Multi-time scale energy storage coordinated control method for pure new energy collection station

Abstract

The application discloses a kind of multi-time scale energy storage coordination control methods for pure new energy gathering station, comprising: obtaining new energy output prediction error sequence, prediction uncertainty index and the operating state data of energy storage unit;Based on error power spectral density distribution determines band demarcation frequency, and the decomposition of prediction error sequence is carried out, and multi-time scale error component sequence is extracted;According to operating state data, the functional layered configuration of energy storage unit is carried out, the reserved state of charge margin of each layer unit is determined in combination with prediction uncertainty index, and adjustable power interval is calculated;Based on operating state data, marginal aging cost is determined, to reduce cost and realize layered state of charge balance as target, under power interval constraint, adjust power is distributed and output instruction.The application realizes the physical matching of signal decomposition and hardware resource, effectively alleviates the control conflict of multiple types of energy storage units, improves the utilization rate and overall operating life of energy storage system.

Description

Technical Field

[0001] This invention relates to the field of energy storage control technology for new energy collection stations, and in particular to a multi-timescale coordinated control method for energy storage in pure new energy collection stations. Background Technology

[0002] With a high proportion of pure renewable energy (such as wind and solar power) integrated into the grid, the output of collection stations exhibits significant randomness and volatility. To meet the grid's requirements for smooth power transmission and dispatch, large-scale energy storage systems are typically configured at the collection station side. Energy storage systems compensate for prediction errors in renewable energy output by absorbing or releasing power, and are a key technical support for ensuring the safe and stable operation of the grid and improving the renewable energy absorption rate. Since the output fluctuations of renewable energy span multiple time scales from seconds to days, how to rationally allocate limited energy storage resources so that they can play an effective supporting role at different time scales has become a core research topic in this field.

[0003] Currently, the industry generally adopts a multi-timescale prediction error compensation strategy to manage energy storage devices. The typical approach is to first acquire predicted wind and solar power data and actual operating data, and obtain a comprehensive error signal by subtracting them. Then, using standard signal processing algorithms such as fixed bandpass filter banks or wavelet transforms, this comprehensive error signal is decomposed into multiple fixed-period frequency band components, including high-frequency, mid-frequency, and low-frequency components. Finally, the power compensation tasks for these different frequency bands are directly distributed to various energy storage units within the station (e.g., supercapacitors handle high frequencies, lithium batteries handle mid- and low-frequency frequencies) according to a pre-set ratio. During execution, a globally unified state of charge (SOC) balancing penalty term is introduced to extend the overall lifespan of the energy storage power station.

[0004] However, the aforementioned conventional control schemes have gradually revealed their limitations in practical engineering applications. On the one hand, the traditional error decomposition process and hardware resource configuration are often decoupled, resulting in a mismatch between the decomposed frequency band task load and the physical carrying capacity of the available energy storage in that frequency band. On the other hand, existing methods typically use a one-size-fits-all static parameter configuration when planning the safe operating boundary and state equilibrium target of the equipment, which is difficult to adapt to the dynamic evolution of system operating risks under different operating conditions, and is prone to mutual interference of control commands when energy storage units with different physical characteristics perform diversified tasks.

[0005] Therefore, there is an urgent need to study a collaborative management framework that can deeply integrate the physical operation status of energy storage with multi-scale adjustment needs, so as to couple the multi-timescale signal decomposition with the physical characteristics of heterogeneous energy storage hardware, thereby improving the overall utilization efficiency and operational safety of energy storage power stations under complex and fluctuating operating conditions. Summary of the Invention

[0006] Purpose of the invention: To provide a multi-timescale energy storage coordination control method for pure new energy collection stations, in order to solve the above-mentioned problems in the prior art.

[0007] Technical solution: A multi-timescale coordinated control method for pure renewable energy collection stations, comprising:

[0008] Acquire the error sequence of new energy output forecast, forecast uncertainty indicators, and collect the operating status data of energy storage units in the station;

[0009] The frequency band boundary frequency is determined based on the error power spectral density distribution of the new energy power output prediction error sequence, and the new energy power output prediction error sequence is decomposed based on the frequency band boundary frequency to extract multi-time scale error component sequences.

[0010] Based on the operational status data, the functional layers of each energy storage unit are configured, and the reserved state of charge margin of each layer of energy storage unit is determined in combination with the prediction uncertainty index. Based on the reserved state of charge margin, the adjustable power range of each energy storage unit is calculated.

[0011] Based on the operational status data, the marginal aging cost of each energy storage unit is determined. With the goal of reducing the marginal aging cost and achieving a balanced state of charge of each layer of energy storage units, under the constraint of the adjustable power range, the power of adjustment for the multi-timescale error component sequence is allocated to each energy storage unit, and the power allocation command is output.

[0012] Beneficial effects: First, addressing the mismatch between workload and physical capacity caused by the decoupling of traditional error signal decomposition and hardware configuration, this solution introduces a frequency reciprocal weighting mechanism to construct an equivalent energy throughput spectrum, and uses the actual available energy storage capacity ratio to look up and match the frequency band boundary frequency. This aligns the workload of the decomposed frequency band with the physical hardware capabilities of each layer, avoiding the phenomenon of some energy storage layers being frequently overloaded while other layers remain idle.

[0013] Secondly, to address the issue that static parameter configurations are difficult to adapt to the dynamic evolution of system operational risks, this solution establishes a dynamic mapping between state of charge deviation and time-varying confidence level, and calculates the reserved margin by combining time-sharing unidirectional error integration. This mechanism can automatically reduce the confidence level to release regulation potential when the energy storage capacity is sufficient, and automatically increase the confidence level to enhance safety defense when approaching the physical boundary, preventing the underlying equipment from experiencing over-limit cut-off failures due to over-response.

[0014] Finally, addressing the issue of control command interference arising from energy storage units with different physical characteristics performing diverse tasks, this solution abandons a globally unified equilibrium benchmark. Instead, it sets differentiated equilibrium targets and weights for different functional layers, pointing towards the mean or intraday time-varying trends, and uses a cross-layer relaxation overflow penalty mechanism for optimization. This approach eliminates the mutual constraints between power-type and energy-type equipment, ensuring the symmetry of the bidirectional charging and discharging capabilities of the fast response unit and enabling the energy transfer unit to accurately track intraday power surplus and deficit trends, thereby improving the overall dispatch response performance of the energy storage power station. Attached Figure Description

[0015] Figure 1 This is a flowchart of the multi-timescale energy storage coordinated control method of the present invention.

[0016] Figure 2 This is a flowchart of the present invention for obtaining the error sequence of new energy output prediction.

[0017] Figure 3 This is a flowchart illustrating the process of determining the frequency band boundary in this invention.

[0018] Figure 4 This is a flowchart of the present invention for solving the frequency band boundary frequency. Detailed Implementation

[0019] Example 1 describes the basic control framework for a pure new energy collection station. In one possible embodiment, it mainly includes the following steps:

[0020] like Figure 1 As shown, in step 101, the new energy output prediction error sequence, prediction uncertainty index, and the operating status data of the energy storage unit in the collection station are obtained.

[0021] In this step, the system acts as the data input terminal for the entire control architecture, centrally collecting the basic data required for control. Among these, the renewable energy output prediction error sequence refers to the time-series data of the deviation between predicted and actual values. Prediction uncertainty indicators are typically expressed as the upper and lower bounds of the prediction interval or statistical parameters of the probability distribution. Operating status data covers the physical parameters and real-time operating conditions of the energy storage equipment. This step provides complete data support for subsequent spectrum decomposition and power allocation.

[0022] Furthermore, in order to resolve the logical conflicts caused by the overlap of prediction data across multiple time scales, the specific process of obtaining the prediction error sequence of new energy power output can be achieved through the following steps.

[0023] Forecast uncertainty indicators can be expressed in the form of forecast standard deviation, forecast interval width, and confidence interval half-width. When using the forecast interval form, the forecast standard deviation can be estimated by dividing the symmetrical interval half-width by the corresponding confidence level quantile, and then used for estimating the maximum error magnitude in subsequent steps. Those skilled in the art can obtain the required forecast standard deviation parameter using conventional statistical transformation methods based on the output format of the actual forecasting system.

[0024] like Figure 2 As shown, in step 102, the power output prediction data and corresponding actual power output data of new energy at each time scale are obtained, and the prediction lead time corresponding to the power output prediction data at each time scale is determined.

[0025] Specifically, the energy management system reads output forecast data from the collection station, encompassing three time scales: ultra-short-term, short-term, and intraday. For example, the forecast lead time for ultra-short-term forecasts can be 15 minutes, while the forecast lead time for short-term forecasts can be several hours. Specifying the time interval is crucial for subsequently selecting the optimal data source with the highest accuracy.

[0026] Step 103: Prioritize the selection of ultra-short-term power output forecast data with the shortest forecast lead time as the target forecast data.

[0027] Under normal communication and operating conditions, since the shorter the prediction interval, the higher the prediction accuracy is usually, the system defaults to using ultra-short-term prediction data at the 15-minute level as the preferred benchmark.

[0028] Step 104: When it is found that the ultra-short-term power output forecast data is missing due to communication delay or failure to reach the update cycle, short-term power output forecast data or intraday power output forecast data are selected in order of forecast lead time from shortest to longest to replace the missing data, so as to ensure the continuity of the data stream when constructing the initial forecast error sequence.

[0029] This is a highly reliable fault-tolerant degradation replacement mechanism. In harsh industrial environments, communication links may experience brief failures. When ultra-short-term data is unavailable, the system automatically falls back to using short-term forecast data. If short-term data is also missing, intraday forecast data is used. This mechanism avoids control interruptions caused by the failure of a single data source, improving the robustness of the entire coordinated control architecture.

[0030] Step 105: Select the power output prediction data corresponding to the shortest available prediction lead time as the target prediction data, and calculate the difference between the target prediction data and the actual power output data to obtain the initial prediction error sequence.

[0031] In this embodiment, after completing the above fault tolerance and degradation judgment, the system performs a mathematical difference operation between the finally selected optimal prediction value and the corresponding actual operating physical quantity to generate an original deviation sequence containing multi-scale mixed wave characteristics.

[0032] Step 106: Perform anomaly removal and missing data imputation on the initial prediction error sequence to obtain the new energy power output prediction error sequence.

[0033] The system uses known data preprocessing algorithms such as local weighted regression to clean the sequence obtained by the aforementioned difference calculation, remove abnormal jump points caused by equipment maintenance, and finally output a continuous and high-quality error sequence to prevent bad data from polluting the subsequent spectrum analysis.

[0034] Step 107: Determine the frequency band boundary frequency based on the error power spectral density distribution of the new energy power output prediction error sequence, and decompose the new energy power output prediction error sequence based on the frequency band boundary frequency to extract the multi-time scale error component sequence.

[0035] This step is a core component of error spectrum decomposition and resource matching. The system first uses well-known algorithms such as the Fast Fourier Transform to perform time-frequency conversion on the pre-cleaned error sequence to obtain the error power spectral density distribution. Then, based on a predefined strategy, the boundary point is calculated, and the original error signal is separated into components with different fluctuation characteristics.

[0036] In some embodiments, the frequency band boundary frequency is determined in conjunction with the rated capacity distribution.

[0037] Optionally, the deep optimization logic for determining the band boundary frequency will be detailed in subsequent embodiments. This embodiment focuses on illustrating the specific process of physical decomposition based on the band boundary frequency, including the following steps.

[0038] Step 108: Configure a multi-channel filter bank with the frequency band boundary frequency as the cutoff frequency. The multi-channel filter bank includes a high-pass filter channel, a band-pass filter channel, and a low-pass filter channel.

[0039] The system constructs three parallel digital filtering channels based on the calculated boundary frequencies. For example, the high-pass channel is used to extract fast fluctuations with shorter periods, the band-pass channel is used to extract fluctuations with medium periods, and the low-pass channel isolates slow fluctuation trends with longer periods.

[0040] Step 109: Configure the multi-channel filter bank as a Butterworth filter architecture to utilize the maximum flat amplitude-frequency response characteristic of the Butterworth filter in the passband for signal separation.

[0041] In this embodiment, the multi-channel filter bank can be implemented using a fourth-order Butterworth filter. The motivation for choosing this architecture is that it has an extremely smooth amplitude-frequency response curve in its passband, which can minimize amplitude distortion during engineering signal separation, thereby ensuring the energy fidelity of error components at each time scale.

[0042] Step 110, and the lower cutoff frequency of the high-pass filter channel, the upper and lower boundary cutoff frequencies of the band-pass filter channel, and the upper cutoff frequency of the low-pass filter channel are all strictly set according to the corresponding frequency band boundary frequencies.

[0043] This parameter configuration step ensures that the physical cutoff boundary of the filter is aligned with the boundary point calculated based on the previous spectrum analysis, eliminating the parameter disconnect between the algorithm layer and the underlying digital signal processing execution layer.

[0044] Step 111: Use a multi-channel filter bank to perform zero-phase bidirectional filtering on the new energy output prediction error sequence to eliminate the phase delay introduced by the filtering and separate the error components with different fluctuation periods to obtain a multi-time scale error component sequence.

[0045] The system performs two filtering operations along the time axis, one forward and one reverse. The principle behind this is that bidirectional processing automatically cancels out the physical phase shift inherent in traditional causal filters. The objective technical effect of this zero-phase processing mechanism is that it ensures that the three decomposed error components can be directly superimposed in the time domain to reconstruct the original error signal, providing a rigorous equation balance basis for subsequent optimization solutions.

[0046] Step 112: Configure the functions of each energy storage unit in layers according to the operating status data, determine the reserved state of charge margin of each layer of energy storage unit in combination with the prediction uncertainty index, and calculate the adjustable power range of each energy storage unit based on the reserved state of charge margin.

[0047] This step establishes a bridge between the energy storage functional hierarchy and system regulation requirements. The system assigns roles based on the inherent properties of the hardware, then locks in a portion of the safe capacity based on risk prediction indicators, and finally converts it into the physical power boundary that can be output or absorbed at the current moment.

[0048] In some specific implementations, the process of functional layer configuration and interval calculation includes the following steps.

[0049] Step 113: Extract the rated power and rated capacity of each energy storage unit based on the operating status data, and calculate the corresponding power-energy ratio index.

[0050] The system calculates the power-energy ratio by dividing the rated power by the rated capacity. This ratio objectively reflects the charge and discharge rate characteristics of the energy storage device (energy storage unit), and a larger ratio indicates that it is more suitable for short-term high-power output.

[0051] Step 114: Based on the numerical distribution of the power-energy ratio of each energy storage unit, classify each energy storage unit into power-type energy storage for high-frequency error compensation, energy-type energy storage for medium- and low-frequency error mitigation, and hybrid energy storage that falls between the two, thus completing the functional stratification configuration. Among them, hybrid energy storage is used for medium-frequency error regulation.

[0052] The system sets preset judgment intervals to assign physical roles. For example, devices with a power-to-energy ratio greater than a first threshold, such as 0.5, can be classified as power-type energy storage, devices with a power-to-energy ratio less than a second threshold, such as 0.25, can be classified as energy-type energy storage, and those between the two thresholds are classified as hybrid energy storage. This matches the hardware physical characteristics with the frequency band of the error to be compensated.

[0053] Step 115: Obtain the physical boundaries that allow each energy storage unit to operate based on the operating status data. Then, extract the current state of charge and rated power of each energy storage unit from the operating status data.

[0054] Physical boundaries typically include the maximum and minimum permissible states of charge (SOCs) set by the battery management system (BMS). The system extracts these absolute boundary values ​​to prevent destructive overcharging or over-discharging.

[0055] Step 116: Calculate the remaining regulation capacity of each energy storage unit in the charging and discharging directions, taking into account the current state of charge, physical boundaries, and allocated reserved state of charge margin.

[0056] The system determines the remaining physical-level regulation capacity by calculating the difference between the absolute usable range and the reserved margin space. Furthermore, a reference regulation range parameter can be introduced, defined as the difference between the maximum allowable state of charge and the minimum allowable state of charge. The system obtains a dimensionless coefficient between 0 and 1, reflecting the current usable depth of the battery, by calculating the ratio of the current state parameter to this reference regulation range.

[0057] Step 117: Determine the corresponding bidirectional adjustable power range based on the remaining regulation capacity and the rated power of the corresponding energy storage unit.

[0058] Multiplying the dimensionless coefficients obtained from the aforementioned calculations by the rated power yields a bidirectional adjustable power range that includes both the upper limit of charging power and the lower limit of discharging power. This range directly defines the search space of the subsequent optimization solver, ensuring the security of system commands.

[0059] Step 118: Based on the operating status data, determine the marginal aging cost of each energy storage unit. With the goal of reducing the marginal aging cost and achieving a balanced state of charge of each layer of energy storage units, under the constraint of the adjustable power range, allocate the adjustment power for the multi-timescale error component sequence to each energy storage unit and output the power allocation command.

[0060] This is the final execution step of this basic control architecture. The system establishes a comprehensive optimization objective that includes equipment aging costs and battery state-of-charge (SOC) balancing deviations, and searches for the optimal solution within the power range calculated in the previous steps. The final generated power allocation command will be directly sent to the underlying energy storage converter for execution. Specific algorithm optimization details regarding lifetime loss calculation and differentiated balancing will be elaborated upon in subsequent embodiments.

[0061] Example 2 describes the core steps of the error spectrum adaptive decomposition mechanism, as a further refinement of the step of determining the frequency band boundary in Example 1.

[0062] like Figure 3 As shown, in step 201, the power output prediction error sequence of new energy sources is subjected to spectral transformation to obtain the error power spectral density distribution.

[0063] After obtaining the prediction error sequence with temporal characteristics, the time-domain signal is converted into a frequency-domain signal using a Fast Fourier Transform. By calculating the power amplitude of each frequency component, a density curve reflecting the energy distribution of the error signal with frequency is obtained, which serves as the reference physical quantity for subsequent frequency band division.

[0064] Step 202: Analyze the error power spectral density distribution according to the set frequency band division strategy, and solve for the frequency band boundary frequency. The frequency band boundary frequency is used to divide the new energy output prediction error sequence into high frequency, medium frequency and low frequency bands.

[0065] This step defines the core logic for dividing the continuous spectrum into discrete frequency bands. Because the output fluctuation characteristics of pure renewable energy collection stations are highly time-varying due to meteorological conditions, fixed frequency band division parameters cannot adapt to real-world operating conditions. Therefore, a specific calculation strategy is configured to dynamically optimize the boundary frequencies. Specifically, there are basic implementation methods based on the morphological characteristics of the error signal itself, and higher-order optimization implementation methods based on matching system energy storage resources.

[0066] In some specific implementations, the basic frequency band division strategy based on the morphological characteristics of the error signal itself includes the following steps.

[0067] Step 203: Extract the morphological features of the error power spectral density distribution, and calculate the spectral centroid index and spectral bandwidth index based on the morphological features.

[0068] Feature extraction of the power spectral density curve was performed using statistical algorithms. The spectral centroid index reflects the concentration of energy in the error signal spectrum, while the spectral bandwidth index reflects the degree of dispersion of energy in the frequency domain.

[0069] Step 204: Using the preset center reference frequency as a benchmark, and combining the spectral centroid index and spectral bandwidth index, determine the first boundary point between high frequency and mid frequency, and the second boundary point between mid frequency and low frequency, thereby obtaining the frequency band boundary frequency.

[0070] Using the morphological indices obtained from the above calculations, combined with pre-calibrated empirical reference center points, two cutoff frequencies are output. The objective technical advantage of this alternative scheme lies in its low computational complexity. However, its applicability is limited. When stable wind conditions cause the prediction error spectrum to exhibit a single-peak dominant characteristic, the spectral centroid can reasonably divide the frequency bands; but when cloudy weather causes photovoltaic power output to exhibit multi-peak fluctuation characteristics, the spectral centroid parameter will lose its physical guiding significance.

[0071] Based on the above embodiments, in order to solve the technical problem that the traditional frequency band allocation results do not match the actual physical resources of energy storage power stations, another preferred implementation method based on the proportion of energy storage resources is provided, which specifically includes the following steps.

[0072] like Figure 4 As shown, in step 205, a frequency reciprocal weighting mechanism is introduced to calculate the error power spectral density distribution, obtain the equivalent energy throughput spectral density distribution, and construct the corresponding normalized cumulative energy throughput function along the frequency axis.

[0073] Conventional signal decomposition algorithms only divide the signal based on power amplitude, ignoring the capacity consumption characteristics of energy storage devices. While low-frequency fluctuations at the physical level have low power amplitudes, their unidirectional duration is extremely long, and their actual consumption of charging and discharging capacity is far greater than that of high-frequency components. Therefore, a frequency reciprocal weight is introduced to scale and reconstruct the power spectrum.

[0074] S E (f k )=(abs(E(f k )) 2 ) / f k ;

[0075] Among them, S E (f k ) represents the equivalent energy throughput spectral density corresponding to the k-th discrete frequency point, abs(E(f) k )) represents the frequency domain power amplitude, f k This represents a discrete frequency value. By dividing by the frequency parameter, the abstract power spectral density is transformed into an equivalent physical quantity that reflects the actual throughput capacity required for energy storage.

[0076] Based on this, numerical integration is performed point-by-point along the frequency axis and divided by the total global area to generate a monotonically increasing cumulative function, described by the following formula:

[0077] R E (f k )=∑1 k (S E (fj)) / ∑1 N (S E (f j ));

[0078] Among them, R E (f k () is the normalized cumulative energy throughput function at frequency f k The value at ∑1 ranges from 0 to 1. k This represents the local accumulation from the first frequency point to the kth frequency point, ∑1 N This indicates a global accumulation of a total of N discrete frequency points.

[0079] Step 206: Based on the pre-stored rated capacity configuration information of each energy storage unit and its functional layer affiliation, i.e. rated capacity distribution, determine the proportion of energy storage capacity corresponding to different functional layers.

[0080] Extract the rated capacity physical parameters of all energy storage units configured as high-frequency, medium-frequency, and low-frequency layers in the aforementioned steps, sum them by layer, and then calculate the ratio of the available capacity of each layer to the total energy storage capacity of the collection station to obtain the resource allocation ratio for the corresponding frequency band regulation tasks.

[0081] In some embodiments, the energy storage units are configured with functional layers based on the operating status data to determine the rated capacity distribution of the energy storage units in each functional layer.

[0082] Step 207: Using the proportion of energy storage capacity as the matching target, reversely query the corresponding frequency node on the normalized cumulative energy throughput function, and determine the frequency node as the initial boundary frequency.

[0083] After obtaining the resource ratio and the monotonically increasing cumulative function, an inverse function mapping solution is performed. For example, assume the extracted low-frequency layer energy storage capacity ratio is 0.5 and the mid-frequency layer capacity ratio is 0.3. The algorithm first locates the position with a vertical coordinate of 0.5 on the normalized cumulative curve, and then uses linear interpolation to look up the corresponding horizontal coordinate frequency value, directly using it as the initial boundary frequency between the mid-frequency and low-frequency layers. Subsequently, 0.5 + 0.3 = 0.8 is used to locate the position with a vertical coordinate of 0.8 on the same curve, and the corresponding horizontal coordinate frequency value is looked up, using it as the initial boundary frequency between the high-frequency and mid-frequency layers. This operation delegates the determination of the signal decomposition boundary to the hardware physical configuration, ensuring an absolute closed-loop match between the adjustment task amount decomposed from each frequency band and the corresponding available capacity of the layer.

[0084] In this application, instead of decomposing the signal first and then matching the hardware, the hardware capacity ratio is read first, and then the signal decomposition boundary is determined in reverse.

[0085] Step 208: Combine the preset energy storage physical response time boundary to truncate the initial boundary frequency, and after performing timing smoothing, output the updated frequency band boundary frequency.

[0086] To prevent the boundary frequencies calculated by purely mathematical mapping from exceeding the physical tracking capabilities of the battery and converter, the boundary constraint range for the cycle transition between low and medium frequencies is set to be 2 to 8 hours, and the boundary constraint range for the cycle transition between medium and high frequencies is set to be 5 to 30 minutes. A limiting function is used to forcibly truncate the initial calculated values ​​within the aforementioned safe physical boundaries.

[0087] Furthermore, to avoid sudden changes in the boundary frequency caused by meteorological changes and resulting in discontinuity in power commands, an exponentially weighted moving average algorithm is used for disturbance rejection filtering.

[0088] ;

[0089] Among them, f ML (t) represents the mid-frequency and low-frequency boundary frequency output in the current update cycle, β is the smoothing weight coefficient between 0 and 1, and f ML_computed (t) represents the initial limit frequency calculated after truncation constraint processing at the current time, f ML (t-dt) represents the historical output frequency of the system in the previous update cycle. The parameter β is preferably in the range of 0.2 to 0.4. The smaller the value, the stronger the ability to suppress abrupt changes. This results in smooth decomposition boundary parameters for updating the multi-channel filter bank.

[0090] In some embodiments, the boundary frequency f between high frequency and intermediate frequency HM (t) Smoothing is performed using the same exponentially weighted moving average method as in the above formula, and the smoothing weight coefficient can be the same as f.ML (t) takes the same value.

[0091] Example 3 describes the calculation of the reserved margin by combining operating status feedback and forward-looking energy integration trend, as a further optimization and refinement of the step of determining the reserved state of charge margin in the previous examples.

[0092] Step 301: Extract the current state of charge of each energy storage unit based on the operating status data, and calculate the deviation index of the current state of charge relative to the midpoint of the available operating range.

[0093] In conventional control logic, the reserved adjustment margin often uses fixed parameters without considering the real-time available capacity of the underlying physical devices. This embodiment introduces a state feedback mechanism to extract the current remaining power status of the energy storage unit. To quantify the urgency of this status, a normalized deviation evaluation system is constructed.

[0094] d m (t)=abs(SOC m (t)-SOC mid ) / ((SOC max -SOC min ) / 2);

[0095] Where, d m (t) represents the state of charge deviation index of the target energy storage unit m at the current time t, SOC. m (t) represents the current state of charge (SOC) of the target energy storage unit m. max With SOC min These are the upper and lower limits of the physical operating state of charge (SOC) set for the battery management system. mid The arithmetic mean of the upper and lower limits of physical operation can be used as the midpoint of the operating interval. abs is the absolute value calculation function. This index is a non-negative value greater than or equal to 0. When the state of charge is at the absolute midpoint, the index takes the value of 0. When the state of charge approaches the physical boundary, the index tends to be 1 or greater.

[0096] Step 302: Map the deviation index to a time-varying confidence level, so that the time-varying confidence level automatically increases when the current state of charge is close to the physical boundary and automatically decreases when it is in the middle region.

[0097] After obtaining the deviation index, it is deeply coupled with statistical risk control parameters. In error compensation control, the confidence level determines the conservatism with which the system mitigates forecast uncertainty. A higher confidence level generates a larger margin of safety, thereby compressing the currently adjustable power range.

[0098] ;

[0099] Among them, z α,m (t) represents the time-varying confidence level parameter of the dynamic mapping, z α min For the preset minimum confidence level, z α max The preset maximum confidence level is defined by `min`, which is a limiting function that takes the minimum value to ensure that the mapping process operates within a linear range. For example, the minimum confidence level can be set to 1.28 to correspond to a 90% statistical confidence level, and the maximum confidence level can be set to 2.33 to correspond to a 99% statistical confidence level. This mapping relationship ensures a positive correlation between the time-varying confidence level and the deviation from the state of charge, thereby releasing regulation potential when energy storage capacity is sufficient and enhancing safety defenses when capacity is tight.

[0100] Step 303: Combine the time-varying confidence level and the prediction uncertainty index to estimate the maximum error amplitude under multiple time scales, and calculate the corresponding reserved state-of-charge margin based on the time-division unidirectional cumulative amount of the maximum error amplitude.

[0101] Using the time-varying parameters generated in the preceding steps, perform an estimation operation for the maximum error amplitude.

[0102] ;

[0103] in, e is the estimate of the maximum error magnitude at a specific time scale k at a future time τ. k (t+τ) represents the corresponding mean prediction error, σ pred_k (t+τ) is the standard deviation parameter representing the forecast uncertainty index.

[0104] The mean of the prediction error can be taken as the moving average of the error components at the corresponding time scale k within the most recent statistical window, or as zero when there is no significant systematic bias in the prediction system.

[0105] After obtaining the error amplitude boundary, in order to prevent positive and negative errors from canceling each other out during time integration calculation and thus producing erroneous margin conclusions, a time-division unidirectional accumulation mechanism is strictly introduced to calculate the state-of-charge margin.

[0106] Calculations were performed separately for each energy storage unit:

[0107] ;

[0108] in, The calculated upper limit margin for the reserved state of charge is defined by max, which is the maximum value cutoff function. This is the rated capacity of the energy storage unit m. This is the estimate of the maximum error amplitude of energy storage unit m at time scale k. k is the time scale index, with values ​​corresponding to the high-frequency, mid-frequency, and low-frequency bands. By limiting the integral term to be greater than or equal to 0 through the maximum value function, this upper limit margin only accumulates the positive portion of the prediction error. Its physical meaning is to specifically reserve charging space for absorbing the portion of power generation exceeding the prediction. Similarly, for the lower limit margin in the discharge direction, the system only accumulates the absolute value of the error amplitude in the negative range.

[0109] Among them, T m The future response time window length corresponds to the energy storage unit m, and its value matches the frequency band characteristic period that the functional layer containing the unit is responsible for compensating. For power-type energy storage, T m The maximum half-cycle corresponding to the high-frequency band can be taken; for energy storage, T m The maximum half-cycle corresponding to the low-frequency band can be taken. For example, T. m = 1 / (2 ×f b ), where f b This refers to the frequency band boundary frequency corresponding to this functional layer. Those skilled in the art can calculate and determine the corresponding window length based on the actual frequency band boundary frequency.

[0110] Based on the above embodiments, for energy storage units classified as energy-type energy storage, relying solely on passive error compensation margin can easily lead to power depletion or overflow. Therefore, in some optional embodiments, an implementation mechanism including forward-looking energy correction logic is further provided, specifically including the following steps.

[0111] Step 304: Obtain the daily output forecast curve and corresponding transmission plan of the collection station in real time, and assess the expected power surplus / deficit status.

[0112] The energy management module acquires data curves over a long period after the current moment in parallel via a communication interface. It then calculates the difference between the projected total power generation sequence of the power plant and the transmission limit sequence issued by the grid dispatch center. A positive surplus is identified when the projected power generation exceeds the transmission limit, and a negative shortage is identified otherwise.

[0113] Step 305: Calculate the intraday trend correction based on the energy integral within the future rolling window according to the power surplus / deficit status.

[0114] Extract the aforementioned power difference and perform energy conversion calculations by product and summation along the time axis.

[0115] ;

[0116] in, This is the intraday trend correction amount output. This indicates the calculation of the time-domain integral within the set future rolling window. The total output prediction data represents the power value at time τ. In order to deliver the specified power value planned at time τ, The rated capacity of a single device. This represents the total number of devices configured as energy storage within the system. The negative constant multiplier at the front end of the formula has a clear physical control orientation. When a positive power generation surplus is predicted in the future, the current output is a negative correction amount, guiding the batteries to discharge early to create absorption space. To prevent over-adjustment, the value range of this intraday trend correction amount can be forcibly truncated between -0.15 and 0.15.

[0117] Step 306: The preset benchmark target value is superimposed with the intraday trend correction amount to obtain the dynamically corrected state of charge target value. The reserved state of charge margin of the corresponding energy storage unit is adjusted using the dynamically corrected state of charge target value.

[0118] After obtaining the trend correction amount, it is superimposed onto the set regular median position.

[0119] ;

[0120] in, The target value for the dynamically corrected state of charge of the energy storage unit m is... The preset benchmark target value is typically set to 0.5. The system re-anchors the reserved state-of-charge margin boundary of the energy storage unit based on the superimposed benchmark target value. This achieves decoupling and coordination between long-term energy transfer tasks and short-term random error compensation tasks at the physical capacity level.

[0121] By establishing a linear mapping between the adaptive confidence level and the SOC deviation, and then calculating the capacity margin through one-way integration, a closed loop is formed that feeds back the real-time physical state of energy storage (SOC deviation) to probabilistic risk assessment parameters (confidence level).

[0122] In some examples, for energy storage units, when calculating the state of charge deviation in step 301, the midpoint of the available operating range (SOC) is used. mid Replace it with a dynamically corrected target value for the state of charge, so that the deviation index reflects the distance of the current state of charge relative to the time-varying target.

[0123] Therefore, when the dynamic target value guides the battery to deviate in a certain direction, the deviation index is automatically adjusted to avoid triggering unnecessary high confidence level protection due to the target value deviation.

[0124] Example 4 describes a comprehensive power allocation process that includes battery aging model construction, hierarchical differentiated equalization target setting, and cross-layer relaxation constraint optimization, as a complete refinement of the optimal power distribution stage in the basic control architecture.

[0125] Step 401: Obtain the benchmark cycle life curve provided by the energy storage battery manufacturer, and perform parameter fitting on the life decay characteristics of the benchmark cycle life curve to obtain the battery characteristic parameters.

[0126] Before performing real-time online power allocation, a mathematical model reflecting the actual degradation characteristics of the physical devices must be constructed offline. The system reads the factory test data of the corresponding batch of cells offline and uses a numerical fitting algorithm to extract constants that characterize the mapping relationship between the depth of discharge and the number of cycles.

[0127] Specifically, this mapping relationship follows a power-law lifetime model and is described by the following formula:

[0128] N life =a·DOD -b ;

[0129] Where, N Life Here, DOD represents the expected total number of cycles at a specific depth of discharge, and a and b are battery characteristic parameters obtained through pre-fitting offline. For example, for a typical lithium iron phosphate battery, parameter a ranges from 5000 to 8000, and parameter b ranges from 1.5 to 2.2.

[0130] Step 402: Obtain the cumulative charge-discharge cycle count and battery health of each energy storage unit based on the operating status data.

[0131] The system reads basic ledger data from the battery management system in real time. This data forms an important prerequisite for assessing the current degradation baseline of each energy storage unit.

[0132] Step 403: Extract the depth of discharge from the historical operating time series of the energy storage unit recorded in the operating status data using the rainflow counting method.

[0133] Because the state-of-charge (POC) time-series curve in actual operation exhibits irregular and random fluctuations, a standard full-charge-discharge lifetime model cannot be directly applied. The system employs the standard rainflow counting method to pair and convert the peaks and troughs of the POC within the previous historical control cycle, effectively transforming the irregular charge-discharge trajectory into a series of standard half-cycle operations, and then outputting the corresponding equivalent depth of discharge value.

[0134] Specifically, the SOC time series of each energy storage unit within the historical control cycle is used as the input signal. The SOC value is used as the counting variable to replace the stress variable in traditional rainflow counting. Half-cycle pairing is performed based on adjacent extreme points (peaks and troughs). For each pair of half-cycles, the amplitude difference is calculated as the equivalent depth of discharge (DOD). halfFor a complete cycle (half-charge, half-discharge pairing), the average of the amplitudes of the two half-cycles is taken as the equivalent DOD. The system calculates the DOD for the most recent statistical window (e.g., the first N). w The weighted average of all equivalent cycles within a control cycle is used to obtain the DOD (Directional Distance) representing the current operational intensity. m .

[0135] For example, a weighted average can be calculated for all equivalent cycles within the most recent statistical window (such as the previous few control periods) according to the damage contribution weight of each cycle, where the weight of each cycle is the normalized damage amount of that cycle under the lifetime model. This yields the DOD representing the current operational intensity. m .

[0136] Step 404: Based on the nonlinear equivalent loss relationship between the depth of discharge and the cycle life of the energy storage battery, and combined with the pre-stored battery replacement cost and battery health, the corresponding marginal aging cost is calculated.

[0137] After obtaining the equivalent depth of discharge (DOD), the system calculates the small lifetime loss introduced by a single action based on the pre-built power-law lifetime model and converts it into an economic marginal cost. When the DOD does not change drastically, the damage rate of the current action can be approximated by the derivative of the historical average operating conditions. The specific calculation is as follows:

[0138] ;

[0139] in, Let m be the marginal aging cost of the target energy storage unit at the current moment. Let a be the total expected replacement cost of the entire energy storage unit m. m With b m For the battery characteristic parameters corresponding to this unit, DOD m The equivalent depth of discharge for the most recent operating cycle is obtained by rainflow counting. This refers to the rated capacity of the unit. The introduction of this physical parameter causes devices with lower health levels to automatically have a higher call cost weight, thus reducing their usage by the system in subsequent optimized allocations.

[0140] In some embodiments, when calculating marginal aging costs, the system adjusts the rated capacity based on the current battery health of the energy storage unit, using the current effective capacity. Replacement rated capacity The battery's health is factored into the calculation. Because declining battery health leads to a decrease in effective capacity, the same power output corresponds to a greater equivalent depth of discharge. This results in devices with lower health automatically exhibiting higher marginal aging costs in the calculation. The cumulative charge-discharge cycle count is used to help verify the reasonableness of the battery health estimate.

[0141] Step 405: For energy storage units classified as power-type energy storage in the functional layer configuration, the average state of charge of the energy storage units in the corresponding functional layer is used as the state of charge balancing target, and a higher balancing weight is configured.

[0142] Traditional strategies often employ a globally unified equalization benchmark, which can lead to mutual interference between control commands. The system manages power-type devices tasked with high-frequency fluctuation compensation independently. Since these devices frequently handle bidirectional power throughput, their ideal physical state is to remain at the midpoint of their capacity range. Therefore, the system extracts the arithmetic mean state of charge within its functional layer as its pull-back benchmark and sets a relatively large weighting parameter, such as 1.5 to 2.0 times the basic equalization weight, to force them to maintain symmetrical bidirectional adjustment capabilities at all times.

[0143] Step 406: For energy storage units classified as energy-type energy storage in the functional tiered configuration, the time-varying target value dynamically corrected based on the intraday power surplus / deficit trend is used as the state-of-charge (SOC) balancing target, and a lower balancing weight is configured. In other words, the time-varying target value dynamically corrected based on the real-time acquired intraday output forecast curve and the power surplus / deficit trend obtained from the transmission plan assessment is used as the SOC balancing target.

[0144] For equipment undertaking long-term energy transfer tasks, the system abandons the fixed mean target and instead uses a previously generated time-varying target value containing forward-looking energy integral information. Simultaneously, a small weighting parameter is set, for example, 0.5 to 0.8 times the basic equilibrium weight. This configuration grants energy-intensive equipment the freedom to move freely within a wider capacity range, preventing stringent static equilibrium commands from interfering with its core task of smoothing out slow fluctuations.

[0145] Step 407: For energy storage units classified as hybrid energy storage in the functional tiered configuration, the weighted average value of the average state of charge and the time-varying target value is used as the state of charge equilibrium target.

[0146] For hybrid devices with intermediate physical characteristics, the system uses a weighted average mathematical approach to generate a compromise target. Specifically, an allocation coefficient with a value range limited to 0.4 to 0.6 is introduced to linearly map and combine the average state of charge with the time-varying target value.

[0147] Step 408: Normalize the marginal aging cost and the deviation of the predicted state of charge (SOC) of each energy storage unit after power allocation based on operating status data from the SOC equilibrium target, and construct an optimization objective function that includes an aging cost term and a hierarchical equilibrium penalty term. Specifically, the predicted SOC of each energy storage unit at the end of the current control cycle is:

[0148] ;

[0149] in, For energy storage unit m, when receiving power distribution P m The predicted state of charge after that, P represents the actual state of charge at the current time t. m Let δt be the power allocation variable to be solved, and δt be the duration of the control cycle. This represents the rated capacity of energy storage unit m.

[0150] Based on this, the dimensionless optimization objective function is constructed as follows:

[0151] ;

[0152] in, The dimensionless optimization objective function is constructed, where M is the total number of energy storage units participating in the control. For marginal aging cost, |P m | represents the absolute value of the power allocation variable, C0 is the predefined normalized benchmark for aging cost in the system, i.e., the average aging rate, specifically the product of the average marginal aging cost coefficient and the average rated power of all energy storage units in the system; λ m This represents the equilibrium weighting coefficient for the corresponding energy storage unit m. To achieve the corresponding hierarchical state-of-charge equilibrium objective, A predefined normalized reference for the state of charge deviation of the system.

[0153] C0 is the predefined normalized benchmark for aging costs in the system, and its dimensions are the same as c. aging , m ×abs(P m Since the product of the two sub-objectives has the same dimensions, those skilled in the art can adaptively set the order of magnitude of the normalized objective function based on the system's rated power, average marginal aging cost, and control cycle duration, so that the two sub-objectives are balanced on the same numerical order of magnitude. m This is the equilibrium weighting coefficient for the corresponding energy storage unit m.

[0154] For units classified as power-type energy storage, λ m Take the basic equilibrium weight λ0 and the power layer weight multiple k P The product of; for units classified as energy storage, λ m Take λ0 and the energy layer weighting factor k E The product of; for hybrid energy storage, λ m Take the product of λ0 and the corresponding compromise factor; SOC m The current state of charge, To achieve the corresponding hierarchical state-of-charge equilibrium objective, A predefined normalized reference for the state of charge deviation of the system.

[0155] In an alternative implementation, the aging cost normalization benchmark C0 can be taken as the average aging cost of all energy storage units in the system operating at rated power for one control cycle, i.e. ; This is the average aging cost coefficient; This represents the average rated power. At this point, the molecule is... .

[0156] Normalized reference for state of charge deviation Typical operating values ​​of the standard deviation of the state of charge of each energy storage unit within the functional layer can be obtained, and those skilled in the art can make adaptive adjustments based on actual system operating data.

[0157] Step 409: Allow each functional layer to overflow its adjustment power to adjacent functional layers when its adjustment capacity is insufficient, and define the corresponding overflow power as a cross-layer relaxation variable.

[0158] Normally, high-frequency error commands are strictly limited to distribution within power-type devices. However, when the load is too large and touches the physical boundaries of the underlying devices, in order to ensure the absolute achievement of the total power target, the system opens up the hierarchical barriers in the mathematical model and sets auxiliary variables to accommodate residual control commands that cannot be digested within the current level.

[0159] Step 410: Add an overflow penalty term for cross-layer slack variables to the optimization objective function, and configure a penalty weight for the overflow penalty term with a value higher than that for the layer equilibrium penalty term, so that the cross-layer overflow power is automatically constrained by the optimization solver to be triggered only when the capacity of the functional layer is exhausted.

[0160] The system adds the cross-layer relaxation variable as a squared term to the dimensionless optimization objective function and assigns it an extremely high penalty multiplier, for example, setting it to 100 times the aforementioned equilibrium weight coefficient λ0. This mechanism ensures, from a purely mathematical constraint level, that the solution algorithm will try its best to suppress the generation of this overflow power variable when searching for the optimal solution, thereby promoting soft hierarchical control logic under hard equilibrium.

[0161] After introducing cross-layer relaxation variables, the complete optimization objective function is updated as follows:

[0162] J obj =J obj base +λ overflow ×∑ l=1 L (sL 2 );

[0163] Among them, J objFor the final complete optimization objective function used to solve, ∑ l=1 L This represents summing over all L functional layers, sL 2 Let λ be the cross-layer slack variable of the l-th layer. overflow The overflow penalty weight is set to 100 times the base equilibrium weight λ0 to ensure that the slack variable is activated only when the current layer capacity is exhausted.

[0164] In some embodiments, sL = sl / Pl, where sl is the cross-layer overflow power of the l-th layer and Pl is the sum of the rated power of all energy storage units within the l-th layer. In some other embodiments, other normalization methods may be used, such as dividing the dimensional parameter by its own reference value.

[0165] Step 411: Using the adjustable power range as the boundary limiting constraint and the total adjustment demand and the hierarchical demand with cross-layer relaxation variables as the power balance constraint, the optimization objective function is optimally solved.

[0166] In some embodiments, the adjustable power range is used as the boundary limiting constraint, the total error value corresponding to the multi-timescale error component sequence is used as the total adjustment requirement, and the hierarchical requirement after introducing cross-layer relaxation variables is used as the power balance constraint to solve the optimization objective function optimally. The cross-layer relaxation variables are the amount of adjustment power overflowing to the adjacent functional layer when the adjustment capacity of each functional layer is insufficient.

[0167] The system aggregates all boundary conditions generated in the preceding steps. The bidirectional adjustable power range of each device is configured as an inequality constraint variable, and the scheduling instructions issued by the central aggregation station are configured as a global equality constraint variable. The standard form derivation of the QP after the expansion of the SOC prediction term has been given in step 408. At the same time, local equality constraints for each independent frequency channel are configured in conjunction with cross-layer relaxation variables.

[0168] For the absolute value term c in the objective function aging,m ×abs(P m This can be achieved by introducing an auxiliary variable P. m + and P m - (The power corresponding to the charging and discharging directions is ≥0 respectively), let abs(P) m )=P m + +P m - And add linear constraint P m =P m + -P m -The absolute value term is equivalently transformed into a linear constraint, so that the overall optimization problem conforms to the standard form of quadratic programming (quadratic objective function + linear constraint).

[0169] Constraints may include two-way interval constraints on power allocation for each energy storage unit; global power balance constraints; and power balance constraints within each functional layer. Those skilled in the art can solve the above problems using conventional quadratic programming solvers (such as OSQP, Gurobi, etc.).

[0170] The system calls the quadratic programming algorithm program to perform multivariate optimization operations.

[0171] Step 412: Determine the optimal output power of each energy storage unit based on the solution results, and generate and issue power allocation instructions.

[0172] The algorithm outputs a specific power matrix that minimizes the dimensionless optimization objective function. The system extracts this power matrix and converts it into control messages recognizable by the underlying hardware communication protocol, sending them to the control panels of each energy storage converter at fixed intervals. For example, the rolling update cycle of this control loop can be set between 1 second and 1 minute, thereby achieving continuous coordinated operation of signals at different time scales.

[0173] In simulation verification of a typical pure wind-solar aggregation station scenario, compared with the traditional scheme of fixed frequency band allocation, the method of this invention reduces the SOC over-limit triggering frequency of power-type energy storage units, and allows for more thorough pre-discharge clearing operations of energy-type energy storage units under conditions of large daily power surpluses and deficits. The overall average aging cost objective function value of each functional layer unit decreases. In a typical simulation scenario (200MW wind power capacity, 50MWh energy storage configuration, including 10 power-type and 8 energy-type energy storage units), compared with the traditional scheme using fixed frequency band allocation, the method of this invention reduces the SOC over-limit triggering frequency of power-type energy storage units by approximately 40% to 60%, and increases the daily capacity utilization rate of energy-type energy storage units by approximately 15% to 25%. Those skilled in the art will understand that the specific performance improvement varies depending on the characteristics of new energy output, the scale of energy storage configuration, and operating conditions. The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A multi-timescale coordinated control method for energy storage in pure renewable energy collection stations, characterized in that, include: Acquire the error sequence of new energy output forecast, forecast uncertainty indicators, and collect the operating status data of energy storage units in the station; The frequency band boundary frequency is determined based on the error power spectral density distribution of the new energy power output prediction error sequence, and the new energy power output prediction error sequence is decomposed based on the frequency band boundary frequency to extract multi-time scale error component sequences. Based on the operational status data, the functional layers of each energy storage unit are configured, and the reserved state of charge margin of each layer of energy storage unit is determined in combination with the prediction uncertainty index. Based on the reserved state of charge margin, the adjustable power range of each energy storage unit is calculated. Based on the operational status data, the marginal aging cost of each energy storage unit is determined. With the goal of reducing the marginal aging cost and achieving a balanced state of charge of each layer of energy storage units, under the constraint of the adjustable power range, the power of adjustment for the multi-timescale error component sequence is allocated to each energy storage unit, and the power allocation command is output. Among them, determining the frequency band boundary frequency based on the error power spectral density distribution of the new energy output prediction error sequence includes: The error power spectral density distribution is obtained by performing a spectral transformation on the error sequence of new energy power output prediction. The error power spectral density distribution is analyzed according to the set frequency band division strategy, and the frequency band boundary frequency is obtained. The frequency band boundary frequency is used to divide the new energy output prediction error sequence into high frequency, medium frequency and low frequency bands. Based on the established frequency band division strategy, the error power spectral density distribution is analyzed to obtain the frequency band boundary frequency, specifically including: A frequency reciprocal weighting mechanism is introduced to calculate the error power spectral density distribution, obtain the equivalent energy throughput spectral density distribution, and construct the corresponding normalized cumulative energy throughput function along the frequency axis; Based on the pre-configured rated capacity distribution of each layer of energy storage units, determine the proportion of energy storage capacity corresponding to different functional layers; Using the proportion of energy storage capacity as the matching target, the corresponding frequency node is queried in reverse on the normalized cumulative energy throughput function, and the frequency node is determined as the initial boundary frequency. The initial boundary frequency is truncated and constrained by the preset energy storage physical response time boundary, and after performing timing smoothing, the updated frequency band boundary frequency is output. The reserved state-of-charge margin of each energy storage unit is determined by combining the prediction uncertainty index, specifically including: The current state of charge of each energy storage unit is extracted based on the operating status data, and the deviation index of the current state of charge relative to the midpoint of the available operating range is calculated. The deviation index is mapped to a time-varying confidence level, so that the time-varying confidence level automatically increases when the current state of charge is close to the physical boundary and automatically decreases when it is in the intermediate region. By combining time-varying confidence level and prediction uncertainty index, the maximum error magnitude under multiple time scales is estimated, and the corresponding reserved state margin of charge is calculated based on the time-division unidirectional cumulative amount of the maximum error magnitude.

2. The method according to claim 1, characterized in that, Obtain the error sequence for predicting new energy power output, including: Acquire the power output forecast data and corresponding actual power output data of new energy at various time scales, and determine the forecast lead time corresponding to the power output forecast data at each time scale. Select the power output prediction data corresponding to the shortest available prediction lead time as the target prediction data, and calculate the difference between the target prediction data and the actual power output data to obtain the initial prediction error sequence; The initial prediction error sequence is processed by anomaly removal and missing data imputation to obtain the new energy output prediction error sequence.

3. The method according to claim 1, characterized in that, The error sequence for predicting new energy output is decomposed based on the frequency band boundary, and multi-time-scale error component sequences are extracted, including: A multi-channel filter bank is configured with the frequency band boundary frequency as the cutoff frequency. The multi-channel filter bank includes a high-pass filter channel, a band-pass filter channel, and a low-pass filter channel. A multi-channel filter bank is used to perform zero-phase bidirectional filtering on the error sequence of new energy output prediction to eliminate the phase delay introduced by the filtering and separate the error components with different fluctuation periods to obtain a multi-time scale error component sequence.

4. The method according to claim 1, characterized in that, Based on operational status data, each energy storage unit is configured with functional hierarchical levels, including: The rated power and rated capacity of each energy storage unit are extracted based on the operating status data, and the corresponding power-energy ratio is calculated. Based on the numerical distribution of the power-energy ratio of each energy storage unit, the energy storage units are divided into power-type energy storage for high-frequency error compensation, energy-type energy storage for medium- and low-frequency error mitigation, and hybrid energy storage that falls between the two, thus completing the functional hierarchical configuration.

5. The method according to claim 1, characterized in that, In determining the reserved state-of-charge margin for each layer of energy storage units, for energy storage units classified as energy-type energy storage, the following also applies: Real-time acquisition of the daily output forecast curve of the collection station and the corresponding transmission plan to assess the expected power surplus or deficit status; The intraday trend correction is calculated based on the energy integral within the future rolling window according to the power surplus / deficit status. The preset benchmark target value is superimposed with the intraday trend correction to obtain the dynamically corrected state of charge target value. The reserved state of charge margin of the corresponding energy storage unit is adjusted using the dynamically corrected state of charge target value.

6. The method according to claim 1, characterized in that, The adjustable power range of each energy storage unit is calculated based on the reserved state-of-charge margin, including: The physical boundaries that allow each energy storage unit to operate are obtained based on the operational status data; Based on the current state of charge, physical boundaries, and allocated reserved state of charge margin, calculate the remaining regulation capacity of each energy storage unit in the charging and discharging directions. Based on the remaining regulation capacity and the rated power of the corresponding energy storage unit, the corresponding bidirectional adjustable power range is determined.

Images