A control method of an energy storage system applied to a transformer district

Abstract

The application discloses a kind of energy storage system control methods applied to transformer area, and the present application relates to the technical field of power system energy storage regulation, based on target transformer area historical operation data to construct benchmark load power fluctuation curve, using adaptive dynamic time warping distance density clustering algorithm to identify typical daily load mode, and generate smooth benchmark curve;According to the historical regulation frequency of energy storage system at each collection time point, the regulation fluctuation interval is divided, the double-index weighted fusion of peak clipping and valley filling effect and state of charge change is used to evaluate the optimal regulation strategy corresponding to each interval;Real-time fitting of current load power curve in operation stage, dynamic similarity matching with benchmark curve is realized using sliding time window and difference index, determine the preferred regulation fluctuation interval to which the current period belongs, call the optimal strategy of this interval to generate regulation instruction, realize the adaptive selection and issue of energy storage system regulation strategy, improve the transformer peak clipping and valley filling effect and energy storage operation economy.

Description

Technical Field

[0001] This invention belongs to the field of power system energy storage and regulation technology, specifically, it relates to a control method for an energy storage system applied to a transformer substation. Background Technology

[0003] As a key node at the end of the power system, the randomness and volatility of the power source and load sides of the distribution area are becoming increasingly prominent. With its flexible charging and discharging characteristics, energy storage system has become an important technical means to smooth out power fluctuations in the distribution area and improve power quality.

[0004] Existing technologies mostly rely on fixed thresholds or simple rules for regulation, such as setting uniform power limits or state of charge ranges, lacking in-depth mining and utilization of historical operating data of the distribution area. They usually directly determine whether to start energy storage charging and discharging based on the current load power, ignoring the periodic fluctuations of load power at different times of the day. This leads to frequent and untargeted regulation actions. During periods of rapid load power fluctuation, the system may trigger a response every time the power limit is exceeded, causing the energy storage system to switch states frequently. This not only shortens battery life but also makes it difficult to achieve true peak shaving and valley filling effects due to the lack of a holistic understanding of the fluctuation range. In addition, most existing technologies adopt a one-size-fits-all regulation strategy, that is, the same control logic is used regardless of whether the distribution area is in a peak, off-peak, or low-load period, which cannot adapt to the differences in energy storage demand at different times. Traditional methods also lack the ability to learn based on feedback from historical control results. Their strategies are often pre-set static rules that cannot be dynamically optimized based on historical control effects. This leads to long-term rigidity of strategies, making it difficult to adapt to the slow evolution of transformer load characteristics with factors such as seasons and user behavior. When matching real-time load with historical patterns, simple point-to-point comparisons are usually used, which are easily affected by instantaneous fluctuations and noise interference, resulting in low matching accuracy. Furthermore, they cannot capture the similarity of load curves within a continuous time window, thus misjudging the current operating status of the transformer area.

[0005] To address the aforementioned problems, this invention proposes a control method for energy storage systems applied to transformer substations. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a control method for energy storage systems applied to transformer substations, solving the problems of disconnect between control strategies and time-varying load fluctuation characteristics, and the lack of historical feedback optimization mechanisms in existing technologies.

[0007] The objective of this invention can be achieved through the following technical solutions: A control method for an energy storage system applied to a transformer substation, the method comprising: Step 1: Extract historical operating data associated with the target transformer area during the predetermined backtracking period from the cloud database, divide the historical operating data in a cyclical manner, construct several load power fluctuation curves associated with the target transformer area, and obtain the benchmark load power fluctuation curve corresponding to the target transformer area through curve fitting. Step 2: Based on the number of times the energy storage system is in a control state at different acquisition time points of the load power fluctuation curve, the control fluctuation range is divided. Combining the feedback results of different control strategies in the same control fluctuation range, a weighted fusion analysis is performed to evaluate the optimal control strategy associated with the corresponding control fluctuation range. Step 3: Based on the load power of the target distribution area at the current time, formulate the real-time load power fluctuation curve associated with the target distribution area, perform a similarity assessment between the curve and the benchmark load power fluctuation curve in real time, determine the preferred control fluctuation range corresponding to the target distribution area at the current time, and implement the optimal control strategy based on the preferred control fluctuation range.

[0008] As a further aspect of the present invention, the specific method for constructing several load power fluctuation curves associated with the target distribution area in step one is as follows: Determine the target transformer area Q; Get the backtracking period Th, where the duration of the backtracking period Th is a preset duration; Using the current time as the end time of the backtracking period Th, a backtracking period Th is determined in the past. Within this backtracking period Th, historical operating data associated with the target distribution area Q is extracted from the cloud database, including the load power P and output power F corresponding to different times. Extract the cycle period Tx, where the cycle period Tx spans 24 hours, with the start time and end time being 0:00 and 24:00 each day, respectively; The backtracking period Th is divided using the cycle period Tx to obtain several cycle periods. The total number is recorded as j. Among them, the division results that are less than a complete cycle period are directly discarded. Arrange the j cycle periods in chronological order, and denote them as the cycle period sequence Tx1, Tx2, ..., Txj; Extract any loop period Txi, where i is the counting index, 1≤i≤j; The load power is locked from historical operating data, and the load power within the cycle Txi is extracted and recorded in chronological order as the load power set Pi={p1,p2,...,pm}, where m is the total load power within a cycle and corresponds to the collection time point within the cycle. Construct a two-dimensional coordinate system with the time line as the horizontal axis and the power value as the vertical axis. Align the m load power values ​​in the load power set LPi={p1,p2,...,pm} with the scale of the horizontal and vertical axes and plot the m data points. Fit the curve to obtain the load power fluctuation curve LSi associated with the cycle period Txi. Similarly, the load power fluctuation curves associated with each cycle are determined as: LS1,LS2,...,LSj.

[0009] As a further aspect of the present invention, the specific method for obtaining the reference load power fluctuation curve corresponding to the target distribution area through curve fitting in step one is as follows: Extract j load power fluctuation curves LS1,LS2,...,LSj; Based on the density clustering algorithm of adaptive dynamic time bending distance, j load power fluctuation curves are divided into several clusters. Outlier curve clusters are removed, and the main cluster with the most curves is retained. The curves in the main cluster are denoted as LS1',LS2',...,LSk', 0≤k≤j; For all curves within the main cluster, the weighted kernel density estimate of the load power is calculated at m acquisition time points, where the kernel function is a Gaussian kernel and the bandwidth is adaptively determined by the Silverman criterion. The load power with the highest kernel density at each acquisition time point is extracted as the reference power value at that acquisition time point. All reference power values ​​are summarized in chronological order to obtain the initial reference load power sequence. The initial reference power sequence is smoothed by cubic Hermitian interpolation, and high-frequency noise components are filtered out by Fourier low-pass filtering to generate a reference load power sequence z1, z2, ..., zm. The reference load power fluctuation curve ZLS is constructed based on the reference load power sequence.

[0010] As a further aspect of the present invention, in step two, the specific method for dividing the control fluctuation range based on the number of times the energy storage system is in a control state at different acquisition time points of the load power fluctuation curve is as follows: Extract the reference load power fluctuation curve ZLS and the reference load power sequence z1, z2, ..., zm; Extract m data collection time points and arrange them into a data collection time point sequence t1, t2, ..., tm; For any data collection time point tn, count the number of times the energy storage system is in a control state at the data collection time point tn among j load power fluctuation curves LS1, LS2, ..., LSj, denoted as Yn, where 0≤Yn≤j, 1≤n≤m; Similarly, determine the number of times the energy storage system is in a controlled state at m data collection points, and summarize them as Y1, Y2, ..., Ym; Extract the preset threshold value Yth for the number of control cycles; For any number of times Yn at any collection time point tn, if Yn≥Yth, then the collection time point tn is marked as a high control frequency point; If 0 < Yn < Yth, mark the acquisition time point tn as the medium control frequency point; if Yn = 0, mark the acquisition time point tn as the low control frequency point. Traverse m collection time points, and merge the collection time points covered by consecutive high control frequency points, medium control frequency points and low control frequency points into high control fluctuation range AM, medium control fluctuation range BM and low control fluctuation range CM. Arrange the various control fluctuation ranges separately to obtain the high control fluctuation range set {AM1,AM2,...,AMa}, the medium control fluctuation range set {BM1,BM2,...,BMb}, and the low control fluctuation range set {BM1,BM2,...,BMb}, where a, b, and c are the total number of high, medium, and low control fluctuation ranges, respectively, and a+b+c≤m.

[0011] As a further aspect of the present invention, the specific method for evaluating the optimal control strategy associated with the corresponding control fluctuation range in step two is as follows: Extract any one of the high-controllability fluctuation intervals AMo from the set of high-controllability fluctuation intervals {AM1,AM2,...,AMa}, where 1≤o≤a; The system acquires the sequence of acquisition time points within the high-regulation fluctuation range AMo, as well as the regulation strategies adopted by the energy storage system at each acquisition time point and the corresponding regulation result feedback. The regulation result feedback includes peak shaving and valley filling effect scores and state of charge change scores. The control strategies used at all acquisition time points within the high control fluctuation range AMo are classified by type, and the number of control times for each control strategy is counted. The sum of the peak shaving and valley filling effect score and the state of charge change score is calculated and divided by the number of control times to obtain the mean, i.e. the mean peak shaving and valley filling effect score and the mean state of charge change score. The average scores for peak shaving and valley filling effects and the average scores for changes in state of charge are assigned weight coefficients α and β, respectively, and the weighted sum is obtained to obtain the comprehensive score for each regulation strategy. The candidate strategy with the highest comprehensive score is selected as the optimal control strategy associated with AMo in the high-control fluctuation range; Similarly, for each control fluctuation range in the set of high control fluctuation ranges {AM1,AM2,...,AMa}, the set of medium control fluctuation ranges {BM1,BM2,...,BMb}, and the set of low control fluctuation ranges {BM1,BM2,...,BMb}, the same operation is performed to evaluate the optimal control strategy.

[0012] As a further aspect of the present invention, in step two, the calculation methods for the peak shaving and valley filling effect score and the state of charge change score are as follows: Extract any one of the control strategies within the high-control fluctuation range AMo and label it CL; Determine the data collection time points covered during the execution of the control strategy CL, and mark them as the control strategy time interval TCL; The difference between the maximum and minimum difference between the load power P and the output power F when using the same control strategy CL in the same control strategy time interval TCL in historical operation data is recorded as the peak shaving and valley filling index. Extract the difference between the maximum and minimum difference between the load power P and the output power F within the same control strategy time interval TCL on the benchmark load power fluctuation curve ZLS, and denot it as the standard peak shaving and valley filling index. Divide the standard peak shaving and valley filling index by the peak shaving and valley filling index to obtain the peak shaving and valley filling effect score. Next, determine the absolute value of the change in state of charge of the energy storage system using the control strategy CL during the control strategy time interval TCL in the historical operation data, and obtain the state of charge change score by subtracting the absolute value of the change from 100%.

[0013] As a further aspect of the present invention, in step three, the specific method for determining the real-time load power fluctuation curve associated with the target distribution area based on the load power of the target distribution area at the current time is as follows: Determine the current time and obtain the load power of the target area at and after the current time in real time. Arrange them in chronological order and denot them as the real-time load power sequence SP1, SP2, ..., SPU, where u is the total real-time load power and u increases with time. Construct a two-dimensional coordinate system with time on the horizontal axis and power values ​​on the vertical axis. Align the real-time load power sequence SP1, SP2, ..., SPU with the scales of the horizontal and vertical axes and plot it as data points to obtain u data points. Fit the data points to a curve and denote it as the real-time load power fluctuation curve SLS.

[0014] As a further aspect of the present invention, in step three, determining the preferred control fluctuation range corresponding to the target station area at the current time, and executing the control using the optimal control strategy of this control fluctuation range, is specifically done as follows: Place the real-time load power fluctuation curve SLS and the reference load power fluctuation curve ZLS in the same two-dimensional coordinate system; Starting from the current time point, a time window of a preset duration T_ys is extracted along the positive horizontal axis, and the end time point is locked as the endpoint. Correspondingly, curve segment R1 on SLS and curve segment R2 on ZLS are extracted. Construct two straight lines, L1 and L2, that pass through the starting point and the ending point, and are perpendicular to the horizontal axis and parallel to the vertical axis, respectively. Obtain the area of ​​one or more closed regions formed by line L1, line L2, curve segment R1, and curve segment R2, and denote it as the difference index between SLS and ZLS within the time window; Obtain the preset difference index threshold. If the difference index is less than the difference index threshold, it is determined that the current real-time load power fluctuation curve and the benchmark load power fluctuation curve are highly matched within the time window. Based on the acquisition time point corresponding to the current time point on ZLS, determine the type of control fluctuation range to which it belongs. If the difference index is greater than or equal to the difference index threshold, the difference index between SLS and ZLS is calculated on ZLS with a fixed step size sliding time window of T_ys / N. The window position corresponding to the minimum value is selected as the best matching position, and the control fluctuation range type is determined by the acquisition time point of this best position on ZLS. N is a preset integer. The determined control fluctuation range is taken as the preferred control fluctuation range corresponding to the current time. The associated optimal control strategy is invoked, control instructions are generated and sent to the energy storage system for control.

[0015] The beneficial effects of this invention are: (1) This invention extracts historical operating data from the cloud and constructs a benchmark load power fluctuation curve to depict the typical power consumption pattern of the target distribution area. It divides the fluctuation range based on the number of control operations, and performs weighted fusion analysis by combining the feedback of different strategies in the same range to select the optimal control strategy corresponding to each range, thereby improving the adaptability and control efficiency of the energy storage system to different operating conditions. The similarity assessment between the real-time load power curve and the benchmark curve quickly identifies the preferred range to which the current operating condition belongs and directly calls the optimized strategy to execute control, reducing online calculation delay and decision bias. It can not only reduce the ineffective charging and discharging losses of the energy storage system and extend the equipment life, but also improve the power supply stability and economy of the distribution area, while reducing the dependence on manual experience parameters. (2) This invention effectively removes outlier curves by using an adaptive dynamic time bending density clustering algorithm, retains the most representative main cluster curves, improves the anti-interference ability and universality of the benchmark load power fluctuation curve, and avoids the distortion of benchmark features by abnormal operating data. On this basis, weighted kernel density estimation and Silverman adaptive bandwidth are used to capture the power probability distribution at each acquisition time point, and the power value with the largest kernel density is extracted as the benchmark, so that the benchmark curve is closer to the mainstream mode of actual operation. Combined with cubic Hermit interpolation and Fourier low-pass filtering, the curve shape is smoothed, high-frequency noise is eliminated, and key power change trends are retained, thereby obtaining the benchmark load power fluctuation curve, enhancing the robustness and accuracy of the load characteristic extraction of the transformer area, and providing a data foundation for subsequent load forecasting, anomaly detection and transformer area health assessment. (3) This invention accurately divides high, medium and low control fluctuation ranges by statistically analyzing the control frequency of the energy storage system at different time points in historical load fluctuations, thereby achieving refined identification of load characteristics and avoiding a one-size-fits-all control method; on this basis, the peak shaving and valley filling effects and state of charge changes of various control strategies are evaluated for each range, and the optimal strategy is selected by weighted comprehensive scoring method, which smooths load fluctuations, improves grid stability, takes into account the state of charge health management of energy storage batteries, and prevents overcharging or over-discharging. This adaptive, range-based optimization method improves the control efficiency and operational reliability of the energy storage system and reduces the number of ineffective controls; (4) This invention dynamically reflects the current operating status of the transformer area by collecting load power in real time and fitting it into a fluctuation curve. The real-time curve and the reference curve are placed in the same coordinate system, and the area of ​​the closed region within the time window is used as the difference index. This achieves intuitive and quantitative matching degree evaluation. When the difference index is small, the corresponding interval of the reference curve is directly matched to ensure rapid response of regulation. When the difference is large, the best matching position is automatically searched through the sliding window to effectively avoid misjudgment caused by phase shift or fluctuation pattern difference, improve the robustness and adaptability of regulation fluctuation interval identification, and ensure that the selected optimal regulation strategy is highly consistent with the actual operating conditions of the transformer area, thereby optimizing the regulation effect of the energy storage system and reducing energy loss. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a flowchart illustrating the method described in this invention; Figure 2 This is a schematic diagram of the structure of the method described in this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention discloses a control method for an energy storage system applied to a transformer substation. It aims to solve the technical problems of poor adaptability and unsatisfactory control effects of energy storage system regulation strategies due to dynamic changes in transformer substation load in existing technologies. By offline analysis of historical operating data of the transformer substation, a baseline load characteristic is constructed and an optimal regulation strategy library for different scenarios is pre-trained. Then, during online operation, by real-time matching of current load fluctuations with the baseline characteristics, the regulation strategy can be accurately and adaptively invoked, improving the peak shaving and valley filling effect and operational economy of the energy storage system.

[0020] like Figure 1 , Figure 2 As shown, this application provides a control method for an energy storage system applied to a transformer substation; As an embodiment 1 of this application, it specifically includes: Step 1: Extract historical operating data associated with the target transformer area during the predetermined backtracking period from the cloud database, divide the historical operating data in a cyclical manner, construct several load power fluctuation curves associated with the target transformer area, and obtain the benchmark load power fluctuation curve corresponding to the target transformer area through curve fitting. Step 2: Based on the number of times the energy storage system is in a control state at different acquisition time points of the load power fluctuation curve, the control fluctuation range is divided. Combining the feedback results of different control strategies in the same control fluctuation range, a weighted fusion analysis is performed to evaluate the optimal control strategy associated with the corresponding control fluctuation range. Step 3: Based on the load power of the target distribution area at the current time, formulate the real-time load power fluctuation curve associated with the target distribution area, perform a similarity assessment between the curve and the benchmark load power fluctuation curve in real time, determine the preferred control fluctuation range corresponding to the target distribution area at the current time, and implement the optimal control strategy based on the preferred control fluctuation range.

[0021] Example 2 This embodiment provides a basic control method for an energy storage system applied to a transformer substation, based on Embodiment 1. It is the basic implementation of the present invention, as detailed below: Before implementation, it needs to be clarified that the local controller of the energy storage system in this application establishes a secure connection with the cloud database deployed on the public cloud through a 4G wireless communication module. The controller extracts historical operating data of the target transformer area Q from the cloud database over the past 90 days, i.e., a preset backtracking period Th, including the load power P of the target transformer area and the output power F of the energy storage system collected every 15 minutes.

[0022] The controller divides 90 days of historical data into 24-hour cycles Tx, for example, from 00:00:00 of each day to 00:00:00 of the next day as one cycle.

[0023] After the above division operation, j=90 complete cycle periods are obtained. Boundary data with less than 24 hours are removed. For each day's cycle period, the controller plots the m load power data points of the day (m=24 hours / 15 minutes=96) in chronological order on a two-dimensional coordinate system with time as the horizontal axis and power value as the vertical axis. The cubic spline interpolation method is used to fit the curve to obtain the load power fluctuation curve corresponding to the day. This operation is repeated to finally obtain 90 load power fluctuation curves.

[0024] Next, the controller fits these 90 curves to generate a baseline load power fluctuation curve. Specifically, the controller calculates the median of the 90 load power at each acquisition time point, connects the calculated 96 medians in chronological order, and performs a Savitzky-Golay filter smoothing process. The final smoothed curve is the baseline load power fluctuation curve of the transformer area during the backtracking period.

[0025] Then, the controller retrieves the 90 historical load power fluctuation curves it has constructed and analyzes them one by one to determine whether the energy storage system is in a control state at each acquisition time point, such as whether it has executed charging or discharging commands. For any given data collection time point tn, count the number of times the energy storage system was in a controlled state Yn among the 90 curves. For example, at 8:00 AM, the energy storage system may have performed charging and discharging operations for 70 days, so Yn = 70. The value of n ranges from 1 to 96 in this embodiment.

[0026] Next, obtain a preset threshold value Yth for the number of control operations, set Yth=45, which is 50% of the total number of days (90); when Yn≥45, mark the collection time point tn as a high control frequency point; When 0 < Yn < 45, the acquisition time point tn is marked as the medium-frequency control point; When Yn=0, the collection time point tn is marked as a low control frequency point.

[0027] Then, the time intervals covered by all consecutive high-frequency control points are merged to obtain the high-frequency control fluctuation interval AM, such as [7:00-8:15]. Similarly, by merging, we obtain the medium-regulation fluctuation range BM and the low-regulation fluctuation range CM.

[0028] Arrange the various control fluctuation ranges separately to obtain the sets of high control fluctuation ranges {AM1,AM2,...,AM1}, medium control fluctuation ranges {BM1,BM2,...,BMb}, and low control fluctuation ranges {BM1,BM2,...,BMb}, where a, b, and c are the total number of high, medium, and low control fluctuation ranges, respectively, and a+b+c≤m. Next, for each defined control fluctuation range, such as a high control fluctuation range AMo=[18:30-20:45], where 1≤o≤a, the controller extracts the control strategies used at all historical acquisition points within the high control fluctuation range AMo and their result feedback. For example, historical operating data recorded that two control strategies were used in this high-regulation fluctuation range: constant power peak shaving strategy (strategy A) and variable power tracking strategy (strategy B).

[0029] The controller calculates the average scores for peak shaving and valley filling effects and state of charge change for both strategy A and strategy B. The results are as follows: The average score for the peak shaving and valley filling effect of Strategy A was 85, and the average score for the change in state of charge was 90. The average scores for the two items in Strategy B were 95 and 80, respectively.

[0030] With preset weighting coefficients α=0.6 and β=0.4, the overall score for strategy A is 85*0.6+90*0.4=87, and the overall score for strategy B is 95*0.6+80*0.4=89.

[0031] Therefore, strategy B, which has a higher overall score, is rated as the optimal control strategy for the control fluctuation range AMo. Similarly, the optimal control strategy is rated and stored for each high, medium, and low control fluctuation range.

[0032] When entering the online operation phase, the energy storage system controller receives load power data from the transformers in the target distribution area in real time and plots the real-time load power fluctuation curve. For example, if the current time is 13:00 in the afternoon and the controller has acquired real-time load power data from 00:00 to 13:00 on the same day, the controller will plot these data points into a real-time load power fluctuation curve.

[0033] The controller then compares the real-time load power fluctuation curve with the generated reference load power fluctuation curve on the same time coordinate system. The specific comparison method is as follows: Calculate the average absolute error of the load power values ​​at all corresponding time points between 00:00 and 13:00 for the two curves. The final calculated average absolute error is 3.2kW. Then, obtain the upper limit of the average absolute error corresponding to the similarity threshold preset by the operator, which is 5kW. Since 3.2kW < 5kW, the controller determines that the current real-time load fluctuation is highly matched with the baseline load fluctuation.

[0034] Therefore, the controller determines the type of control fluctuation range that the current time 13:00 belongs to on the reference load power fluctuation curve. For example, if data analysis shows that 13:00 falls within the medium control fluctuation range BM2, the controller will then call the optimal control strategy associated with the medium control fluctuation range BM2 from the locally stored strategy library, such as the shallow charge and discharge smoothing strategy, and automatically generate specific charge and discharge commands (this step can be adaptively implemented in combination with existing energy storage systems, so it is considered existing technology and will not be elaborated on in this solution). For example, charge at 20kW power for 30 minutes and execute the control through the CAN bus or by sending it to the power conversion system of the energy storage system.

[0035] Based on the content described in this embodiment, by constructing a baseline curve and a pre-trained strategy library offline, and matching and calling the optimal strategy online in real time, adaptive control of the energy storage system is realized. Compared with the traditional fixed timing control method, it can dynamically adapt to the changes in the load characteristics of the transformer area, thereby improving the accuracy and effectiveness of the control strategy and avoiding the efficiency loss caused by strategy mismatch.

[0036] Example 3 This embodiment introduces a data processing and curve fitting method based on Embodiment 2. The difference from Embodiment 2 is that this embodiment uses more advanced clustering, kernel density estimation, and filtering algorithms in the construction of the benchmark load power fluctuation curve to remove outlier data and improve the robustness and representativeness of the benchmark curve, as detailed below: In this embodiment, the construction method of the load power fluctuation curve is basically the same as that in embodiment 2. The difference is that this embodiment clarifies the data elimination rules. That is, the controller obtains the historical operating data of Th=120 days of the backtracking period from the cloud, divides it into 120 cycles with 24-hour cycle period Tx. If the data of the last day is less than 24 hours, the data of the incomplete cycle is directly discarded and not included in the subsequent calculation, and finally 120 load power fluctuation curves are constructed.

[0037] Next, after the controller obtains 120 load power fluctuation curves, it does not directly calculate the mean or median. Instead, it first uses a density clustering algorithm based on adaptive dynamic time warp (ADTW) distance to cluster the curves. The principle is that ADTW distance is more tolerant of small deviations of the curves on the time axis than traditional Euclidean distance, so this method is adopted. Then, the controller automatically determines the clustering parameters through grid search, including the neighborhood radius Eps and the minimum number of points MinPts. After clustering, the main cluster containing the most curves is selected. The curves in the main cluster represent the most typical load pattern of the transformer area. Outliers caused by holidays, sudden accidents, etc., such as clusters with a small number of curves or noise points, are directly removed. Therefore, a specific example is given to facilitate the subsequent operation process: out of 120 curves, the main cluster contains 105 typical curves, and the remaining 15 outlier curves are removed.

[0038] For the remaining 105 typical curves, the controller performs weighted kernel density estimation on these 105 load power values ​​at each acquisition time point. The kernel function is a standard Gaussian kernel function, and the bandwidth is adaptively determined by the Silverman criterion. At each acquisition time point, the controller calculates the power value with the largest kernel density estimate as the baseline power value for that acquisition time point, because it represents the load power most likely to occur at that acquisition time point. Finally, the initial baseline load power of 105 acquisition time points is summarized to obtain the initial baseline load power sequence. To mitigate any minor jitter that may exist in the initial reference load power sequence, the controller first upsamples the initial reference load power sequence using cubic Hermite interpolation to make it smoother. Then, it filters the upsampled initial reference load power sequence using a zero-phase Fourier low-pass filter with a cutoff frequency of twice the fundamental frequency to remove high-frequency noise components caused by measurement noise or random fluctuations, ultimately generating a smooth and stable reference load power fluctuation curve ZLS.

[0039] This embodiment introduces ADTW density clustering to remove outlier curves, thus avoiding the contamination of baseline characteristics by abnormal historical data. It uses kernel density estimation to select the most likely power value, which reflects the typical state of the load better than the mean or median. Finally, by combining interpolation and filtering, the generated baseline curve is smoother and more robust.

[0040] Example 4

[0041] This embodiment further refines the specific implementation logic of the control fluctuation range division and optimal strategy evaluation based on embodiment 3, and discloses the specific calculation formulas for peak shaving and valley filling effect scores and state of charge change scores, as follows: After the controller completes the construction of the reference load power fluctuation curve ZLS, it performs statistical analysis on the number of times the energy storage system is controlled at each acquisition time point in the historical operation data, Yn. The analysis is performed by setting two thresholds: the high control threshold Yth_high=0.6*j and the low control threshold Yth_low=0.1*j, where j is the total number of cycle periods. If Yn≥Yth_high, then mark the collection time point tn as a high control frequency point; If Yth_low < Yn < Yth_high, mark the collection time point tn as the medium-frequency control point; If Yn≤Yth_low, then mark the collection time point tn as a low control frequency point.

[0042] Finally, the controller iterates through all the data collection time points in chronological order to complete the mark merging operation. For example, all data collection time points are marked and arranged in chronological order as: [high, high, high, medium, medium, high, high, low, low...]. Consecutive high marks are merged to form a high control fluctuation range, such as the range A1=[t1,t3]. Similarly, consecutive medium marks are merged to form a medium control fluctuation range, and consecutive low marks are merged to form a low control fluctuation range. If there is a single data collection time point, it is directly regarded as an outlier and discarded.

[0043] For example, the high-regulation fluctuation range AMo described in Example 3 is extracted, and the optimal strategy evaluation is illustrated using examples based on the high-regulation fluctuation range AMo. The cloud-based historical records show that a constant power peak-shaving strategy CL was executed during the high-regulation fluctuation range AMo. The controller finds the historical moment when the constant power peak-shaving strategy CL was executed, records the time interval it covers, and denotes it as the regulation strategy time interval TCL, for example, 19:00-20:00.

[0044] From historical operating data, extract the actual load power P and energy storage output power F of the target transformer area Q within the control strategy time interval TCL, and calculate the peak shaving and valley filling index = max(PF) - min(PF), which is the difference between the maximum difference and the minimum difference between the load power P and the output power F using the control strategy CL in the same control strategy time interval TCL. The peak shaving and valley filling index reflects the peak-valley difference of the net load of the target distribution area under the effect of energy storage. At the same time, on the benchmark load power fluctuation curve ZLS, the load power P_base in the TCL is extracted, and the standard peak shaving and valley filling index = max(P_base) - min(P_base) is calculated. Then the peak shaving and valley filling effect score = (standard peak shaving and valley filling index / peak shaving and valley filling index) * 100. The higher the value of the peak shaving and valley filling effect score, the better the peak shaving and valley filling effect of energy storage.

[0045] Retrieve the state of charge (SOC) of the energy storage system before and after executing strategy CL, denoted as SOC_start and SOC_end respectively, and calculate the absolute value of the change in SOC = |SOC_end - SOC_start|. The state of charge change score is calculated as (100% - absolute value of SOC change) * 100. The higher the state of charge change score, the smaller the change in the energy storage state of charge while completing the regulation task, and the more sufficient the reserve for subsequent regulation capabilities.

[0046] The controller then compiles statistics on all historically used strategies within the high-fluctuation range AMo. Assuming there are strategies CL1, CL2, ..., for each strategy, the average score of peak shaving and valley filling effect and the average score of state of charge change are calculated based on the above method. Then, weighting coefficients are set according to the current target area's operating target, such as α=0.7 and β=0.3, and the comprehensive score of each strategy is calculated as α*average score of peak shaving and valley filling effect + β*average score of state of charge change. Finally, the controller selects the strategy with the highest comprehensive score as the optimal control strategy for the AMo correlation in the high control fluctuation range and stores it in the pre-built strategy library. It should be noted that the strategies in the strategy library are updated every predetermined update period.

[0047] This embodiment introduces a dual-threshold-based interval partitioning method and a quantitative weighted scoring model, transforming the strategy selection process from experience-based judgment to data-driven objective calculation. This improves the accuracy of selecting the optimal strategy and enables the strategy library to be dynamically optimized based on historical feedback, thus possessing self-evolution capabilities.

[0048] Example 5 This embodiment further discloses a method for formulating real-time load power fluctuation curves and a method for dynamic similarity evaluation and strategy matching based on embodiment 4. The goal is to solve the problem of incomplete temporal alignment between real-time load fluctuations and historical benchmark patterns. By using a sliding time window matching algorithm, the optimal matching position of the current load fluctuation pattern in the historical benchmark is accurately found, thereby correctly selecting the control strategy, as detailed below: First, the controller of the energy storage system establishes a 4-hour circular buffer to store the load power of the target area for each acquisition operation. Starting from the current time, the controller retrieves the real-time load power stored in the buffer in chronological order to form a time-continuous real-time load power sequence SP1, SP2, ..., SPU, where u is the total real-time load power and u increases with time. The controller plots each data point in the real-time load power sequence SP1, SP2, ..., SPU in a two-dimensional coordinate system with time as the horizontal axis and power value as the vertical axis, and fits adjacent data points with curves to form a continuously extending real-time load power fluctuation curve SLS. Place the real-time load power fluctuation curve SLS and the reference load power fluctuation curve ZLS in the same two-dimensional coordinate system; Let the current time be t_now. The controller uses [t_now-T_ys,t_now] as the time window on the reference load power fluctuation curve ZLS. It extracts a curve segment R1 of the same time length on SLS and a curve segment R2 of the corresponding time period on ZLS. Here, T_ys is the preset duration, which is set to 30 minutes.

[0049] Construct two straight lines, L1 and L2, that pass through the starting point and the ending point, and are perpendicular to the horizontal axis and parallel to the vertical axis, respectively. Calculate the total area of ​​one or more closed regions enclosed by straight line L1, straight line L2, curve segment R1, and curve segment R2, and denote it as the SLS and ZLS difference index within the time window. Specifically, for each pair of sampling points corresponding to the time on the two curves within the time window, calculate the absolute value of the power difference, and then integrate this absolute difference within the time window. The integration result is the difference index. The smaller the difference index, the more similar the two curves are within the time window. Obtain the operator's preset difference index threshold, such as 20kW·min; If the calculated difference index is less than the difference index threshold, it is determined that the real-time curve and the benchmark curve are highly matched within the current time window. The controller directly uses the acquisition time point corresponding to the current time point t_now on ZLS to determine the type of control fluctuation range to which it belongs. If the calculated difference index is greater than or equal to the difference index threshold, it indicates that the current load pattern has shifted in time relative to the baseline historical pattern. At this point, the controller starts sliding window matching, sets N=10, and sets a fixed step size of T_ys / N=3 minutes. The controller is on ZLS, with [t_now-T_ys,t_now] as the initial sliding window, and then slides at most N / 2=5 times in the historical time direction and the future time direction, generating a maximum of 11 candidate window positions in total. For each candidate window, repeat the above steps to calculate its difference index with SLS, and select the window position with the smallest difference index among all candidate windows as the best matching position of the current real-time load fluctuation on the baseline curve. After determining the optimal matching position, the acquisition time point of the matching sliding window on the ZLS corresponding to the current time t_now is obtained. Based on the acquisition time point, the control fluctuation range to which it belongs is determined. Finally, the optimal control strategy associated with the corresponding control fluctuation range is called from the strategy library and immediately issued for execution.

[0050] This embodiment enhances the adaptability to the time offset of the load pattern in the distribution area by introducing a sliding time window and a difference area minimization matching algorithm. Even when the peak and valley periods of the load are not completely consistent with historical patterns due to holidays, sudden weather changes, etc., it can accurately match the most similar scenario in history and adopt the verified optimal strategy to improve the intelligence level of the control system.

[0051] All data in the formulas described above have been calculated with dimensions removed. Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0052] The above description is merely an example and illustration of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0053] It should be stated that all user data collected in this application was collected with the user's consent and authorization. Furthermore, the uses of user data are legal and compliant, and the use and processing of user data comply with the relevant laws, regulations, and standards of the relevant regions.

Claims

1. A control method for an energy storage system applied to a transformer substation, characterized in that, The method includes: Step 1: Extract historical operating data associated with the target transformer area during the predetermined backtracking period from the cloud database, divide the historical operating data in a cyclical manner, construct several load power fluctuation curves associated with the target transformer area, and obtain the benchmark load power fluctuation curve corresponding to the target transformer area through curve fitting. Step 2: Based on the number of times the energy storage system is in a control state at different acquisition time points of the load power fluctuation curve, the control fluctuation range is divided. Combining the feedback results of different control strategies in the same control fluctuation range, a weighted fusion analysis is performed to evaluate the optimal control strategy associated with the corresponding control fluctuation range. Step 3: Based on the load power of the target distribution area at the current time, formulate the real-time load power fluctuation curve associated with the target distribution area, perform a similarity assessment between the curve and the benchmark load power fluctuation curve in real time, determine the preferred control fluctuation range corresponding to the target distribution area at the current time, and implement the optimal control strategy based on the preferred control fluctuation range.

2. The method according to claim 1, characterized in that, In step one, the specific method for constructing several load power fluctuation curves associated with the target transformer area is as follows: Determine the target transformer area Q; Get the backtracking period Th, where the duration of the backtracking period Th is a preset duration; Using the current time as the end time of the backtracking period Th, a backtracking period Th is determined in the past. Within this backtracking period Th, historical operating data associated with the target distribution area Q is extracted from the cloud database, including the load power P and output power F corresponding to different times. Extract the cycle period Tx, where the cycle period Tx spans 24 hours, with the start time and end time being 0:00 and 24:00 each day, respectively; The backtracking period Th is divided using the cycle period Tx to obtain several cycle periods. The total number is recorded as j. The division results that are less than a complete cycle period are directly discarded. Arrange the j cycle periods in chronological order, and denote them as the cycle period sequence Tx1, Tx2, ..., Txj; Extract any loop period Txi, where i is the counting index, 1≤i≤j; The load power is locked from historical operating data, and the load power within the cycle Txi is extracted and recorded in chronological order as the load power set Pi={p1,p2,...,pm}, where m is the total load power within a cycle and corresponds to the collection time point within the cycle. Construct a two-dimensional coordinate system with the time line as the horizontal axis and the power value as the vertical axis. Align the m load power values ​​in the load power set LPi={p1,p2,...,pm} with the scale of the horizontal and vertical axes and plot the m data points. Fit the curve to obtain the load power fluctuation curve LSi associated with the cycle period Txi. Similarly, the load power fluctuation curves associated with each cycle are determined as: LS1,LS2,...,LSj.

3. The method according to claim 2, characterized in that, In step one, the specific method for obtaining the reference load power fluctuation curve corresponding to the target distribution area through curve fitting is as follows: Extract j load power fluctuation curves LS1,LS2,...,LSj; Based on the density clustering algorithm of adaptive dynamic time bending distance, j load power fluctuation curves are divided into several clusters. Outlier curve clusters are removed, and the main cluster with the most curves is retained. The curves in the main cluster are denoted as LS1',LS2',...,LSk', 0≤k≤j; For all curves within the main cluster, the weighted kernel density estimate of the load power is calculated at m acquisition time points, where the kernel function is a Gaussian kernel and the bandwidth is adaptively determined by the Silverman criterion. The load power with the highest kernel density at each acquisition time point is extracted as the reference power value at that acquisition time point. All reference power values ​​are summarized in chronological order to obtain the initial reference load power sequence. The initial reference power sequence is smoothed by cubic Hermitian interpolation, and high-frequency noise components are filtered out by Fourier low-pass filtering to generate a reference load power sequence z1, z2, ..., zm. The reference load power fluctuation curve ZLS is constructed based on the reference load power sequence.

4. The method according to claim 3, characterized in that, In step two, the specific method for dividing the control fluctuation range based on the number of times the energy storage system is in a control state at different acquisition time points of the load power fluctuation curve is as follows: Extract the reference load power fluctuation curve ZLS and the reference load power sequence z1, z2, ..., zm; Extract m data collection time points and arrange them into a data collection time point sequence t1, t2, ..., tm; For any data collection time point tn, count the number of times the energy storage system is in a control state at the data collection time point tn among j load power fluctuation curves LS1, LS2, ..., LSj, denoted as Yn, where 0≤Yn≤j, 1≤n≤m; Similarly, determine the number of times the energy storage system is in a controlled state at m data collection points, and summarize them as Y1, Y2, ..., Ym; Extract the preset threshold value Yth for the number of control cycles; For any number of times Yn at any collection time point tn, if Yn≥Yth, then the collection time point tn is marked as a high control frequency point; If 0 < Yn < Yth, mark the acquisition time point tn as the medium control frequency point; if Yn = 0, mark the acquisition time point tn as the low control frequency point. Traverse m collection time points, and merge the collection time points covered by consecutive high control frequency points, medium control frequency points and low control frequency points into high control fluctuation range AM, medium control fluctuation range BM and low control fluctuation range CM. Arrange the various control fluctuation ranges separately to obtain the high control fluctuation range set {AM1,AM2,...,AMa}, the medium control fluctuation range set {BM1,BM2,...,BMb}, and the low control fluctuation range set {BM1,BM2,...,BMb}, where a, b, and c are the total number of high, medium, and low control fluctuation ranges, respectively, and a+b+c≤m.

5. The method according to claim 4, characterized in that, In step two, the specific method for evaluating the optimal control strategy associated with the corresponding control fluctuation range is as follows: Extract any one of the high-controllability fluctuation intervals AMo from the set of high-controllability fluctuation intervals {AM1,AM2,...,AMa}, where 1≤o≤a; The system acquires the sequence of acquisition time points within the high-regulation fluctuation range AMo, as well as the regulation strategies adopted by the energy storage system at each acquisition time point and the corresponding regulation result feedback. The regulation result feedback includes peak shaving and valley filling effect scores and state of charge change scores. The control strategies used at all acquisition time points within the high control fluctuation range AMo are classified by type, and the number of control times for each control strategy is counted. The sum of the peak shaving and valley filling effect score and the state of charge change score is calculated and divided by the number of control times to obtain the mean, i.e. the mean peak shaving and valley filling effect score and the mean state of charge change score. The average scores for peak shaving and valley filling effects and the average scores for changes in state of charge are assigned weight coefficients α and β, respectively, and the weighted sum is obtained to obtain the comprehensive score for each regulation strategy. The candidate strategy with the highest comprehensive score is selected as the optimal control strategy associated with AMo in the high-control fluctuation range; Similarly, for each control fluctuation range in the set of high control fluctuation ranges {AM1,AM2,...,AMa}, the set of medium control fluctuation ranges {BM1,BM2,...,BMb}, and the set of low control fluctuation ranges {BM1,BM2,...,BMb}, the same operation is performed to evaluate the optimal control strategy.

6. The method according to claim 5, characterized in that, In step two, the calculation methods for the peak shaving and valley filling effect score and the state of charge change score are as follows: Extract any one of the control strategies within the high-control fluctuation range AMo and label it CL; Determine the data collection time points covered during the execution of the control strategy CL, and mark them as the control strategy time interval TCL; The difference between the maximum and minimum difference between the load power P and the output power F when using the same control strategy CL in the same control strategy time interval TCL in historical operation data is recorded as the peak shaving and valley filling index. Extract the difference between the maximum and minimum difference between the load power P and the output power F within the same control strategy time interval TCL on the benchmark load power fluctuation curve ZLS, and denot it as the standard peak shaving and valley filling index. Divide the standard peak shaving and valley filling index by the peak shaving and valley filling index to obtain the peak shaving and valley filling effect score. Next, determine the absolute value of the change in state of charge of the energy storage system using the control strategy CL during the control strategy time interval TCL in the historical operation data, and obtain the state of charge change score by subtracting the absolute value of the change from 100%.

7. The method according to claim 5, characterized in that, In step three, the specific method for determining the real-time load power fluctuation curve associated with the target distribution area based on the load power of the target distribution area at the current time is as follows: Determine the current time and obtain the load power of the target area at and after the current time in real time. Arrange them in chronological order and denot them as the real-time load power sequence SP1, SP2, ..., SPU, where u is the total real-time load power and u increases with time. Construct a two-dimensional coordinate system with time on the horizontal axis and power values ​​on the vertical axis. Align the real-time load power sequence SP1, SP2, ..., SPU with the scales of the horizontal and vertical axes and plot it as data points to obtain u data points. Fit the data points to a curve and denote it as the real-time load power fluctuation curve SLS.

8. The method according to claim 7, characterized in that, In step three, the preferred control fluctuation range corresponding to the target area at the current time is determined, and the specific method for implementing control based on the optimal control strategy of this control fluctuation range is as follows: Place the real-time load power fluctuation curve SLS and the reference load power fluctuation curve ZLS in the same two-dimensional coordinate system; Starting from the current time point, a time window of a preset duration T_ys is extracted along the positive horizontal axis, and the end time point is locked as the endpoint. Correspondingly, curve segment R1 on SLS and curve segment R2 on ZLS are extracted. Construct two straight lines, L1 and L2, that pass through the starting point and the ending point, and are perpendicular to the horizontal axis and parallel to the vertical axis, respectively. Obtain the area of ​​one or more closed regions formed by line L1, line L2, curve segment R1, and curve segment R2, and denote it as the difference index between SLS and ZLS within the time window; Obtain the preset difference index threshold. If the difference index is less than the difference index threshold, it is determined that the current real-time load power fluctuation curve and the benchmark load power fluctuation curve are highly matched within the time window. Based on the acquisition time point corresponding to the current time point on ZLS, determine the type of control fluctuation range to which it belongs. If the difference index is greater than or equal to the difference index threshold, the difference index between SLS and ZLS is calculated on ZLS with a fixed step size sliding time window of T_ys / N. The window position corresponding to the minimum value is selected as the best matching position, and the control fluctuation range type is determined by the acquisition time point of this best position on ZLS. N is a preset integer. The determined control fluctuation range is taken as the preferred control fluctuation range corresponding to the current time. The associated optimal control strategy is invoked, control instructions are generated and sent to the energy storage system for control.

Images