementioned problems, this application provides a multi-timescale energy management method for energy storage power stations, employing the following technical solution: A multi-timescale energy management method for energy storage power stations includes the following steps: During the dead time of the energy storage converter's charge-discharge switching, a wide-band small-amplitude excitation pulse is injected into the battery cluster, and the voltage response signal of the battery cluster is collected simultaneously. The voltage response signal is subjected to a fast Fourier transform to extract the broadband electrochemical impedance spectrum of the battery cluster. Based on the current SOC value of the battery cluster, an equivalent circuit model is adaptively selected to fit the electrochemical impedance spectrum online to obtain the key aging parameters of the battery cluster. The key aging parameters include at least the ohmic internal resistance R0 and the charge transfer impedance Rct. Based on the real-time change rate of the key aging parameters, the safety scheduling boundary of the battery cluster is dynamically generated. The safety scheduling boundary includes the maximum allowable charge and discharge power and the available SOC range. The safety scheduling boundary is used as a constraint and embedded into the multi-timescale optimization scheduling model of the energy management system to generate and issue scheduling instructions for each battery cluster.

[0006] Preferably, the adaptive selection of the equivalent circuit model includes: When the SOC is below the first threshold or above the second threshold, a second-order RC equivalent circuit model is adopted. When the SOC is between the first threshold and the second threshold, a first-order RC equivalent circuit model is used. The first threshold is 20%, and the second threshold is 80%.

[0007] Preferred options also include, When the rate of change of the ohmic internal resistance R0 exceeds 5% in a single charge-discharge cycle, the equivalent circuit model is refitted, and the equivalent circuit parameters are recalibrated.

[0008] Preferably, the step of dynamically generating the safety scheduling boundary based on the real-time change rate of key aging parameters includes: Monitor the instantaneous change slope of the ohmic internal resistance R0. If it exceeds the first preset threshold, it is determined that there is a risk of lithium plating. The maximum allowable charge and discharge power of the corresponding battery cluster is reduced to less than 50% of the rated power, and the battery cluster is controlled to enter the static recovery mode. The rate of increase of charge transfer impedance Rct is monitored. If it exceeds the second preset threshold, it is determined that the active material is decaying rapidly, and the SOC usable range of the battery cluster is dynamically reduced.

[0009] Preferably, the amplitude of the excitation pulse is 1% to 3% of the rated voltage of the battery cluster, the frequency range is 0.1Hz to 1000Hz, and the pulse duration is 10ms to 100ms. The Fast Fourier Transform is executed by the FPGA hardware module built into the edge controller, and the calculation time for a single electrochemical impedance spectroscopy inversion is no more than 5ms.

[0010] Preferably, the multi-timescale optimization scheduling model is a model predictive control model, which performs global optimization with a 15-minute rolling cycle and performs real-time power tracking with a second-level cycle. The scheduling instructions are sent from the cloud-based centralized optimizer to the edge controllers of each battery cluster. The edge controllers dynamically adjust their local control parameters based on the latest safety scheduling boundaries.

[0011] Preferred options also include: The edge controllers of each battery cluster report key aging parameters to the cloud-based centralized optimizer in real time. The cloud-based centralized optimizer prioritizes each battery cluster based on its health status and prioritizes battery clusters with lighter aging levels to participate in high-frequency frequency modulation tasks in subsequent scheduling.

[0012] Preferred options also include a self-governing mode that is enabled even when the network is down: When communication between the cloud-based centralized optimizer and the edge controller is interrupted, the edge controller will autonomously determine whether to participate in frequency regulation based on the local electrochemical impedance spectroscopy inversion results and the preset grid frequency-power droop curve. Once communication is restored, the edge controller will synchronize its local operating status to the cloud-based centralized optimizer.

[0013] Preferred options also include: The real-time ohmic internal resistance R0, charge transfer impedance Rct, and their rate of change of each battery cluster are used as state variables to build a digital twin model of the entire battery system in the cloud. This model is used to predict the future aging trend of the battery clusters and optimize the medium- and long-term scheduling strategy of the energy storage power station.

[0014] Preferably, the energy storage converter has a built-in harmonic injection module and a high-speed ADC acquisition circuit. The injection of excitation pulses and the acquisition of voltage response signals are triggered by the edge controller, and the entire process does not interrupt the normal charging and discharging operation of the energy storage converter.

[0015] In summary, this application includes at least one of the following beneficial technical effects: 1. By injecting wideband, small-amplitude excitation pulses during the inherent dead time of the energy storage converter's charge-discharge switching, and combining this with FPGA hardware-accelerated Fast Fourier Transform, online inversion of electrochemical impedance spectroscopy is achieved without interrupting the normal operation of the energy storage system. Compared to traditional methods that rely solely on cumulative throughput or open-circuit voltage to estimate health status, this application can acquire key aging parameters such as ohmic internal resistance R0 and charge transfer impedance Rct in real time, elevating battery life management from statistical posterior to real-time state perception, and providing precise physical basis for scheduling decisions.

[0016] 2. A safe scheduling boundary is dynamically generated based on the real-time change rates of R0 and Rct. When R0 suddenly changes beyond the threshold, the power is actively reduced and the battery is left stagnant to avoid the risk of lithium plating. When Rct continues to rise, the usable SOC range is dynamically contracted to delay activity decay. This application abandons the conservative approach of fixing the SOC window and power limit in traditional methods, maximizing the release of battery potential while ensuring safety, preventing the battery from operating in the degradation range from the source, and significantly extending the battery cycle life. A dynamic mapping mechanism between electrochemical state and scheduling constraints is established, realizing the transition from fixed boundary to adaptive boundary.

[0017] 3. The dynamically generated safety scheduling boundary is embedded as a core constraint into the Model Predictive Control (MPC) optimization model, making the generation of scheduling instructions directly constrained by the real-time electrochemical state of the battery. The cloud performs global optimization with a rolling cycle of 15 minutes, while the edge controller performs real-time tracking with a cycle of seconds, forming a complete closed loop of "perception-constraint-scheduling-execution", realizing the synergistic optimization of battery life protection and power plant economic benefits.

[0018] 4. This application designs a cloud-edge collaborative distributed control architecture that balances global optimization with local response reliability. The edge controller possesses capabilities such as EIS inversion, security boundary generation, and local autonomy, while the cloud-based centralized optimizer performs health ranking and power allocation based on aging parameters reported by each cluster. When communication is interrupted, the edge controller automatically switches to a network-off-grid autonomous mode, dynamically adjusting the droop coefficient based on local EIS sensing results, and can still autonomously participate in grid frequency regulation even offline. This architecture ensures both the economic efficiency of cloud-based global optimization and operational safety and grid responsiveness during communication failures.

[0019] 5. A comprehensive health index is constructed based on the real-time reported R0 and Rct values ​​of each battery cluster. In subsequent scheduling, the cloud prioritizes clusters with lower health indices (lighter aging) for high-frequency frequency regulation tasks, while assigning clusters with heavier aging to low-frequency tasks such as peak shaving and standby. This strategy effectively balances the aging rate of all battery clusters across the entire station, preventing some clusters from decaying too quickly due to frequent calls, improving the overall lifespan and capacity utilization of the energy storage power station, and achieving differentiated scheduling and balanced aging of multiple battery clusters.

[0020] 6. This application establishes a closed-loop energy management system covering the entire lifecycle and all time scales. By constructing a digital twin model based on historical EIS data in the cloud, it utilizes time-series prediction algorithms to accurately predict battery aging trends and inputs medium- to long-term scheduling strategies (such as monthly charge-discharge plans) as boundary conditions into the short-term MPC model, forming a complete technology chain of "long-term planning - short-term optimization - real-time control." This application breaks through the limitations of traditional energy management that only focuses on short-term economic benefits, achieving full-domain coordination from millisecond-level inertial response to weekly energy arbitrage, providing a systematic solution for the full lifecycle economic optimization of energy storage power stations. Attached Figure Description

[0021] Figure 1 This is a flowchart of a multi-timescale energy management method for an energy storage power station according to an embodiment of this application. Detailed Implementation

[0022] The following is in conjunction with the appendix Figure 1 This application will be described in further detail.

[0023] This application provides a multi-timescale energy management method for energy storage power stations. The core of this method lies in: utilizing the dead time of charge-discharge switching in the energy storage converter to inject wideband, small-amplitude excitation pulses into the battery clusters and acquiring the voltage response; extracting the electrochemical impedance spectrum through fast Fourier transform; adaptively selecting an equivalent circuit model based on the battery's state of charge (SOC) to fit key aging parameters such as ohmic internal resistance R0 and charge transfer impedance Rct; dynamically generating a safe scheduling boundary (including maximum allowable power and available SOC range) based on the real-time change rate of these parameters; and finally embedding this boundary as a constraint into a multi-timescale optimization scheduling model to generate scheduling instructions for each battery cluster. This method achieves closed-loop linkage between online sensing of the battery's internal state and scheduling decision-making, effectively avoiding operation in deterioration zones and improving the overall lifecycle economy and operational safety of the energy storage power station. The implementation of each step is described in detail below with specific embodiments.

[0024] Example 1: Complete Implementation Process of this Application The energy storage power station used in this embodiment consists of 100 battery clusters connected in parallel. Each battery cluster has a rated voltage of 800V and a rated capacity of 200kWh, using lithium iron phosphate batteries. The energy storage converter (PCS) adopts a bidirectional two-level topology and has a built-in harmonic injection module and a 16-bit high-speed ADC acquisition circuit. Each battery cluster is equipped with an edge controller, which uses a Xilinx Zynq UltraScale+MPSoC chip. Its FPGA logic part is used to implement high-speed calculations such as fast Fourier transform, and the ARM core part runs lightweight control algorithms. A centralized optimization server is deployed in the cloud and communicates with each edge controller via industrial Ethernet.

[0025] Reference Figure 1 A multi-timescale energy management method for energy storage power stations includes the following steps: Step S1: Inject an excitation pulse and acquire the response during the dead time of PCS charge / discharge switching: The edge controller monitors the operating status of the PCS in real time. When the PCS switches from a discharging state to a charging state, or vice versa, a brief dead-time window occurs. This dead-time is an inherent interval set to ensure the safety of the power devices, typically ranging from 10ms to 100ms. In this embodiment, the PCS dead-time is set to 50ms.

[0026] Upon detecting the arrival of the dead time, the edge controller immediately triggers the harmonic injection module built into the PCS to inject a wideband, small-amplitude excitation pulse into the target battery cluster. The amplitude of the excitation pulse needs to meet the following requirements: too low an amplitude will result in insufficient signal-to-noise ratio of the response signal, affecting impedance extraction accuracy; too high an amplitude may cause battery polarization or affect normal operation. Therefore, the excitation pulse amplitude in this application is set to 1% to 3% of the rated voltage of the battery cluster. Within this range, sufficient signal-to-noise ratio is ensured without causing significant interference to the battery. In this embodiment, considering that the rated voltage of the battery cluster is 800V, the middle value of 2.5% is taken, i.e., the excitation pulse amplitude is 20V.

[0027] The frequency range of the excitation pulse needs to cover the key frequency bands of the electrochemical impedance spectroscopy: the low-frequency band (0.1Hz to 1Hz) reflects the diffusion process, the mid-frequency band (1Hz to 100Hz) reflects the charge transfer process, and the high-frequency band (100Hz to 1000Hz) reflects the ohmic process. Therefore, this application sets the frequency range of the excitation pulse to 0.1Hz to 1000Hz to obtain complete impedance information. In this embodiment, a sine wave sequence with a logarithmic sweep from 0.1Hz to 1000Hz is generated using direct digital frequency synthesis (DDS) technology, containing a total of 31 frequency points, each lasting approximately 1.3ms, for a total duration of 40ms, which falls within the 10ms to 100ms range defined in the claims.

[0028] Simultaneously with the excitation pulse injection, the PCS's built-in high-speed ADC acquisition circuit synchronously acquires the terminal voltage response signal of the battery cluster at a sampling rate of 10kHz. The acquired raw data, after being filtered for anti-aliasing, is stored in the FPGA's on-chip BRAM of the edge controller, awaiting further processing. The entire injection and acquisition process is precisely triggered by the edge controller, and due to the utilization of the dead time during charge-discharge switching, the normal charge-discharge operation of the PCS is not interrupted.

[0029] This step achieves excitation pulse injection by reusing the inherent dead time of the PCS, without requiring additional energy storage system operating time. It provides a data foundation for online sensing of the battery's electrochemical state without affecting normal charging and discharging. At the same time, the wide-band, small-amplitude pulse design balances impedance detection accuracy and battery operation safety.

[0030] Step S2: Perform a fast Fourier transform on the voltage response signal to extract the electrochemical impedance spectroscopy: The edge controller's built-in FPGA hardware module performs Fast Fourier Transform (FFT) on the acquired excitation and response signals. In this embodiment, the FFT uses 1024 points and is implemented using a pipelined architecture, with each transformation taking approximately 0.3ms. Before performing the FFT, a Hanning window is applied to the signal to reduce spectral leakage.

[0031] After the FFT is completed, the complex ratio of the response signal to the excitation signal is calculated for each frequency point to obtain the impedance amplitude and phase at that frequency, thus forming a complete broadband electrochemical impedance spectroscopy. To ensure the reliability of the impedance data, frequency points with a signal-to-noise ratio below 3dB are discarded. In this embodiment, the entire EIS inversion calculation process is completed within the FPGA, and the calculation time for a single operation is controlled within 5ms, fully meeting the real-time requirements.

[0032] This step utilizes FPGA hardware to implement high-speed FFT operations, controlling the inversion time of electrochemical impedance spectroscopy to the millisecond level, meeting the real-time requirements of energy management across multiple time scales. At the same time, signal-to-noise ratio filtering improves the accuracy of impedance data, providing reliable data support for subsequent aging parameter fitting.

[0033] Step S3: Adaptively select an equivalent circuit model to fit key aging parameters based on the current SOC: The edge controller reads the current SOC value of the battery cluster from the battery management system (BMS) and ensures that this SOC value is aligned with the EIS measurement time. Based on the range of the SOC value, it adaptively selects different equivalent circuit models to fit the impedance spectrum. Specifically, when the SOC is below the first threshold (20%) or above the second threshold (80%), the battery is in a region with significant polarization effects, requiring a second-order RC equivalent circuit model for fitting. This model includes the ohmic internal resistance R0, the charge transfer impedance Rct, and a constant phase angle element Q connected in parallel with Rct, which can accurately reflect the concentration polarization characteristics in the low SOC region and the lithium plating risk in the high SOC region.

[0034] When the state of charge (SOC) is between 20% and 80%, the battery is in a relatively linear range of electrochemical characteristics, and a first-order RC equivalent circuit model can meet the fitting accuracy requirements. This model includes the ohmic internal resistance R0 and the charge transfer impedance Rct, which has a relatively small computational load and is beneficial for improving real-time performance.

[0035] The aforementioned first threshold of 20% and second threshold of 80% are based on the electrochemical characteristics of lithium-ion batteries: when the state of charge (SOC) is below 20%, the negative electrode potential increases, and the polarization resistance increases significantly; when the SOC is above 80%, the difficulty of lithium delithiation at the positive electrode increases, and the risk of lithium plating increases. Therefore, more refined models are needed to capture electrochemical changes in these two ranges.

[0036] In this embodiment, the current SOC is 65%, falling between 20% and 80%, so a first-order RC model is adopted. The Levenberg-Marquardt algorithm is used for fitting, with an upper limit of 50 iterations and a convergence condition that the change in the sum of squared residuals between two adjacent iterations is less than 1e-6. The fitting yields the following key aging parameters: ohmic internal resistance R0 = 5.2mΩ, and charge transfer impedance Rct = 8.7mΩ.

[0037] If the current SOC is below 20% or above 80%, a second-order RC equivalent circuit model is used for fitting. Taking SOC=15% as an example, the Levenberg-Marquardt algorithm is still used for fitting, with an upper limit of 50 iterations and the convergence condition remains unchanged. However, the model includes ohmic internal resistance R0, charge transfer impedance Rct, and constant phase angle element Q and diffusion impedance W connected in parallel with Rct, for a total of 5 parameters to be fitted. The fitted R0 and Rct are used as key aging parameters for subsequent steps.

[0038] This step, based on the SOC range adaptive switching equivalent circuit model, ensures fitting accuracy in the significant polarization range and reduces computation in the linear range, achieving a balance between detection accuracy and computational efficiency. It can accurately extract the ohmic internal resistance and charge transfer impedance that reflect the true aging state of the battery.

[0039] Step S4: Monitor the rate of change of R0 and trigger model refitting: The edge controller maintains a historical data queue, storing the R0 and Rct values ​​and their timestamps obtained from the most recent 10 EIS inversions. It calculates the relative rate of change of R0 within a single charge-discharge cycle (i.e., the percentage change compared to the previous cycle). When the relative rate of change of R0 exceeds 5%, it indicates a possible abrupt change in the battery's internal state (such as lithium plating, internal short circuit, etc.), at which point it is necessary to trigger the refitting of the equivalent circuit model.

[0040] The 5% threshold is derived from statistical analysis of lithium iron phosphate battery cycle aging experiments: during normal aging, the change rate of R0 per cycle is usually within 3%; when it exceeds 5%, it is often accompanied by abnormal aging phenomena such as lithium plating, and the model parameters need to be recalibrated.

[0041] In this embodiment, the rate of change of R0 relative to the previous cycle is 2%, which does not exceed 5%, so refitting is not triggered. If refitting is triggered, the edge controller will clear the fitting parameters of the current model and re-execute the model selection and fitting process in step S3 until convergence.

[0042] This step achieves adaptive triggering of model refitting by monitoring the relative rate of change of R0, which can respond promptly to sudden changes in the internal state of the battery, avoid aging perception bias caused by inaccurate model parameters, and improve the robustness of the entire state perception system.

[0043] Step S5: Dynamically generate a safe scheduling boundary based on key aging parameters: The edge controller dynamically generates the safety scheduling boundary of the battery cluster based on the real-time change rates of R0 and Rct, including the maximum allowable charge / discharge power and the available SOC range. The specific rules are as follows: (1) Monitor the instantaneous rate of change of the ohmic internal resistance R0, which is defined here as the relative rate of change of R0 between two adjacent EIS measurements (30 seconds apart in this embodiment) (i.e., the difference divided by the previous value). If the relative rate of change exceeds the first preset threshold, it is determined that there is a risk of lithium plating. The value range of the first preset threshold is 8% to 15%, and 10% is taken in this embodiment. When the threshold is exceeded, the maximum allowable charge and discharge power of the corresponding battery cluster is linearly reduced, and the reduction coefficient is proportional to the multiple of the threshold, but not less than 50% of the rated power. At the same time, the battery cluster is forced to enter the static recovery mode, and the static time is not less than 30 minutes. The 30-minute static time is based on the experimental statistics of the internal resistance recovery after lithium iron phosphate battery plating: the internal resistance can be restored to more than 90% of the stable value within 30 minutes.

[0044] (2) Monitor the continuous rise rate of charge transfer impedance Rct, defined as the linear fitting slope of Rct over the past 10 minutes (expressed as a percentage change per minute). If the rate exceeds a second preset threshold, it is determined that the active material is accelerating its decay. The second preset threshold ranges from 0.5% / minute to 1.5% / minute, and is set to 1.0% / minute in this embodiment. When the threshold is exceeded, the usable SOC range of the battery cluster is dynamically reduced: from the default [20%, 90%] to [30%, 80%], with the reduction magnitude proportional to the multiple of the threshold. If the continuous rise rate exceeds 2.0% / minute, it can be further reduced to [35%, 75%].

[0045] In this embodiment, the instantaneous relative change rate of R0 is 2%, which is less than 10%; the continuous rise rate of Rct is 0.8% / minute, which is less than 1.0% / minute. Therefore, the safety scheduling boundary remains at the default value: the maximum allowable charging and discharging power is 100% of the rated power, and the SOC usable range is 20% to 90%.

[0046] This step transforms the real-time electrochemical aging state of the battery into a safety boundary that can be directly used for scheduling, replacing the pseudo-constraints of traditional fixed power and fixed SOC range. It can identify abnormal risks such as lithium plating and active material degradation in advance, thus avoiding excessive battery aging and safety hazards from the source.

[0047] Step S6: Embed the safety scheduling boundary into the multi-timescale optimization scheduling model: The cloud-based centralized optimizer employs a Model Predictive Control (MPC) model for multi-timescale optimization scheduling. The MPC model uses a 15-minute rolling cycle to predict grid dispatch demand, spot electricity prices, and renewable energy output for the next four hours, solving for the power allocation scheme for each battery cluster. The 15-minute rolling cycle is chosen because: the clearing cycle of the domestic electricity spot market is typically 15 minutes, and the short-term forecast error for wind and solar power reaches an inflection point at the 15-minute scale (the 15-minute forecast error is usually less than 5%, and the error increases rapidly after 15 minutes). Therefore, 15 minutes is the engineering optimal value that balances forecast accuracy and real-time scheduling.

[0048] The objective function of the MPC model is to maximize revenue minus a lifetime loss penalty term, where revenue includes frequency regulation revenue, peak shaving revenue, and spot arbitrage revenue, etc. Constraints include: Power balance constraint: The sum of the power of each cluster equals the power grid dispatch command; SOC dynamic equation: SOC(k+1) = SOC(k) - η·P(k)·Δt / E; Dynamic safety boundary constraints: The power and SOC of each cluster must satisfy the dynamic boundary generated in step S5.

[0049] The cloud-based optimizer solves the optimization problem every 15 minutes and sends the generated scheduling instructions to each edge controller via the MQTT protocol. The edge controllers perform real-time power point tracking at second-level intervals and dynamically adjust local control parameters (such as the droop coefficient) according to the latest safety scheduling boundaries to ensure that safety constraints are always met during real-time operation.

[0050] This step embeds the dynamic safety boundary as a core constraint into the multi-timescale MPC scheduling model, achieving synergistic optimization of battery life protection and power plant economic benefits, while also taking into account the needs of medium- and long-term scheduling planning and real-time power point tracking.

[0051] Step S7: The edge controller reports aging parameters, and the cloud prioritizes them. Each battery cluster's edge controller reports R0, Rct, and their rate of change obtained from each EIS inversion to the cloud-based centralized optimizer in real time. The cloud then constructs a comprehensive health index based on these parameters, calculated using the following formula: HI=w1·(R0 / R0base)+w2·(Rct / Rctbase); Where R0base and Rctbase are factory baseline values, and w1 and w2 are weighting coefficients. In this embodiment, w1=w2=0.5. This equal weighting setting is based on experimental research: during the cycle aging process of lithium iron phosphate batteries, R0 and Rct contribute similarly to capacity decay, so equal weighting is used for comprehensive evaluation. The lower the health index, the lighter the battery aging.

[0052] In subsequent scheduling, the cloud prioritizes battery clusters with lower health indices for high-frequency frequency regulation tasks to balance the aging rate of each cluster. For battery clusters with higher health indices, low-frequency tasks such as peak shaving and backup are prioritized.

[0053] This step implements cluster-level scheduling priority sorting based on the real-time aging status of the batteries, which can balance the aging rate of each battery cluster in the entire station, prevent some clusters from decaying too quickly due to frequent calls, and improve the overall service life and capacity utilization of the energy storage power station.

[0054] Step S8: Switch to autonomous mode when communication is interrupted: In this embodiment, the cloud-based centralized optimizer and the edge controller maintain a connection via heartbeat packets. When the edge controller fails to receive a heartbeat response from the cloud three times consecutively, it determines that the communication has been interrupted and automatically switches to the offline autonomous mode.

[0055] In autonomous mode, the edge controller operates autonomously based on local EIS inversion results and a preset grid frequency-power droop curve. The baseline coefficient K0 of the droop curve is set as the ratio of the deviation between rated power and rated frequency. To balance lifetime protection in autonomous mode, the droop coefficient is dynamically adjusted based on R0: when R0 increases by no more than 10% from the baseline value, K=K0; when R0 increases by more than 10% but less than 20% from the baseline value, K=0.8K0, i.e., the frequency regulation participation depth is reduced by 20%; when R0 increases by more than 20% from the baseline value, K=0.5K0. The effect of Rct is similar, and the final droop coefficient is the smaller value after the two adjustments. Simultaneously, the edge controller locally records operating data, including power output, SOC changes, and EIS measurement results.

[0056] Once communication is restored, the edge controller synchronizes data with the cloud. Synchronization uses a timestamp mechanism to ensure seamless integration between local records and historical data in the cloud, avoiding data conflicts.

[0057] The autonomous mode set up in this step ensures that the energy storage system can still participate in grid frequency regulation autonomously when communication fails. At the same time, it dynamically adjusts control parameters based on local EIS sensing results, thus achieving a balance between operational safety and grid response capability in offline mode.

[0058] Step S9: Build a digital twin model in the cloud to optimize medium- and long-term scheduling: The cloud uses the R0, Rct, and their rate of change reported by each battery cluster over a long period as state variables to construct a digital twin model of the entire battery system. This model employs a time-series prediction algorithm based on Long Short-Term Memory (LSTM) networks and uses historical EIS data to train a battery aging trend prediction model.

[0059] Based on the prediction results of the digital twin model, the cloud optimizes the medium- and long-term scheduling strategies of energy storage power stations, such as monthly charging and discharging plans and annual maintenance plans. The medium- and long-term scheduling results are used as boundary conditions input to the short-term MPC model, forming a complete closed loop of long-term planning, short-term optimization, and real-time control.

[0060] This step builds a digital twin model based on real-time EIS sensing data, enabling accurate prediction of battery aging trends, providing data support for medium- and long-term scheduling strategy optimization, and forming a closed loop of energy management across the entire time scale and life cycle.

[0061] Example 2: Lithium plating risk triggering scenario This embodiment describes the system's response process when the instantaneous relative change rate of R0 exceeds a first preset threshold.

[0062] In two consecutive EIS measurements (30 seconds apart), the R0 of a certain battery cluster jumped from 5.2 mΩ to 5.8 mΩ, with a calculated relative change rate of (5.8 - 5.2) / 5.2 ≈ 11.5%, exceeding the first preset threshold of 10%. The edge controller determined that there was a risk of lithium plating and immediately performed the following actions: 1. The maximum permissible charge / discharge power of the battery cluster is linearly reduced from 100% to 50%; 2. Force the battery cluster to exit the current scheduling task and enter a static recovery mode for no less than 30 minutes; 3. Send an early warning message to the cloud, marking the cluster as a high-risk state.

[0063] After receiving the warning, the cloud will suspend the power allocation of the cluster in subsequent scheduling until it is idle and R0 returns to stability (the change rate of R0 is less than 5% in two consecutive measurements).

[0064] This embodiment achieves early identification and proactive intervention of lithium plating risk by real-time monitoring of R0 mutations, which can effectively prevent the further expansion of lithium plating risk, protect the safety of the battery itself, and reduce the probability of battery failure.

[0065] Example 3: Model refitting scenario triggered when the rate of change of R0 exceeds 5%. For a certain battery cluster, the relative change rate of R0 in 5 consecutive cycles is +1.2%, +1.5%, +1.8%, +2.3%, and +5.1%. The change rate in the 5th cycle exceeds the 5% threshold, triggering model refitting.

[0066] The edge controller clears the fitting parameters of the current equivalent circuit model and re-executes the adaptive model selection and fitting process in step S3. After refitting, it is found that the fitting error of the second-order RC model is significantly smaller than that of the first-order model, indicating that the battery polarization characteristics have changed. The system automatically switches to the second-order RC model and recalibrates the baseline value. Subsequent scheduling will be based on the updated model parameters.

[0067] This embodiment adaptively updates the equivalent circuit model for sudden changes in battery state, ensuring the continuous accuracy of aging parameter perception and providing a reliable basis for the dynamic generation of subsequent scheduling boundaries.

[0068] Through the above embodiments, this application combines online electrochemical impedance spectroscopy sensing technology with multi-timescale energy management, overcoming the limitations of traditional energy storage management that relies on fixed constraints and is detached from the actual electrochemical state of the battery. This achieves closed-loop coordination between real-time battery state sensing and scheduling decisions. The overall solution can accurately identify battery aging anomalies and safety risks while ensuring that energy storage power stations participate in grid dispatch and obtain economic benefits. It balances the aging rate of each battery cluster, effectively extending the cycle life of the battery clusters and improving the safety, reliability, and overall lifecycle economy of the energy storage power station. Simultaneously, the cloud-edge collaborative architecture and grid-outage autonomous design further enhance the system's operational stability and environmental adaptability, meeting the practical application needs of various energy storage power stations on both the grid and user sides.

[0069] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Although this application has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features in the various embodiments of this application as appropriate without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of this application. These technical solutions are also within the scope of protection of this application.

Claims

1. A multi-timescale energy management method for an energy storage power station, characterized in that, Includes the following steps: During the dead time of the energy storage converter's charge-discharge switching, a wide-band small-amplitude excitation pulse is injected into the battery cluster, and the voltage response signal of the battery cluster is collected simultaneously. The voltage response signal is subjected to a fast Fourier transform to extract the broadband electrochemical impedance spectrum of the battery cluster. Based on the current SOC value of the battery cluster, an equivalent circuit model is adaptively selected to fit the electrochemical impedance spectrum online to obtain the key aging parameters of the battery cluster. The key aging parameters include at least the ohmic internal resistance R0 and the charge transfer impedance Rct. Based on the real-time change rate of the key aging parameters, the safety scheduling boundary of the battery cluster is dynamically generated. The safety scheduling boundary includes the maximum allowable charge and discharge power and the available SOC range. The safety scheduling boundary is used as a constraint and embedded into the multi-timescale optimization scheduling model of the energy management system to generate and issue scheduling instructions for each battery cluster.

2. The multi-timescale energy management method for an energy storage power station according to claim 1, characterized in that, The adaptive selection of the equivalent circuit model includes: When the SOC is below the first threshold or above the second threshold, a second-order RC equivalent circuit model is adopted. When the SOC is between the first threshold and the second threshold, a first-order RC equivalent circuit model is used. The first threshold is 20%, and the second threshold is 80%.

3. The multi-timescale energy management method for an energy storage power station according to claim 2, characterized in that, Also includes: When the rate of change of the ohmic internal resistance R0 exceeds 5% in a single charge-discharge cycle, the equivalent circuit model is refitted, and the equivalent circuit parameters are recalibrated.

4. The multi-timescale energy management method for an energy storage power station according to claim 1, characterized in that, The dynamic generation of the safety scheduling boundary based on the real-time change rate of key aging parameters includes: Monitor the instantaneous change slope of the ohmic internal resistance R0. If it exceeds the first preset threshold, it is determined that there is a risk of lithium plating. The maximum allowable charge and discharge power of the corresponding battery cluster is reduced to less than 50% of the rated power, and the battery cluster is controlled to enter the static recovery mode. The rate of increase of charge transfer impedance Rct is monitored. If it exceeds the second preset threshold, it is determined that the active material is decaying rapidly, and the SOC usable range of the battery cluster is dynamically reduced.

5. The multi-timescale energy management method for an energy storage power station according to claim 1, characterized in that, The amplitude of the excitation pulse is 1% to 3% of the rated voltage of the battery cluster, the frequency range is 0.1Hz to 1000Hz, and the pulse duration is 10ms to 100ms. The Fast Fourier Transform is executed by the FPGA hardware module built into the edge controller, and the calculation time for a single electrochemical impedance spectroscopy inversion is no more than 5ms.

6. The multi-timescale energy management method for an energy storage power station according to claim 1, characterized in that, The multi-timescale optimization scheduling model is a model predictive control model that performs global optimization with a 15-minute rolling cycle and performs real-time power tracking with a second-level cycle. The scheduling instructions are sent from the cloud-based centralized optimizer to the edge controllers of each battery cluster. The edge controllers dynamically adjust their local control parameters based on the latest safety scheduling boundaries.

7. The multi-timescale energy management method for an energy storage power station according to claim 6, characterized in that, Also includes: The edge controllers of each battery cluster report key aging parameters to the cloud-based centralized optimizer in real time. The cloud-based centralized optimizer prioritizes each battery cluster based on its health status and prioritizes battery clusters with lighter aging levels to participate in high-frequency frequency modulation tasks in subsequent scheduling.

8. The multi-timescale energy management method for an energy storage power station according to claim 6, characterized in that, It also includes a self-governing mode that is disconnected from the internet: When communication between the cloud-based centralized optimizer and the edge controller is interrupted, the edge controller will autonomously determine whether to participate in frequency regulation based on the local electrochemical impedance spectroscopy inversion results and the preset grid frequency-power droop curve. Once communication is restored, the edge controller will synchronize its local operating status to the cloud-based centralized optimizer.

9. The multi-timescale energy management method for an energy storage power station according to claim 1, characterized in that, Also includes: The real-time ohmic internal resistance R0, charge transfer impedance Rct, and their rate of change of each battery cluster are used as state variables to build a digital twin model of the entire battery system in the cloud. This model is used to predict the future aging trend of the battery clusters and optimize the medium- and long-term scheduling strategy of the energy storage power station.

10. The multi-timescale energy management method for an energy storage power station according to claim 1, characterized in that, The energy storage converter has a built-in harmonic injection module and a high-speed ADC acquisition circuit. The injection of excitation pulses and the acquisition of voltage response signals are triggered by the edge controller, and the entire process does not interrupt the normal charging and discharging operation of the energy storage converter.

Images