A power grid seasonal load balance regulation method based on energy storage optimization

Abstract

The application discloses a kind of grid seasonal load balancing regulation methods based on energy storage optimization, belong to power system dispatching technical field, obtain energy storage equipment operating parameter, construct multidimensional response characteristic mapping system, determine response boundary by response capability vector space and combination threshold template combination;Through double time scale load variation monitoring, the real-time calculation and trend judgment of grid load variation rate are carried out, generate load surge early warning signal, implement cross-node collaborative scheduling to realize multi-energy storage station distributed energy allocation;According to the correlation analysis of environmental parameters and load characteristics, seasonal adaptability regulation is carried out, and the seasonal switching threshold value in the dispatching instruction is dynamically adjusted;Response deviation detection is carried out by scheduling execution monitoring, identifies scheduling deviation segment and marks, to use scheduling deviation segment as input by adaptive learning to update scheduling parameter set and instruction generation logic, realize scheduling logic closed loop optimization.

Description

Technical Field

[0001] This invention relates to the field of power system dispatching technology, and specifically to a power grid seasonal load balance regulation method based on energy storage optimization. Background Technology

[0002] During periods of high summer temperatures and low winter temperatures, the concentrated surge in air conditioning and heating loads leads to a sharp increase in the peak-to-valley difference in the power grid. Energy storage technology, as a crucial means of power grid flexibility regulation, can release electricity during peak load periods and absorb electricity during off-peak periods, thereby achieving peak shaving and valley filling and alleviating grid operational pressure. However, with the large-scale deployment of energy storage devices in the power grid, several pressing technical challenges remain: how to achieve coordinated scheduling of multiple energy storage nodes, how to accurately allocate regulation tasks based on the real-time operating status of energy storage devices, and how to dynamically adjust scheduling strategies to adapt to seasonal load changes.

[0003] Existing energy storage dispatching methods mostly employ fixed threshold triggering and single-node control strategies, lacking a dynamic evaluation mechanism for the response capabilities of energy storage devices. In actual operation, operating parameters such as the state of charge, cell temperature, and charge / discharge cycle count of energy storage devices change constantly, directly affecting their response speed and adjustable power range. However, traditional methods cannot quantify the impact of these changes on dispatching capabilities in real time, leading to a mismatch between dispatching commands and the actual capabilities of the devices, resulting in response delays or even dispatching failures. Furthermore, existing methods rely primarily on threshold judgments at a single time scale for load forecasting and early warning, lacking the ability to identify sudden load changes in advance and failing to allow sufficient preparation time for energy storage devices to respond.

[0004] In terms of collaborative scheduling of multiple energy storage nodes, existing technologies generally adopt equal-sharing strategies or static priority allocation methods, failing to comprehensively consider multiple factors such as the remaining capacity, response efficiency, and regional redundancy of each node. This results in some nodes operating under overload while others remain idle, leading to low overall system regulation efficiency. Especially during seasonal transitions, there is a complex coupling relationship between temperature changes and load fluctuations. Traditional methods, which use fixed seasonal switching thresholds, cannot adapt to actual climate change trends, resulting in discontinuous strategy switching and causing regulation instability.

[0005] More critically, existing energy storage scheduling methods lack closed-loop feedback and adaptive optimization capabilities. After scheduling is executed, it is impossible to identify response deviations, analyze the causes of failures, or automatically correct scheduling parameters and optimize allocation strategies based on historical operating data. This results in the system operating in a fixed mode for a long time, unable to continuously improve with the accumulation of operating experience. Summary of the Invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] Therefore, this invention mainly addresses the technical problems of existing energy storage scheduling methods, such as inaccurate response capability assessment, untimely load change early warning, unreasonable multi-node collaborative allocation, poor seasonal adaptability, and lack of adaptive optimization capabilities.

[0008] To solve the above technical problems, the present invention provides the following technical solution: a method for seasonal load balancing regulation of power grid based on energy storage optimization, which includes: obtaining the current operating parameters of energy storage devices, constructing a multi-dimensional response feature mapping system, and determining the response boundary of energy storage devices under the current operating conditions by combining the response capability vector space with the combined threshold template.

[0009] By monitoring load changes on dual time scales, the power grid load change rate is calculated and trend is judged in real time, and a load surge warning signal is generated. Based on the load surge warning signal, cross-node collaborative scheduling is performed, and distributed energy allocation of multiple energy storage stations is obtained through optimized allocation.

[0010] During the energy storage node regulation process, seasonal adaptive regulation is performed based on the correlation analysis between environmental parameters and load characteristics, and the seasonal switching threshold value in the dispatching instructions is dynamically adjusted.

[0011] During the scheduling execution process, response deviations are detected through scheduling execution monitoring, scheduling deviation segments are identified and marked, and scheduling deviation segments are used as input to update the scheduling parameter set and instruction generation logic through adaptive learning, thereby obtaining closed-loop optimization of the scheduling logic.

[0012] As a preferred embodiment of the power grid seasonal load balancing regulation method based on energy storage optimization described in this invention, the step of obtaining the current operating parameters of the energy storage device includes obtaining the state of charge, cell temperature distribution, output voltage fluctuation range, output current change rate, and historical discharge capacity of the energy storage device as operating parameters.

[0013] The construction of the multidimensional response feature mapping system includes inputting the running parameters into the response capability evaluation matrix for weighted calculation and selecting an appropriate boundary threshold template.

[0014] Construct a response capability vector space with operating parameters as coordinate axes, and calibrate the minimum adjustment cycle and maximum power change amplitude corresponding to the current equipment state in this vector space as response boundaries;

[0015] By matching the current state point in the vector space with the historical operating trajectory, the current response level of the energy storage device is automatically classified.

[0016] As a preferred embodiment of the power grid seasonal load balancing regulation method based on energy storage optimization described in this invention, the response capability evaluation matrix adopts weighted calculation, sets weight coefficients for multiple operating parameters, and introduces external ambient temperature, load change trend of scheduling cycle and battery charge and discharge cycle number to adjust the weight coefficients;

[0017] The boundary threshold template is pre-classified and constructed according to temperature range, load change rate range and battery cycle aging level. An evaluation reference template is selected by calculating the distance between the current feature combination and the central feature value of the template.

[0018] As a preferred embodiment of the power grid seasonal load balance adjustment method based on energy storage optimization described in this invention, the load change monitoring through dual time scales includes: constructing continuous time-series load data, generating a trigger threshold by combining the response boundary, and setting a recovery threshold based on the trigger threshold to form a dual threshold set;

[0019] A short-term sliding time window and a long-term sliding time window are set to be executed in parallel. The short-term sliding time window is used to detect abrupt changes in the load change rate, and the long-term sliding time window is used to extract the load change trend.

[0020] When the load change rate exceeds the trigger threshold within the short-term sliding time window, and the average change rate within the long-term sliding time window meets the trend judgment condition, the advance response time is estimated based on the current change rate and compared with the equipment adjustment capability.

[0021] The load surge warning signal is generated when the criterion is met.

[0022] As a preferred embodiment of the power grid seasonal load balancing regulation method based on energy storage optimization described in this invention, the execution of cross-node collaborative scheduling includes: acquiring load surge warning signals, extracting response capability level labels of each energy storage node, and filtering out a set of target energy storage nodes with adjustable capabilities.

[0023] Perform a regulation capability assessment on the target energy storage node set and construct a comprehensive score, and construct an allocation weight vector based on the comprehensive score;

[0024] By combining the weighted vector and the target response time period in the early warning signal, a unified scheduling instruction table is generated and sent to the target energy storage node;

[0025] During the scheduling process, the execution status of each node is monitored, and a replacement allocation is triggered when an abnormal node response is detected.

[0026] As a preferred embodiment of the power grid seasonal load balancing regulation method based on energy storage optimization described in this invention, the following steps are included: 1) Analyzing the correlation between environmental parameters and load characteristics, including continuously collecting external temperature data for a preset time period and corresponding daytime and nighttime load data within the daily cycle; 2) Calculating the daytime and nighttime load ratio based on the daytime and nighttime load data according to the daily cycle, and calculating the moving average of temperature using a sliding time window; 3) Constructing a bivariate time series matrix by aligning the temperature moving average sequence and the daytime and nighttime load ratio sequence according to timestamps, and calculating the linear correlation coefficient; 4) When the linear correlation coefficient exceeds a preset coupling threshold, extracting the directional parameter of the trend of the daytime and nighttime load ratio changing with the temperature moving average, and constructing a seasonal trend index vector.

[0027] As a preferred embodiment of the power grid seasonal load balance regulation method based on energy storage optimization described in this invention, the method of dynamically adjusting the seasonal switching threshold value in the dispatching instruction includes comparing the seasonal switching threshold value in the dispatching instruction with the seasonal trend index vector and calculating the deviation value of the trend direction and trend intensity in each threshold type.

[0028] For threshold types where the deviation value exceeds the preset offset threshold, a dynamic correction coefficient is determined based on the deviation value. The corresponding seasonal switching threshold value is then corrected based on the dynamic correction coefficient, generating a corrected threshold set and replacing the seasonal switching threshold value in the original scheduling instruction.

[0029] As a preferred embodiment of the power grid seasonal load balancing regulation method based on energy storage optimization described in this invention, the step of detecting response deviation through scheduling execution monitoring includes:

[0030] Extract the energy storage load response commands and the actual power output timing data of the corresponding energy storage nodes within the current execution cycle;

[0031] Align the scheduling target curve with the actual response curve to divide the instruction execution into multiple instruction execution segments, and construct a segment comparison dataset;

[0032] For each scheduling segment, the changing trends of the target response curve and the actual response curve are calculated, and the differences in power values ​​are compared.

[0033] When the power deviation of multiple consecutive sampling points exceeds the set threshold, and the trend direction is opposite or the delay time exceeds the equipment's adjustment capability, the segment is identified as a scheduling deviation segment and marked.

[0034] As a preferred embodiment of the power grid seasonal load balancing regulation method based on energy storage optimization described in this invention, when marking scheduling deviation segments, the scheduling deviation segments are classified into response delay type deviation, response missing type deviation, and reverse response type deviation. The deviation segments are weighted and quantified by four dimensions: maximum power deviation value, deviation duration, trend conflict ratio, and delay duration. The deviation segments are classified into three levels: mild, moderate, and severe according to the scoring results.

[0035] As a preferred embodiment of the power grid seasonal load balance adjustment method based on energy storage optimization described in this invention, the step of updating scheduling parameters through adaptive learning includes classifying and organizing the scheduling deviation segments according to time period, energy storage node identifier, and response capability level label to construct a training sample set.

[0036] The correspondence between the trigger thresholds corresponding to scheduling deviation segments and the actual response behavior is extracted from the training sample set.

[0037] The beneficial effects of this invention are as follows: By constructing a multi-dimensional response feature mapping system, this invention quantifies and calibrates the state of charge, cell temperature, voltage fluctuation, current change rate, and historical discharge capacity of energy storage devices in a five-dimensional vector space. Combined with a combined threshold template, it achieves accurate assessment of the regulation capability of energy storage devices under different operating conditions, solving the scheduling mismatch problem caused by the inability of traditional fixed threshold methods to adapt to changes in the dynamic response capability of devices. Through dual-timescale load change monitoring, a short-time sliding window captures load mutation characteristics, and a long-time sliding window extracts trend changes. Under the joint triggering criteria, early warning signals are generated in advance, reserving sufficient response preparation time for energy storage devices and avoiding the problem of insufficient identification of load mutations by single-timescale monitoring. Based on the remaining capacity margin, unit capacity response efficiency, and adjacent node redundancy ratio, a dynamic weight allocation vector is constructed to achieve accurate collaborative scheduling of multiple energy storage stations, avoiding node overload and resource idleness problems caused by equal distribution strategies or static priority methods. Through linear correlation analysis of temperature sliding mean and day-night load ratio, a seasonal trend index vector is constructed and the seasonal switching threshold value is dynamically corrected, so that the scheduling strategy remains continuous during seasonal transitions and eliminates strategy mutations caused by fixed thresholds. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1The above is a general flowchart of a power grid seasonal load balancing regulation method based on energy storage optimization, provided as an embodiment of the present invention.

[0040] Figure 2 This is a flowchart illustrating Embodiment 2 of a power grid seasonal load balancing regulation method based on energy storage optimization, which is an embodiment of the present invention.

[0041] Figure 3 This is another schematic diagram of a power grid seasonal load balancing regulation method based on energy storage optimization, provided as an embodiment of the present invention. Detailed Implementation

[0042] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0043] Example 1, referring to Figures 1-3 This is one embodiment of the present invention, which provides a grid seasonal load balancing regulation method based on energy storage optimization, comprising:

[0044] S100: Obtain the current operating parameters of the energy storage device, construct a multi-dimensional response feature mapping system, and determine the response boundary of the energy storage device under the current operating conditions by combining the response capability vector space with the combined threshold template.

[0045] S200: By monitoring load changes on dual time scales, it performs real-time calculation and trend judgment on the rate of change of grid load, generates a load surge warning signal, and performs cross-node collaborative scheduling based on the load surge warning signal, thereby obtaining distributed energy allocation of multiple energy storage stations through optimized allocation.

[0046] S300: During the energy storage node regulation process, based on the correlation analysis between environmental parameters and load characteristics, seasonal adaptive regulation is performed, and the seasonal switching threshold value in the dispatch command is dynamically adjusted.

[0047] S400: During the scheduling execution process, response deviation is detected through scheduling execution monitoring, scheduling deviation segments are identified and marked, and the scheduling parameter set and instruction generation logic are updated through adaptive learning using scheduling deviation segments as input, so as to obtain closed-loop optimization of scheduling logic.

[0048] It should be noted that the response capability of energy storage devices dynamically changes with parameters such as state of charge, cell temperature, and charge / discharge cycle count. Traditional fixed threshold triggering methods cannot adapt to this dynamism, leading to a mismatch between dispatch commands and the actual capacity of the equipment. Secondly, the grid load exhibits complex fluctuation characteristics during seasonal transitions. Single-time-scale monitoring methods are insufficient to identify load abrupt changes in advance, making it difficult to reserve sufficient response preparation time for energy storage devices. Thirdly, when multiple energy storage nodes are coordinated for dispatch, existing equal-distribution strategies or static priority methods fail to comprehensively consider the remaining capacity, response efficiency, and regional redundancy of each node, resulting in low overall regulation efficiency. Furthermore, the seasonal coupling relationship between temperature changes and load fluctuations is not fully utilized, and fixed seasonal switching thresholds cannot adapt to actual climate change trends, leading to discontinuous strategy switching and regulation instability. Finally, the lack of closed-loop feedback and adaptive optimization capabilities based on historical operating data prevents continuous improvement with accumulated operating experience.

[0049] Therefore, to address the aforementioned issues such as inaccurate response capability assessment, untimely load warnings, unreasonable multi-node collaborative allocation, poor seasonal adaptability, and lack of adaptive optimization capabilities, the following steps from S100 to S400 are employed: First, a multi-dimensional response feature mapping system is constructed. Through the linkage processing of a five-dimensional response capability vector space and a combined threshold template, precise quantification and hierarchical control of the energy storage device's adjustment capability under different operating conditions are achieved. Second, dual-timescale load change monitoring is adopted. Short-term sliding windows capture load abrupt changes, while long-term sliding windows extract trend changes. Combined with joint triggering criteria, dynamic identification and control of seasonal load changes in the power grid are realized. Early warning is provided; secondly, a dynamic weight vector is constructed based on remaining capacity margin, unit capacity response efficiency, and adjacent node redundancy ratio to achieve precise coordinated scheduling of multiple energy storage stations; then, through cross-analysis of temperature moving average sequence and day-night load ratio sequence, a seasonal trend index vector is constructed and the seasonal switching threshold is dynamically adjusted to avoid strategy abrupt changes and regulation failures caused by temperature changes and day-night fluctuations; finally, through response deviation identification and closed-loop scheduling logic update mechanism, parameter configuration and instruction generation logic can be optimized based on actual response after each scheduling execution, thus constructing an energy storage system load regulation platform with self-learning, self-adaptation, and self-evolution capabilities.

[0050] Example 2, refer to Figures 1-3 This is one embodiment of the present invention. Based on the previous embodiment, a method for seasonal load balancing regulation of the power grid based on energy storage optimization is provided, including: obtaining the current operating parameters of the energy storage device in step S100, constructing a multi-dimensional response feature mapping system, and determining the response boundary of the energy storage device under the current operating conditions by combining the response capability vector space with the combined threshold template. Step S100 includes the following steps A1-A3:

[0051] A1: Obtaining the current operating parameters of the energy storage device includes obtaining the energy storage device's state of charge, cell temperature distribution, output voltage fluctuation range, output current change rate, and historical discharge capacity as operating parameters.

[0052] Specifically, multiple sensors are deployed in each energy storage unit of the energy storage station to achieve real-time acquisition of operating parameters. The state of charge (SOC) is acquired using a bidirectional current integration method. This method monitors the charging and discharging current in real time using a high-precision current sensor, performing continuous integration calculations with a sampling period of 1 second. The SOC resolution of a single energy storage unit reaches 0.5%, ensuring accurate control of the battery's state of charge. Cell temperature distribution is acquired using an embedded multi-point thermistor array. Each energy storage unit has 16 temperature acquisition points evenly distributed across different locations within the cell, including the central region, edge regions, and electrode connections. Temperature values ​​are collected once per second, forming a temperature distribution matrix that reflects the internal temperature gradient distribution of the cell.

[0053] The measurement of output voltage fluctuation range is achieved through a grid-side digital voltage monitoring chip, which measures voltage with millisecond-level sampling resolution. The voltage measurement range covers 0 to 1000V, and the measurement accuracy reaches ±0.2%, accurately capturing the transient voltage changes of the energy storage unit during charging and discharging. The acquisition of the output current change rate is achieved through a combination of a high-precision current sensor and an FPGA (Field-Programmable Gate Array) processor. The current sensor measures the output current value in real time, and the FPGA processor performs differential calculations on the current data every second to calculate the current derivative, i.e., the current change rate. The calculation accuracy of this rate reaches 0.1A / ms, which can reflect the dynamic response speed of the energy storage unit to power regulation commands.

[0054] Historical discharge capacity is acquired by a data logger built into the control center. This logger retrieves charge and discharge data from the past 90 days on an hourly basis, including the start and end times of each charge and discharge cycle, charge and discharge current, voltage variation curves, and cumulative charge. The data is then processed using an error correction algorithm to eliminate the effects of temperature, self-discharge, and measurement errors, ultimately generating a capacity decay curve that reflects the energy storage unit's capacity degradation trend. This curve visually reflects the changing health status of the energy storage unit during long-term operation.

[0055] A2: Construct a multi-dimensional response feature mapping system, including inputting the running parameters into the response capability evaluation matrix for weighted calculation, and selecting an appropriate boundary threshold template;

[0056] Construct a response capability vector space with operating parameters as coordinate axes, and calibrate the minimum adjustment cycle and maximum power change amplitude corresponding to the current equipment state in this vector space as response boundaries;

[0057] By matching the current state point in the vector space with the historical operating trajectory, the current response level of the energy storage device is automatically classified.

[0058] Specifically, the method for establishing the five-dimensional response capability vector space uses state of charge, cell temperature, voltage fluctuation, current change rate, and discharge capacity as five orthogonal coordinate axes, respectively... , , , , To eliminate the influence of different dimensions of the parameters on the spatial projection results, each parameter is first normalized using the following standardized formula:

[0059]

[0060] In the formula, This represents the original measured value of the j-th parameter; and These are the minimum and maximum values ​​of the parameter in historical samples or the current period, respectively. The normalized value of the j-th parameter is 0 to 1.

[0061] The five obtained after normalization Corresponding to:

[0062] State of charge (SOC); Cell temperature; Output voltage fluctuation range; : Rate of change of output current; Historical discharge capacity;

[0063] Subsequently, a standardized coordinate system is constructed, and the five normalized parameters can be mapped to a five-dimensional response capability vector space through a transformation matrix, M, with vector points V:

[0064] V=M·X;

[0065]

[0066] in, This is the transpose operator. M is a 5×5 identity matrix or a tuned characteristic matrix;

[0067] The formula for calculating the position of a vector point in five-dimensional space is:

[0068]

[0069] in, These are the first-dimensional coordinate values, corresponding to the mapped coordinates of the charged state in vector space; The second-dimensional coordinate value corresponds to the mapped coordinates of the cell temperature in vector space; The third-dimensional coordinate value corresponds to the mapped coordinates of the voltage fluctuation in vector space; The fourth dimension coordinate value corresponds to the mapped coordinates of the current change rate in vector space; The fifth dimension coordinate value corresponds to the mapping coordinates of the historical discharge capacity in vector space.

[0070] Each This represents the coordinate value of the corresponding parameter in the standardized coordinate system, and the Euclidean distance is used as the similarity score between the response vector point and the template reference value. The formula is:

[0071]

[0072] In the formula, This represents the score value of the j-th dimension in the reference template (i.e., the standard response value under this working condition in historical experience or model). This represents the Euclidean distance between the current response vector point and the template reference value; a smaller distance indicates a higher similarity.

[0073] Finally, based on the comparison results between distance d and similarity threshold, the closest template center value is selected to extract the minimum adjustment cycle and maximum power change range under the current operating state.

[0074] A3: The response capability assessment matrix adopts a weighted calculation, sets weight coefficients for multiple operating parameters, and introduces external ambient temperature, scheduling cycle load change trend and battery charge and discharge cycle number to adjust the weight coefficients;

[0075] The boundary threshold template is pre-classified and constructed according to temperature range, load change rate range and battery cycle aging level. The evaluation reference template is selected by calculating the distance between the current feature combination and the central feature value of the template.

[0076] Specifically, different weighting coefficients are set for state of charge, cell temperature distribution, output voltage fluctuation range, output current change rate, and historical discharge capacity. For example, under the current high-temperature summer conditions, the weight of cell temperature distribution is set to 0.35, the weight of state of charge is 0.25, the weight of output current change rate is 0.2, the weight of voltage fluctuation range is 0.1, and the weight of historical discharge capacity is 0.1. The weights are then adjusted by combining the external ambient temperature, the load change trend in the current scheduling cycle, and the number of battery charge and discharge cycles, so that the evaluation results can dynamically perceive environmental factors.

[0077] The comprehensive response score of each parameter under the current state is calculated by a linear weighting function. For example, when a certain unit has a state of charge of 72%, an average cell temperature of 45℃, an output current change rate of 2.5A / ms, and a historical discharge capacity decay of 8%, the calculated comprehensive response score is 0.67, indicating that the current response capability of the energy storage unit is at a medium level.

[0078] Subsequently, based on the above comprehensive response score, the feature vector of the current energy storage unit is matched with the pre-constructed combined threshold template;

[0079] The combined threshold template is pre-classified and constructed according to three features: temperature range, load change rate range, and battery cycle aging level. Each feature is set with multiple range segments. For example, the temperature range is set to three segments: <20℃, 20-35℃, and >35℃; the load change rate range is set to three segments: <0.1, 0.1-0.3, and >0.3; and the battery cycle aging level is divided into three levels: mild, moderate, and severe.

[0080] For the current comprehensive score feature vector, the template closest to the template center feature value is selected by calculating the Euclidean distance. Let the minimum Euclidean distance obtained be 0.18, and the corresponding template recommends a minimum adjustment period of 35 seconds and a maximum power change range of 200kW.

[0081] Meanwhile, the five operating parameters are normalized and then used as five orthogonal coordinate axes as inputs to construct a five-dimensional response capability vector space, and the response capability vector points corresponding to the current device status are generated in this vector space.

[0082] Finally, the above response capability vector points are projected onto the reference value coordinates in the combined threshold template. The template is generated by combining the temperature grading threshold template, the load gradient threshold template, and the cyclic aging threshold template as a three-dimensional index. Its reference values ​​include the average minimum adjustment cycle and the average maximum controllable power change amplitude obtained by statistical analysis of the template's historical samples under this operating condition category.

[0083] By calculating the spatial distance between vector points and template reference values, the minimum adjustment cycle and maximum power change range of the current operating state under the template are extracted. For example, the projection results show that the minimum adjustment cycle of the energy storage unit is 33 seconds, the maximum controllable power change range is 195kW, and the similarity score is 0.91, indicating a high degree of matching with the template. The current coordinate points in the vector space are matched with the historical operating trajectories to automatically classify the current response level of the energy storage device, such as classifying it as "Level B response level".

[0084] Registering the response level label in the scheduling logic limits the upper and lower boundaries of the adjustment strategies it can participate in during the current time period, thus avoiding scheduling failures or overload risks caused by response capability mismatch.

[0085] In this embodiment, step S200 involves monitoring load changes at dual time scales to perform real-time calculations and trend judgments on the grid load change rate, generating a load surge warning signal. Based on the load surge warning signal, cross-node collaborative scheduling is performed, and distributed energy allocation of multiple energy storage stations is obtained through optimized allocation. This includes the following steps B1-B2:

[0086] B1: Load change monitoring through dual time scales includes constructing continuous time-series load data, generating trigger thresholds by combining response boundaries, and setting recovery thresholds based on the trigger thresholds to form a dual threshold set;

[0087] Set up short-time sliding time windows and long-time sliding time windows to be executed in parallel. The short-time sliding time window is used to detect abrupt changes in the load change rate, while the long-time sliding time window is used to extract the load change trend.

[0088] When the load change rate exceeds the trigger threshold within the short-term sliding time window, and the average change rate within the long-term sliding time window meets the trend judgment condition, the advance response time is estimated based on the current change rate and compared with the equipment adjustment capability.

[0089] A load surge warning signal is generated when the criterion is met.

[0090] Specifically, to determine whether the trigger threshold is met, a short-term sliding time window and a long-term sliding time window are set to be executed in parallel.

[0091] A short-time sliding window is used to detect the abrupt slope of the load change rate.

[0092] Long-term sliding time windows are used to extract load change trends. The average value of the rate of change is calculated in two time windows, and a joint triggering criterion is constructed.

[0093] When the load change rate within the short-term sliding time window exceeds the trigger threshold, and the average change rate within the long-term sliding time window is non-negative, the advance response time required to reach the response boundary is further estimated based on the current change rate, and it is determined whether the advance response time is less than the minimum adjustment cycle in step one, which serves as the final basis for whether to trigger the load surge warning signal.

[0094] Specifically, the short-time sliding window length is denoted as... It is used to capture the sudden slope of load changes. In this embodiment, it is set to a 1-5 minute interval to quickly track the load change trend with a second-level resolution.

[0095] Long-term sliding time window is denoted as This is used to extract stable trend change trajectories. In this embodiment, it is set to a 15-30 minute interval and dynamically adjusted according to the standard deviation of historical load fluctuations during daily operation.

[0096] Let the current load time series be P(t), then:

[0097] Short-time rate of change The calculation formula is:

[0098]

[0099] This represents the length of the short-time sliding window. Let be the short-term load change rate at time t.

[0100] Long-term average rate of change The calculation formula is:

[0101]

[0102] This represents the number of short-term sampling points contained within a long-term sliding window. Used as a circular index variable; Let be the long-term average rate of change at time t.

[0103] Where N= / for , which is the number of short-time sampling points contained within the sliding window;

[0104] Let the load change rate trigger threshold and the long-term window average change rate judgment lower limit be respectively... and ,but:

[0105] like ≥ ,and ≥ If so, it is determined to be a sudden upward trend;

[0106] After satisfying the joint triggering criterion, further based on the current short-term rate of change The formula for calculating the prediction time is: [Formula for predicting the time it will take for the load to reach the regulation response boundary].

[0107]

[0108] In the formula, This represents the response power threshold set by the scheduling system; Let t be the time predicted to reach the response boundary at time t.

[0109] Forecast time Minimum adjustment cycle of equipment Comparison:

[0110] like ≤ This will trigger a load surge warning signal;

[0111] B2: Perform cross-node collaborative scheduling, including acquiring load surge warning signals, extracting response capability level tags of each energy storage node, and filtering out a set of target energy storage nodes with adjustable capabilities.

[0112] Perform a regulation capability assessment on the target energy storage node set and construct a comprehensive score, then construct an allocation weight vector based on the comprehensive score;

[0113] By combining the weight vector and the target response time period in the early warning signal, a unified scheduling instruction table is generated and sent to the target energy storage node;

[0114] During the scheduling process, the execution status of each node is monitored, and a replacement allocation is triggered when an abnormal node response is detected.

[0115] Specifically, when assessing the regulation capacity of the target energy storage node set, a comprehensive score is constructed from three metrics (remaining capacity margin, unit capacity response efficiency, and adjacent node redundancy ratio).

[0116]

[0117] In the formula, , , These are the weighting coefficients for the three scores, satisfying... + + =1; Scoring the remaining capacity margin; Scoring the response efficiency per unit capacity; Scoring the redundancy ratio of adjacent nodes; Let i be the overall score of node i.

[0118] The formula for calculating the weight vector and the final power allocation weight vector. The calculation formula is:

[0119]

[0120] In the formula, Ω represents the set of target energy storage nodes; This represents the overall score of the j-th node in the set. Assign a weight to the power of node i, which represents the proportion of the target power that node i should bear to the total target power.

[0121] In this embodiment, during the energy storage node adjustment process in step S300, seasonal adaptive adjustment is performed based on the correlation analysis between environmental parameters and load characteristics, and the seasonal switching threshold value in the dispatch command is dynamically adjusted, including the following steps C1-C2:

[0122] C1: Based on the correlation analysis between environmental parameters and load characteristics, this includes continuously collecting external temperature data for a preset time period and diurnal load data within the corresponding daily cycle; calculating the diurnal load ratio based on the diurnal load data according to the daily cycle, and calculating the moving average of temperature using a sliding time window; aligning the temperature moving average sequence and the diurnal load ratio sequence by timestamp to construct a bivariate time series matrix and calculating the linear correlation coefficient; when the linear correlation coefficient exceeds a preset coupling threshold, extracting the directional parameter of the diurnal load ratio changing with the temperature moving average, and constructing a seasonal trend index vector.

[0123] Specifically, seasonal trend analysis is the core component of achieving seasonal adaptive regulation. This involves continuously collecting external temperature data for a preset time period (e.g., 30 consecutive days) and corresponding diurnal load data within that daily cycle. External temperature data is provided by meteorological monitoring stations deployed around the energy storage station, with sampling occurring hourly. The data includes real-time temperature values, daily maximum and minimum temperatures, and temperature trends. Diurnal load data is provided by the power grid dispatch center, recording the total power grid load value with a 15-minute sampling period, divided into daytime load (6:00 to 18:00) and nighttime load (18:00 to 6:00 the next day).

[0124] The daytime and nighttime load data are statistically processed according to the daily cycle to calculate the daily average daytime load value and the nighttime average load value. Then, the daytime and nighttime load ratio is calculated, which is the ratio of the average daytime load value to the average nighttime load value.

[0125] The temperature moving average series and the day-night load ratio series are precisely aligned by timestamp to ensure a one-to-one correspondence between each temperature moving average and the corresponding day-night load ratio, constructing a bivariate time series matrix. The rows of this matrix represent the time series, and the columns represent the temperature moving average and the day-night load ratio, respectively. Each element of the matrix records the temperature and load characteristics for a specific date. Statistical analysis is performed on this bivariate time series matrix to calculate the linear correlation coefficient between the temperature moving average series and the day-night load ratio series. The linear correlation coefficient is calculated using the Pearson correlation coefficient formula. First, the means of both series are calculated. Then, the deviation of each data point from the mean is calculated. The products of the deviations at corresponding positions in the two series are summed to obtain the centered covariance. Finally, this covariance is divided by the product of the standard deviations of the two series to obtain the linear correlation coefficient. The coefficient ranges from -1 to 1, with positive values ​​indicating a positive correlation (the day-night load ratio increases with rising temperatures) and negative values ​​indicating a negative correlation. The closer the absolute value is to 1, the stronger the correlation.

[0126] The linear correlation coefficient between the moving average temperature series and the diurnal load ratio series was calculated using the Pearson correlation coefficient:

[0127]

[0128] In the formula, Let be the moving average of the temperature on day i. The day-night load ratio for day i; This represents the average value of the temperature series. is the average value of the load ratio sequence; n is the number of statistical days; r is the linear correlation coefficient, ranging from -1 to 1.

[0129] When the linear correlation coefficient exceeds a preset coupling threshold (e.g., 0.7), a linear relationship between temperature and the day-night load ratio is determined, and the directional parameter of the trend of the day-night load ratio changing with the moving average of temperature is extracted. A linear regression is performed on the bivariate time series matrix, with the moving average of temperature as the independent variable and the day-night load ratio as the dependent variable, to fit an optimal fitted line. The slope of this line is the trend direction parameter. If the slope is positive, it indicates that rising temperature leads to an increase in the day-night load ratio, i.e., a pronounced summer characteristic; if the slope is negative, it indicates that falling temperature leads to an increase in the day-night load ratio, i.e., a pronounced winter heating characteristic.

[0130] C2: Dynamically adjusting the seasonal switching threshold value in the scheduling instruction includes comparing the seasonal switching threshold value in the scheduling instruction with the seasonal trend indicator vector, and calculating the deviation value of the trend direction and trend strength in each threshold type.

[0131] For threshold types where the deviation value exceeds the preset offset threshold, a dynamic correction coefficient is determined based on the deviation value. The corresponding seasonal switching threshold value is then corrected based on the dynamic correction coefficient, generating a corrected threshold set and replacing the seasonal switching threshold value in the original scheduling instruction.

[0132] Specifically, the dynamic adjustment of seasonal switching thresholds is a key step in achieving seasonal adaptive regulation. First, seasonal switching thresholds are extracted from the currently executed dispatch instructions. These thresholds include temperature range thresholds (e.g., a temperature threshold of 30℃ for initiating the summer high-temperature mode and 5℃ for initiating the winter low-temperature mode), day-night load ratio thresholds (e.g., a day-night load ratio greater than 1.5 for the summer mode and less than 1.2 for the winter mode), and load change rate thresholds (e.g., a load change rate threshold of 10kW / min for the summer mode and 8kW / min for the winter mode). These thresholds are then compared and analyzed with the seasonal trend index vector constructed in step C1.

[0133] For each threshold type (temperature range threshold, day-night load ratio threshold, and load change rate threshold), a deviation value is calculated. For threshold types where the deviation value exceeds a preset offset threshold (e.g., temperature threshold deviation exceeds 5℃, day-night load ratio threshold deviation exceeds 0.3, and load change rate threshold deviation exceeds 2kW / min), a threshold adjustment process is initiated. The adjustment process sets a dynamic correction coefficient based on the magnitude of the deviation value; the larger the deviation value, the larger the dynamic correction coefficient and the larger the threshold adjustment range. Simultaneously, a comprehensive adjustment is made by combining the trend confidence and trend strength in the seasonal trend indicator vector.

[0134] The formula for calculating the dynamic correction coefficient is: Correction coefficient = Basic correction coefficient × Trend confidence level × Trend strength weight. The basic correction coefficient is set based on the magnitude of the deviation; the larger the deviation, the larger the basic correction coefficient. The trend confidence level comes from the seasonal trend indicator vector. The trend strength weight is determined based on the ratio of the actual trend strength (i.e., the absolute value of the slope of the linear regression of temperature and diurnal load ratio) to the historical average trend strength.

[0135] The seasonal switching threshold is adjusted based on a dynamic correction factor:

[0136]

[0137] In the formula, This is the original seasonal switching threshold value; Δ is the dynamic correction factor; Δ is the deviation value. This is the corrected seasonal switching threshold value. The dynamic correction coefficient is included. It is calculated based on trend confidence and trend strength.

[0138] For example, if the current summer high temperature mode starts at a temperature threshold of 30℃, but the seasonal trend analysis shows that the actual summer characteristics appear earlier and are more intense, with a deviation of -5℃ (meaning that the summer mode should be started 5℃ earlier), and the dynamic correction coefficient is calculated to be 0.8, then the corrected temperature threshold is 30℃ + 0.8 × (-5℃) = 26℃, that is, the summer mode start temperature threshold is lowered to 26℃.

[0139] Perform the above correction calculations on all threshold types that require adjustment to generate a corrected threshold set. This set includes all updated temperature range thresholds, day-night load ratio thresholds, load change rate thresholds, and other relevant seasonal scheduling parameters. Write the corrected threshold set into the scheduling instruction configuration file, replacing the seasonal switching threshold values ​​in the original scheduling instructions, so that subsequent scheduling decisions are executed based on the updated threshold values.

[0140] In this embodiment, during the scheduling execution process in step S400, response deviation detection is performed through scheduling execution monitoring, scheduling deviation segments are identified and marked, and the scheduling parameter set and instruction generation logic are updated through adaptive learning using scheduling deviation segments as input to obtain closed-loop optimization of the scheduling logic, including the following steps D1-D3:

[0141] D1: Response deviation detection is performed through scheduling execution monitoring, including...

[0142] Extract the energy storage load response commands and the actual power output timing data of the corresponding energy storage nodes within the current execution cycle;

[0143] Align the scheduling target curve with the actual response curve to divide the instruction execution into multiple instruction execution segments, and construct a segment comparison dataset;

[0144] For each scheduling segment, the changing trends of the target response curve and the actual response curve are calculated, and the differences in power values ​​are compared.

[0145] When the power deviation of multiple consecutive sampling points exceeds the set threshold, and the trend direction is opposite or the delay time exceeds the equipment's adjustment capability, the segment is identified as a scheduling deviation segment and marked.

[0146] Specifically, the actual power output timing data of the corresponding energy storage node during command execution is obtained from the data acquisition terminal of the energy storage device. This data is recorded with a sampling period of 30 seconds and includes information such as timestamp, actual output power value, power change rate and node operating status identifier.

[0147] For each scheduling segment, a comparative analysis of execution behavior trends is conducted.

[0148] First-order difference operations are performed on the target scheduling curve and the actual response curve to calculate the power change rate at each sampling point, which is the power difference between adjacent sampling points divided by the time interval, resulting in the target power change rate sequence and the actual power change rate sequence. By comparing the signs of the two sequences (positive values ​​indicate power increase, negative values ​​indicate power decrease), it is determined whether the direction of change in the actual response is consistent with the target. If, within a certain scheduling segment, the signs of the actual power change rate at multiple consecutive sampling points (e.g., four consecutive sampling points, i.e., 2 minutes) are opposite to the signs of the target power change rate, then an anomaly with an opposite trend direction is determined for that segment.

[0149] By comparing the power values ​​of the target scheduling curve and the actual response curve at the same time point, the power deviation value is calculated, which is the difference between the actual power output and the target power value. A power deviation threshold is set at 10% of the node's maximum response capability. For example, if the maximum controllable power variation of the node is 200kW, then the power deviation threshold is 20kW. If, within a certain scheduling segment, the absolute value of the power deviation at multiple consecutive sampling points (e.g., 5 consecutive sampling points, i.e., 2.5 minutes) exceeds the power deviation threshold, it is preliminarily determined that there is an anomaly of insufficient response amplitude in that segment.

[0150] Analyze the response delay. Identify the point in the target scheduling curve where power begins to change (i.e., the start time in the command), and find the point in the actual response curve where power first changes (identified by determining when the absolute value of the power change rate first exceeds the preset start threshold). The time difference between these two points is the response delay. Compare the response delay with the minimum adjustment period of that node. If the response delay exceeds 50% of the minimum adjustment period (e.g., if the minimum adjustment period is 33 seconds, then the delay threshold is 16.5 seconds), an anomaly in response delay is determined for that segment.

[0151] Taking into account the three types of abnormal situations mentioned above, a joint judgment rule is established: when a certain scheduling segment simultaneously meets both conditions of "power deviation of multiple consecutive sampling points exceeding the threshold" and "opposite trend direction or response delay time exceeding 50% of the minimum adjustment cycle," the segment is judged as a scheduling deviation segment and labeled "scheduling deviation." This label is attached to the recording information of the scheduling segment, indicating that the scheduling execution quality of the segment is substandard and needs to enter the subsequent deviation analysis and parameter optimization process.

[0152] D2: When marking scheduling deviation segments, the scheduling deviation segments are classified into response delay type deviation, response missing type deviation, and reverse response type deviation. The deviation segments are weighted and quantified by four dimensions: maximum power deviation value, deviation duration, trend conflict ratio, and delay duration. The deviation segments are divided into three levels: mild, moderate, and severe according to the scoring results.

[0153] Specifically, after identifying scheduling deviation segments, each segment undergoes in-depth analysis and classification. First, the abnormal features of the scheduling deviation segments are extracted, and the deviation pattern of the actual response curve relative to the target scheduling curve is analyzed. Based on the different deviation patterns, the scheduling deviation segments are classified into three typical types: response delay deviation, response missing deviation, and reverse response deviation.

[0154] Response delay-type deviation refers to a significant lag in the initiation of a response by an energy storage node after receiving a dispatch command, but the final response direction is correct and the dispatch target is partially or fully achieved. The identifying characteristics of this type of deviation are: the response initiation time is delayed by more than 50% of the node's minimum adjustment cycle relative to the command start time; however, after the delayed initiation, the actual response curve's trend is consistent with the target dispatch curve, and the power gradually approaches the target value. The delay duration of this type of deviation (the time difference between the command start time and the actual response initiation time) is calculated as a quantitative indicator.

[0155] Response deficiency bias refers to a situation where an energy storage node fails to generate significant power output changes during the dispatch cycle, or the magnitude of the power output change is far below the target requirement. The identifying characteristics of this type of bias are: the actual power output curve is approximately flat throughout the dispatch segment (the absolute value of the power change rate remains below the activation threshold), or although there is power change, the cumulative response power is less than 30% of the target power. The ratio of the actual response power to the target power is calculated as an indicator of response completion; segments with a response completion rate below 30% are classified as response deficiency bias.

[0156] Reverse response bias refers to the situation where the power change direction of an energy storage node during a scheduling cycle is opposite to the direction required by the command. For example, the command may require discharging, but the node actually charges, or vice versa. The identifying characteristics of this type of bias are: the sign of the power change rate of the actual response curve is opposite to the sign of the power change rate of the target scheduling curve for most of the scheduling period (more than 60% of the sampling points), and the relative positions of the two curves on the power coordinate axis show a diverging trend. The percentage of trend conflict, i.e., the percentage of sampling points with opposite signs out of the total number of sampling points, is calculated as a quantitative indicator of reverse response.

[0157] After classifying the types, each scheduling deviation segment is quantitatively scored in multiple dimensions. The scoring system includes four dimensions: First, the maximum power deviation value, which is the absolute value of the maximum deviation between the actual power output and the target power value within the entire scheduling segment. This indicator reflects the severity of the peak deviation. Second, the deviation duration, which is the cumulative duration for which the power deviation exceeds the threshold. This indicator reflects the persistence of the deviation. Third, the trend conflict ratio, which is the proportion of sampling points where the actual response trend is opposite to the target trend. This indicator reflects the degree of directional error of the deviation. Fourth, the delay duration, which is the delay between the response start time and the instruction start time. This indicator reflects the timeliness of the response.

[0158] The four dimensions of indicators are standardized, converting the original values ​​into normalized scores between 0 and 1. The standardization method uses min-maximum normalization, which involves subtracting the historical minimum value for each dimension from the original value and then dividing by the difference between the historical maximum and minimum values. Weighting coefficients are assigned to the four dimensions; for example, maximum power deviation is weighted at 0.3, deviation duration at 0.3, trend conflict percentage at 0.25, and delay at 0.15. These weighting coefficients reflect the varying importance of different dimensions to scheduling quality. The normalized scores for the four dimensions are multiplied by their corresponding weighting coefficients and then summed to obtain a comprehensive weighted quantitative score for the scheduling deviation segment. A higher score indicates a more severe deviation.

[0159] Multi-dimensional quantitative scoring of scheduling deviation segments:

[0160]

[0161] In the formula, The weight coefficient for the k-th dimension; The standardized score for the kth dimension (including four dimensions: maximum power deviation, deviation duration, trend conflict ratio, and delay duration). The score is a comprehensive weighted quantitative score, ranging from 0 to 1. A higher score indicates a more severe deviation.

[0162] Based on the overall score, scheduling deviation segments are categorized into three severity levels: minor deviation (overall score between 0 and 0.4), moderate deviation (overall score between 0.4 and 0.7), and severe deviation (overall score between 0.7 and 1). Each scheduling deviation segment is labeled with a severity level, and the type label (delayed response, missing response, reverse response), severity level label, overall score, and the original values ​​of the four dimensions are recorded together in the structured data record of the deviation segment. This detailed labeling information provides precise problem localization and improvement guidance for subsequent scheduling parameter optimization and strategy adjustment.

[0163] D3: Update scheduling parameters through adaptive learning, including classifying and organizing scheduling deviation segments according to time period, energy storage node identifier, and response capability level label to construct a training sample set;

[0164] The correspondence between the trigger thresholds corresponding to scheduling deviation segments and the actual response behavior is extracted from the training sample set. Based on the clustering results of response capability similarity, a mapping model between deviation scenarios and failure factors is constructed.

[0165] The out-of-limit factor of the original response threshold is derived by using the mapping model, the optimal threshold range is refitted, and the seasonal combined threshold template is updated.

[0166] Specifically, all scheduling deviation segments identified and marked in step D2 are used as the main input data source and classified and organized according to three dimensions: The first dimension is time period, in which scheduling deviation segments are divided into different time periods according to their occurrence time, such as by hour, by day, and by week, in order to analyze the time distribution pattern of deviation occurrence; The second dimension is energy storage node identification, in which scheduling deviation segments are grouped according to the energy storage nodes involved, and the frequency and severity of deviation occurrence of each node are counted to identify high-frequency failure nodes; The third dimension is response capability level label, in which scheduling deviation segments are grouped according to the response capability level of the nodes (Level A, Level B, Level C, etc.) to analyze the scheduling failure characteristics of nodes with different response levels.

[0167] The data after 3D classification is used to construct a training sample set. Each training sample contains the following fields: scheduling time period identifier (start and end time), energy storage node ID (unique number), response capability level label (A / B / C level), target power change slope (kW / min), actual response curve data (power time series array), response delay duration (seconds), failure type label (delay / missing / reverse), comprehensive score value (0 to 1), environmental status label (including temperature, day-night load ratio, load change rate, etc.), and deviation numerical feature vector (a four-dimensional vector containing maximum power deviation, deviation duration, trend conflict ratio, and delay duration).

[0168] The correspondence between the trigger thresholds and actual response behaviors corresponding to scheduling deviation segments is extracted from the training sample set. Specifically, the scheduling parameter configuration at the time each deviation segment occurred is traced back, including the load change rate trigger threshold, power deviation threshold, response delay threshold, and seasonal switching threshold, etc., which were in effect at that time. These parameter values ​​are used as input features; the failure type, comprehensive score, and whether a substitute allocation was triggered for the deviation segment are used as output labels. A mapping matrix between the input features and output labels is constructed, reflecting the probability and severity of scheduling deviations under specific parameter configurations.

[0169] Based on the historical performance of energy storage nodes in the training sample set, a dynamic scoring matrix for node response capability is constructed.

[0170] Dynamic scoring of node response capabilities based on historical performance:

[0171]

[0172] In the formula, This represents the number of successful responses from node i. Let i be the total number of scheduling attempts for node i. This represents the average response delay; Maximum allowable delay; This represents the average power deviation. Power deviation threshold; This is the number of times a substitute can be triggered; to These are the weighting coefficients for each evaluation dimension; The comprehensive dynamic score for node i ranges from 0 to 1.

[0173] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for seasonal load balancing regulation of power grids based on energy storage optimization, characterized in that: This includes obtaining the current operating parameters of the energy storage device, constructing a multi-dimensional response feature mapping system, and determining the response boundary of the energy storage device under the current operating conditions by combining the response capability vector space with the combined threshold template. By monitoring load changes on dual time scales, the power grid load change rate is calculated and trend is judged in real time, and a load surge warning signal is generated. Based on the load surge warning signal, cross-node collaborative scheduling is performed, and distributed energy allocation of multiple energy storage stations is obtained through optimized allocation. During the energy storage node regulation process, seasonal adaptive regulation is performed based on the correlation analysis between environmental parameters and load characteristics, and the seasonal switching threshold value in the dispatching instructions is dynamically adjusted. During the scheduling execution process, response deviations are detected through scheduling execution monitoring, scheduling deviation segments are identified and marked, and scheduling deviation segments are used as input to update the scheduling parameter set and instruction generation logic through adaptive learning, thereby obtaining closed-loop optimization of the scheduling logic.

2. The method for seasonal load balancing regulation of power grids based on energy storage optimization as described in claim 1, characterized in that: The acquisition of the current operating parameters of the energy storage device includes acquiring the energy storage device's state of charge, cell temperature distribution, output voltage fluctuation range, output current change rate, and historical discharge capacity as operating parameters. The construction of the multidimensional response feature mapping system includes inputting the running parameters into the response capability evaluation matrix for weighted calculation and selecting an appropriate boundary threshold template. Construct a response capability vector space with operating parameters as coordinate axes, and calibrate the minimum adjustment cycle and maximum power change amplitude corresponding to the current equipment state in this vector space as response boundaries; By matching the current state point in the vector space with the historical operating trajectory, the current response level of the energy storage device is automatically classified.

3. The method for seasonal load balancing regulation of power grids based on energy storage optimization as described in claim 2, characterized in that: The response capability evaluation matrix adopts a weighted calculation, sets weight coefficients for multiple operating parameters, and introduces external ambient temperature, scheduling cycle load change trend and battery charge and discharge cycle number to adjust the weight coefficients; The boundary threshold template is pre-classified and constructed according to temperature range, load change rate range and battery cycle aging level. An evaluation reference template is selected by calculating the distance between the current feature combination and the central feature value of the template.

4. The method for seasonal load balancing regulation of power grids based on energy storage optimization as described in claim 3, characterized in that: The load change monitoring through dual time scales includes: constructing continuous time-series load data, generating a trigger threshold by combining the response boundary, and setting a recovery threshold based on the trigger threshold to form a dual threshold set; A short-term sliding time window and a long-term sliding time window are set to be executed in parallel. The short-term sliding time window is used to detect abrupt changes in the load change rate, and the long-term sliding time window is used to extract the load change trend. When the load change rate exceeds the trigger threshold within the short-term sliding time window, and the average change rate within the long-term sliding time window meets the trend judgment condition, the advance response time is estimated based on the current change rate and compared with the equipment adjustment capability. The load surge warning signal is generated when the criterion is met.

5. A method for seasonal load balancing regulation of power grids based on energy storage optimization as described in claim 4, characterized in that: The execution of cross-node collaborative scheduling includes acquiring load surge warning signals, extracting response capability level tags of each energy storage node, and filtering out a set of target energy storage nodes with adjustable capabilities. Perform a regulation capability assessment on the target energy storage node set and construct a comprehensive score, and construct an allocation weight vector based on the comprehensive score; By combining the weighted vector and the target response time period in the early warning signal, a unified scheduling instruction table is generated and sent to the target energy storage node; During the scheduling process, the execution status of each node is monitored, and a replacement allocation is triggered when an abnormal node response is detected.

6. The method for seasonal load balancing regulation of power grids based on energy storage optimization as described in claim 5, characterized in that: The correlation analysis based on environmental parameters and load characteristics includes continuously collecting external temperature data for a preset time period and diurnal load data within the corresponding daily cycle; The diurnal load ratio is calculated by dividing the diurnal load data into daily cycles, and the moving average of temperature is calculated using a sliding time window. A bivariate time series matrix was constructed by aligning the temperature moving average series and the day-night load ratio series according to the timestamps, and the linear correlation coefficient was calculated. When the linear correlation coefficient exceeds the preset coupling threshold, the directional parameter of the diurnal load ratio changing with the sliding mean of temperature is extracted to construct a seasonal trend index vector.

7. A method for seasonal load balancing regulation of a power grid based on energy storage optimization as described in claim 6, characterized in that: The seasonal switching threshold value in the dynamic adjustment scheduling instruction includes comparing the seasonal switching threshold value in the scheduling instruction with the seasonal trend index vector, and calculating the deviation value of the trend direction and trend intensity in each threshold type. For threshold types where the deviation value exceeds the preset offset threshold, a dynamic correction coefficient is determined based on the deviation value. The corresponding seasonal switching threshold value is then corrected based on the dynamic correction coefficient, generating a corrected threshold set and replacing the seasonal switching threshold value in the original scheduling instruction.

8. A method for seasonal load balancing regulation of power grids based on energy storage optimization as described in claim 7, characterized in that: The method of detecting response deviations through scheduling execution monitoring includes, Extract the energy storage load response commands and the actual power output timing data of the corresponding energy storage nodes within the current execution cycle; Align the scheduling target curve with the actual response curve to divide the instruction execution into multiple instruction execution segments, and construct a segment comparison dataset; For each scheduling segment, the changing trends of the target response curve and the actual response curve are calculated, and the differences in power values ​​are compared. When the power deviation of multiple consecutive sampling points exceeds the set threshold, and the trend direction is opposite or the delay time exceeds the equipment's adjustment capability, the segment is identified as a scheduling deviation segment and marked.

9. A method for seasonal load balancing regulation of a power grid based on energy storage optimization as described in claim 8, characterized in that: When marking scheduling deviation segments, the scheduling deviation segments are classified into response delay type deviation, response missing type deviation, and reverse response type deviation. The segments are weighted and quantified by four dimensions: maximum power deviation value, deviation duration, trend conflict ratio, and delay duration. The segments are divided into three levels: mild, moderate, and severe according to the scoring results.

10. A method for seasonal load balancing regulation of a power grid based on energy storage optimization as described in claim 9, characterized in that: The step of updating scheduling parameters through adaptive learning includes classifying and organizing the scheduling deviation segments according to time period, energy storage node identifier, and response capability level label to construct a training sample set; The correspondence between the trigger thresholds corresponding to scheduling deviation segments and the actual response behavior is extracted from the training sample set.

Images