Energy storage capacity configuration method and system based on multi-scale data and adaptive optimization

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an energy storage capacity configuration method and system based on multi-scale data and adaptive optimization. According to the method, historical new energy output data and load data are collected and cleaned, the data are decomposed into short-term, medium-term and long-term components according to the time scale, the difference between each scale component and a local mean value is calculated, a weight attenuation coefficient is combined, and an adaptive weight coefficient is obtained through exponential function mapping normalization; and scaling the corresponding scale component based on the time window extreme value and combining the adaptive weight coefficient to obtain multi-scale normalized data. And calculating load power shortage power, new energy power abandoning power and an energy storage life loss factor based on the normalized data, and performing weighted combination on the three to construct a comprehensive objective function. An improved whale optimization algorithm is utilized, population initialization is performed through multi-scale data statistical information, control parameters are dynamically updated, a search strategy is selected based on a population distribution state, and accurate optimization configuration of the energy storage capacity is realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method and system for configuring energy storage capacity based on multi-scale data and adaptive optimization. Background Technology

[0002] As the proportion of renewable energy sources such as wind and solar power in the power system continues to increase, their inherent intermittency and strong volatility pose a severe challenge to the stable operation of the power grid. Energy storage technology, as a key means to smooth out fluctuations and promote the consumption of new energy sources, has its capacity configuration directly affecting the economy and reliability of the system.

[0003] Existing energy storage capacity configuration methods suffer from the following shortcomings: Traditional normalization techniques rely on global statistics, making it difficult to adapt to the local mutations and multi-scale fluctuations in new energy data, resulting in low data preprocessing quality; optimization objective functions often focus on single indicators, failing to comprehensively balance multiple objectives such as new energy consumption, load guarantee, and energy storage lifetime; conventional optimization algorithms lack utilization of problem domain characteristics, easily getting trapped in local optima and exhibiting low search efficiency under high-dimensional complex constraints; and convergence criteria are too simplistic, making it difficult to accurately assess whether the optimization process has truly reached the optimum. These problems lead to over-configuration or under-configuration of energy storage systems, affecting overall economic efficiency and practicality. Summary of the Invention

[0004] This invention provides a method and system for configuring energy storage capacity based on multi-scale data and adaptive optimization, which solves the problems of inaccurate configuration and high cost of traditional methods when facing the strong volatility of new energy sources.

[0005] To achieve the above objectives, the present invention provides the following technical solution: This invention relates to an energy storage capacity configuration method based on multi-scale data and adaptive optimization, comprising: S1: Collect historical renewable energy output data and historical load data and perform cleaning processing; S2: Decompose the cleaned data into short-term, medium-term and long-term components according to time scale. Calculate the local mean of each time scale component through a sliding window. Calculate the difference between each time scale component and the corresponding local mean, and normalize it by combining the weight decay coefficient with an exponential function to obtain adaptive weight coefficients. Scale the corresponding scale components based on the extreme values ​​within the time window of each time scale, and combine them with the corresponding adaptive weight coefficients to obtain multi-scale normalized data. S3: Calculate the load power shortage, renewable energy curtailment and energy storage life loss factor based on multi-scale normalized data, and construct a comprehensive objective function by weighting the three and defining the operation constraints of the energy storage system. S4: Initialize the whale population position using the adaptive weight coefficients, local mean, and standard deviation of each time scale component at the reference time. Calculate the step size correction amount using the standard deviation of each time scale component at the reference time to dynamically update the control parameters. Calculate the global search probability and select a search strategy based on the average distance between the current population and the historical best position using an exponential function. When the relative change in the objective function value and the average distance of the population are both less than the threshold, output the best position as the energy storage capacity configuration result.

[0006] As a preferred embodiment of the present invention, the calculation steps of the adaptive weighting coefficients include: Calculate the absolute difference between each time scale component and its corresponding local mean; The negative of the product of the absolute difference and the weight decay coefficient is used as the input to the exponential function to obtain the exponential value. The adaptive weighting coefficient of each time scale component at each time point is obtained by dividing the exponent value of each time scale component by the sum of the exponent values ​​of all time scale components.

[0007] As a preferred embodiment of the present invention, the calculation steps of the multi-scale normalized data include: Obtain the maximum and minimum values ​​of the corresponding scale components within the time window of each time scale; The scaling value is obtained by dividing the difference between each time scale component and the minimum value within the corresponding time window by the difference between the maximum value and the minimum value. Multiplying the scaling value by the adaptive weighting coefficient at the corresponding time point yields the multi-scale normalized data for each time scale.

[0008] As a preferred embodiment of the present invention, the calculation steps for the load power shortage include: The difference between the load data in the multi-scale normalized data and the sum of the grid interaction power and the energy storage discharge power is compared with zero, and the larger of the two values ​​is taken as the load power shortage.

[0009] As a preferred embodiment of the present invention, the calculation steps for the abandoned power of renewable energy include: The difference between the sum of renewable energy output data and the sum of energy storage charging power and grid interaction power in the multi-scale normalized data is compared with zero, and the larger of the two values ​​is taken as the renewable energy curtailment power.

[0010] As a preferred embodiment of the present invention, the calculation steps of the energy storage lifetime loss factor include: Calculate the absolute value of the difference between the energy storage charging power and the energy storage discharging power at each time point and sum them up. Divide the summation result by the energy storage capacity to obtain the energy storage lifetime loss factor.

[0011] As a preferred embodiment of the present invention, the step of initializing the whale population location includes: Select the midpoint of the data time range as the reference time; Calculate the product of the adaptive weight coefficient of each time scale component at the reference time and the local mean, and sum them up. Divide the summation result by the maximum value of the local mean of all time scale components at the reference time to obtain the normalization factor. The standard deviation of each time scale component at the reference time is obtained by using a sliding window. The product of the standard deviation of each time scale component at the reference time and the random number is calculated and summed. The result is multiplied by the population initialization adjustment rate to obtain the individual correction increment. The initial position of each whale individual is obtained by multiplying the difference between the lower and upper bounds of the optimization variable by the normalization factor, adding the lower bound of the optimization variable, and then adding the individual correction increment.

[0012] As a preferred embodiment of the present invention, the step of dynamically updating the control parameters includes: The standard deviation of each time scale component at the reference time is calculated using a sliding window. The step size correction is obtained by summing the standard deviations of each time scale component at the reference time, dividing by the total number of time scales, and then multiplying by the step size adjustment factor. The updated control parameters are obtained by subtracting the product of the current iteration number and the difference between the initial and final control parameter values, divided by the maximum iteration number, and then adding the step size correction.

[0013] As a preferred embodiment of the present invention, the step of selecting a search strategy includes: Calculate the L2 norm of the distance between each individual whale's location and its historical best location, and then average them to obtain the population average distance. The negative of the product of the average population distance and the search balance coefficient is used as the input of the exponential function. The global search probability is obtained by adding 1 to the value of the exponential function and taking the reciprocal. A random number is generated and compared with the global search probability. When the random number is less than the global search probability, the coefficient vector is calculated based on the updated control parameters and the random vector, and the whale individual position is updated by selecting either the encirclement predation strategy or the spiral update strategy. Otherwise, the gradient of the objective function at the current individual position is calculated by the finite difference method and multiplied by the local learning rate before updating the whale individual position.

[0014] This invention also proposes an energy storage capacity configuration system based on multi-scale data and adaptive optimization, comprising: The data acquisition module is used to collect historical renewable energy output data and historical load data and perform cleaning processing. The multi-scale normalization module is used to decompose the cleaned data into short-term, medium-term and long-term components according to time scale. It calculates the local mean of each time scale component through a sliding window, calculates the difference between each time scale component and the corresponding local mean, and normalizes it by combining the weight decay coefficient through an exponential function to obtain the adaptive weight coefficient. Based on the extreme values ​​within the time window of each time scale, it scales the corresponding scale component and combines it with the corresponding adaptive weight coefficient to obtain multi-scale normalized data. The objective function construction module is used to calculate the load power shortage, renewable energy curtailment, and energy storage lifetime loss factor based on multi-scale normalized data, and to construct a comprehensive objective function by weighting and combining the three and defining the operating constraints of the energy storage system. The adaptive optimization module initializes the whale population position using the adaptive weight coefficients, local mean, and standard deviation of each time scale component at the reference time. It dynamically updates the control parameters by calculating the step size correction amount through the standard deviation of each time scale component at the reference time. Based on the average distance between the current population and the historical best position, it calculates the global search probability through an exponential function and selects a search strategy. When the relative change of the objective function value and the average distance of the population are both less than the threshold, it outputs the best position as the energy storage capacity configuration result.

[0015] The beneficial effects of this invention are: 1. This invention innovatively proposes a multi-scale normalization processing technique, which effectively solves the feature distortion problem of traditional methods when processing highly volatile new energy data by combining time scale decomposition with adaptive dynamic adjustment of weight coefficients. This technique enables the optimization model to accurately capture the fluctuation patterns at different time scales and avoids configuration deviations caused by improper data preprocessing.

[0016] 2. This invention constructs a multi-objective optimization function that comprehensively considers renewable energy consumption, load guarantee, and energy storage lifetime. It also proposes a whale optimization algorithm improved based on multi-scale data characteristics. Through the synergistic effect of intelligent population initialization, adaptive step size adjustment, and dynamic search strategy selection, it achieves efficient global optimization in complex constraint spaces. This method effectively avoids the over-configuration problem caused by traditional single-objective optimization and random search, and can achieve significant savings in energy storage capacity and overall cost under different renewable energy penetration scenarios. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the energy storage capacity configuration method based on multi-scale data and adaptive optimization according to the present invention. Figure 2This is a schematic diagram of the energy storage capacity configuration system based on multi-scale data and adaptive optimization according to the present invention. Figure 3 This is a time series diagram of the original new energy output and load data in an embodiment of the present invention; Figure 4 This is a comparison chart of the new energy output normalization method on the first day in the embodiments of the present invention; Figure 5 This is a comparison chart of the load data normalization methods for the first day in this invention embodiment; Figure 6 This is a comparison chart of energy storage capacity configurations under different new energy penetration rates in embodiments of the present invention; Figure 7 This is a comparison chart of comprehensive costs under different new energy penetration rates in embodiments of the present invention; Figure 8 This is a probability density distribution diagram of energy storage capacity saving rate in an embodiment of the present invention; Figure 9 This is a probability density distribution diagram of the overall cost saving rate in an embodiment of the present invention. Detailed Implementation

[0018] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0019] Example 1: As Figure 1 As shown, the present invention provides a method for configuring energy storage capacity based on multi-scale data and adaptive optimization, comprising: S1: Collect historical renewable energy output data and historical load data and perform cleaning processing; Specifically, historical renewable energy output data covers the time series of power output from renewable energy sources such as wind power and photovoltaic power, while historical load data covers the time series of electricity demand in the power grid or specific regions at different times. This data is typically extracted from the SCADA systems of renewable energy power plants, the EMS systems of the power grid dispatch center, and historical databases to ensure the authenticity and completeness of the data.

[0020] To fully reflect the seasonal characteristics, weather conditions, and intermittent patterns of new energy output and load, the collected data should have a sufficiently long time span, preferably several years. The temporal resolution of the data should be determined based on the actual optimization accuracy requirements, and can be selected at the hourly or minute level to capture fluctuation characteristics at different time scales.

[0021] After obtaining the raw data, necessary cleaning processes are required. For missing values ​​in the data series, linear interpolation is used to fill in the gaps and maintain the integrity of the time series. For anomalous jumps or noise interference in the data, moving averages are used for smoothing to eliminate the interference of extreme outliers on subsequent analysis. Furthermore, the time bases of different data sources must be verified to ensure accurate correspondence between the data series in the time dimension.

[0022] The cleaned data is arranged chronologically to form a continuous time series. The optimization method used in this invention directly uses historical time series data as input, thus eliminating the need for additional annotation. Based on actual needs, the complete dataset can be divided into several scenario segments according to time windows to simulate the changing patterns of renewable energy output and load under various typical operating conditions, thereby providing sufficient raw data support for subsequent multi-scale normalization processing.

[0023] The new energy output and load data obtained after the above processing have good quality and continuity, and can be used as input data for subsequent multi-scale decomposition and capacity optimization configuration.

[0024] S2: Decompose the cleaned data into short-term, medium-term and long-term components according to time scale. Calculate the local mean of each time scale component through a sliding window. Calculate the difference between each time scale component and the corresponding local mean, and normalize it by combining the weight decay coefficient with an exponential function to obtain adaptive weight coefficients. Scale the corresponding scale components based on the extreme values ​​within the time window of each time scale, and combine them with the corresponding adaptive weight coefficients to obtain multi-scale normalized data. Furthermore, the calculation steps for the adaptive weighting coefficients include: Calculate the absolute difference between each time scale component and its corresponding local mean; The negative of the product of the absolute difference and the weight decay coefficient is used as the input to the exponential function to obtain the exponential value. The adaptive weighting coefficient of each time scale component at each time point is obtained by dividing the exponent value of each time scale component by the sum of the exponent values ​​of all time scale components.

[0025] Specifically, the cleaned renewable energy output and load data are first decomposed according to time scales. Multi-scale decomposition technology is used to decompose the original time series into three time scales: short-term, medium-term, and long-term components. The short-term component reflects rapid fluctuations at the hourly or minute level, the medium-term component reflects regular changes at the daily or weekly level, and the long-term component reflects slow changes in seasonality or multi-day trends. After decomposition, the following is obtained: Each time scale Raw renewable energy output load data at any given time ,in For time-scale indexing. For time indexing.

[0026] It should be noted that, It is a scalar value, representing the original data after multi-scale decomposition at the [number]th [scale]. The time scale component is at time 1. The specific numerical value. For example, when decomposing the daily power output data of a photovoltaic power station, in... Short-term scale component values ​​may be obtained at time 2. Mid-scale component values Long-term scale component values , representing the contributions of hourly fluctuations, daily cycle fluctuations, and multi-day trends at that moment, respectively.

[0027] After obtaining the components at each time scale, adaptive weighting coefficients need to be calculated to dynamically balance the contributions of each scale. The sliding window technique is used to calculate the weighting coefficients at each time scale. Each time scale Local mean at time This local mean reflects the local average level of the data at that scale. Then, the components at each time scale are calculated. and corresponding local mean absolute difference This difference value reflects the degree to which the data at the current moment deviates from the local average level. This absolute difference is then compared with the weight decay coefficient. The product is multiplied and the negative value is used as input to the natural exponential function. Weight decay coefficient This value is used to adjust the sensitivity of the current data to the difference between the current data and the local mean in the weight calculation. The larger the value, the more the weights tend to favor the scale with smaller fluctuations. In this invention, it is preferably set to 0.95. Finally, the exponent value of each time scale is divided by the sum of the exponent values ​​of all time scales for normalization, to obtain the... Each time scale Adaptive weighting coefficients at time intervals The calculation formula is: ; in This represents the natural exponential function. The adaptive weighting coefficients characterize the contribution of that time scale to the normalization result, dynamically adjusting the weights of each scale based on the local fluctuations in the data, thus enhancing robustness to local abrupt changes.

[0028] Furthermore, the calculation steps for the multi-scale normalized data include: Obtain the maximum and minimum values ​​of the corresponding scale components within the time window of each time scale; The scaling value is obtained by dividing the difference between each time scale component and the minimum value within the corresponding time window by the difference between the maximum value and the minimum value. Multiplying the scaling value by the adaptive weighting coefficient at the corresponding time point yields the multi-scale normalized data for each time scale.

[0029] Specifically, after obtaining the adaptive weight coefficients, multi-scale normalized data is further calculated. First, the first... The set of time windows corresponding to each time scale This time window represents the time range at that scale; for example, a short-term scale corresponds to an hourly window, a medium-term scale to a daily window, and a long-term scale to a monthly window. Within the time window... Retrieve raw data maximum value and minimum value ,in This is the index of the moment within the time window. The [time index] will be... Each time scale Raw data of time Dividing the difference between the minimum value and the maximum value within the time window by the difference between the minimum and maximum values ​​completes the data scaling based on the time window, yielding a normalized intermediate result. This scaled value is then compared with the adaptive weight coefficient at the corresponding time point. Multiply to get the first... Each time scale Multiscale normalized data at different times The calculation formula is: ; This multi-scale normalized data maps the original data to a standard interval, while retaining the multi-scale fluctuation characteristics through adaptive adjustment of weight coefficients. Compared with traditional normalization methods, it can better adapt to the strong fluctuations and multi-time-scale correlation characteristics of new energy output and load data.

[0030] While performing multi-scale normalization, the standard deviation of each time-scale component at each time point is calculated using the sliding window technique. The standard deviation reflects the degree of data fluctuation at various time scales and is obtained by calculating the root mean square deviation of the data within a time window. This standard deviation data will be used in subsequent optimization algorithms for population initialization and dynamic updates of control parameters to enhance the algorithm's adaptability to data fluctuation characteristics.

[0031] Through the above multi-scale normalization process, the data is standardized and mapped, while the fluctuation characteristics of the original data at different time scales are preserved. This avoids the distortion problem that occurs when traditional methods deal with highly volatile data, and provides higher quality input data for subsequent energy storage capacity optimization.

[0032] S3: Calculate the load power shortage, renewable energy curtailment and energy storage life loss factor based on multi-scale normalized data, and construct a comprehensive objective function by weighting the three and defining the operation constraints of the energy storage system. Furthermore, the calculation steps for the load power shortage include: The difference between the load data in the multi-scale normalized data and the sum of the grid interaction power and the energy storage discharge power is compared with zero, and the larger of the two values ​​is taken as the load power shortage.

[0033] Specifically, the load power shortage is first calculated based on multi-scale normalized data. At any given time, the normalized load data will be... Power interaction with the grid and energy storage discharge power Compare the sums. Among them... It represents the power exchange between the system and the power grid, with positive values ​​indicating power absorbed from the power grid. This characterizes the power supplied by the energy storage system to the load. The larger of zero and this difference is obtained. Power shortage at any given time The calculation formula is: ; in This indicates that the value is the larger of 0 and the value within parentheses, ensuring that the load power shortage is non-negative. This load power shortage quantifies the gap between load demand and actual power supply, and is used to assess the system's ability to meet load demands.

[0034] Furthermore, the calculation steps for the abandoned power of the new energy source include: The difference between the sum of renewable energy output data and the sum of energy storage charging power and grid interaction power in the multi-scale normalized data is compared with zero, and the larger of the two values ​​is taken as the renewable energy curtailment power.

[0035] Specifically, in At any time, the normalized new energy output data will be used. With energy storage charging power and grid interaction power Compare the sums. Among them... Characterizes the power absorbed by the energy storage system from new energy sources. This characterizes the power delivered by the system to the grid. The larger of zero and this difference is obtained. Power curtailment of renewable energy at any given time The calculation formula is: ; This quantification of renewable energy curtailment power refers to the portion of renewable energy output that is wasted due to its inability to be absorbed.

[0036] Furthermore, the calculation steps for the energy storage lifetime loss factor include: Calculate the absolute value of the difference between the energy storage charging power and the energy storage discharging power at each time point and sum them up. Divide the summation result by the energy storage capacity to obtain the energy storage lifetime loss factor.

[0037] Specifically, within the optimized timeframe, the charge-discharge cycles of an energy storage system will cause capacity degradation and lifespan loss. The energy storage charging power at each moment will be calculated. With energy storage discharge power absolute value of the difference This value reflects the charging and discharging intensity of the energy storage system at that moment. The charging and discharging intensity at all moments is then considered within the optimized time range. The total cycle loss is obtained by summing the results. Then, the sum is divided by the energy storage capacity. Normalization is performed to obtain the energy storage lifetime loss factor. The calculation formula is: ; in To optimize the total number of time points within the time range, This represents the energy storage capacity and is one of the optimization variables. The energy storage lifetime loss factor characterizes the degree of cyclic loss of the energy storage system during the optimization period and is used to assess the impact of the configuration scheme on the energy storage lifetime.

[0038] After obtaining the three evaluation indicators mentioned above, a comprehensive objective function is constructed to achieve unified optimization of multiple objectives. A single optimization objective is constructed by weighting the sum of squares of the load shortage power at all times, the sum of squares of the renewable energy curtailment power at all times, and the energy storage lifetime loss factor. The squared form is used to increase the penalty for large deviations, making the optimization focus more on avoiding severe power shortages or curtailment. Comprehensive Objective Function The calculation formula is: ; in The weighting coefficient for the load power shortage is used to adjust the importance of the load satisfaction rate in the objective function. In this invention, it is preferably set to 0.5. This is the weighting coefficient for the curtailed power of new energy sources, used to adjust the importance of the new energy absorption rate in the objective function, and is preferably set to 0.3. These are the weighting coefficients for the energy storage lifetime loss factor, used to adjust the importance of energy storage lifetime in the objective function, and are preferably set to 0.2. By adjusting these three weighting coefficients, different balances can be achieved between renewable energy consumption, load guarantee, and energy storage lifetime according to actual needs.

[0039] To ensure the physical feasibility and operational safety of the optimization results, it is necessary to define the operational constraints of the energy storage system. These constraints include three aspects: power constraints, capacity constraints, and state constraints. The power constraint requires the energy storage system to have the required charging power. and discharge power All do not exceed the rated power of energy storage , represented as and ,in It is also one of the optimization variables, characterizing the maximum charge and discharge capacity of the energy storage system. Capacity constraints require that the energy storage state... Within a reasonable range, it is represented as ,in Characterizing energy storage systems in The remaining energy percentage at any given time. The state of energy storage is calculated using a recursive formula, expressed as: ; in for The energy storage state at a given moment represents the proportion of remaining energy in the energy storage system at the previous moment. Energy storage charging efficiency, which characterizes the proportion of energy loss during the charging process, is preferably set to 0.1 in this invention. The energy storage discharge efficiency, characterizing the proportion of energy loss during the discharge process, is preferably set to 0.1. This recursive formula reflects the dynamic change of the energy state of the energy storage system with its charging and discharging behavior.

[0040] By constructing the above comprehensive objective function and defining the operational constraints, a complete energy storage capacity optimization configuration model is established. This model can comprehensively balance multiple optimization objectives such as new energy consumption, load satisfaction, and energy storage lifespan, while ensuring the feasibility of system operation.

[0041] S4: Initialize the whale population position using the adaptive weight coefficients, local mean, and standard deviation of each time scale component at the reference time. Calculate the step size correction amount using the standard deviation of each time scale component at the reference time to dynamically update the control parameters. Calculate the global search probability and select a search strategy based on the average distance between the current population and the historical best position using an exponential function. When the relative change in the objective function value and the average distance of the population are both less than the threshold, output the best position as the energy storage capacity configuration result.

[0042] Furthermore, the step of initializing the whale population location includes: Select the midpoint of the data time range as the reference time; Calculate the product of the adaptive weight coefficient of each time scale component at the reference time and the local mean, and sum them up. Divide the summation result by the maximum value of the local mean of all time scale components at the reference time to obtain the normalization factor. The standard deviation of each time scale component at the reference time is obtained by using a sliding window. The product of the standard deviation of each time scale component at the reference time and the random number is calculated and summed. The result is multiplied by the population initialization adjustment rate to obtain the individual correction increment. The initial position of each whale individual is obtained by multiplying the difference between the lower and upper bounds of the optimization variable by the normalization factor, adding the lower bound of the optimization variable, and then adding the individual correction increment.

[0043] Specifically, the first step is to initialize the whale population location. Traditional random initialization methods cannot utilize the multi-scale statistical characteristics of renewable energy output and load data, resulting in poor initial population quality. This invention guides population initialization by combining statistical information from multi-scale data at a reference time. The midpoint of the data time range is selected as the reference time. This moment can well represent the overall data characteristics. Calculate the adaptive weighting coefficients of each timescale component at the reference time. With local mean The product of these products, summed over all time scales, yields the result. The summation result comprehensively reflects the weighted contribution of each time scale at the reference time. Then, the summation result is divided by the maximum value of the local mean of all time scale components at the reference time. This yields a normalization factor, which is used to map multi-scale statistical information to the optimization variable space.

[0044] To enhance population diversity, individual correction increments also need to be calculated. The standard deviation of each timescale component at the reference time is obtained using a sliding window technique. This standard deviation reflects the degree of data fluctuation at that scale. Calculate the standard deviation of each time scale component at the reference time. With random numbers The product of these products and summed over all time scales, where for Random numbers within a certain range are used to introduce randomness and ensure that the adjustment increment is different for each individual. The summation result above is then compared with the population initialization adjustment rate. Multiplication yields the individual correction increment. The calculation method is as follows: ; in The response rate, which characterizes the data fluctuation features, is preferably set to 0.01 in this invention.

[0045] Finally, optimize the lower bound of the variable. With the upper realm The difference, multiplied by the normalization factor, plus the lower bound of the optimization variable, plus the individual correction increment, yields the first... The initial position of each individual whale The calculation formula is: ; in For individual whale indexes, , Let be the population size. The lower and upper bounds of the optimization variables are respectively... and This represents the minimum and maximum allowable values ​​for energy storage capacity and rated power. This initial location allows the initial population to cover potentially valuable areas of the search space and respond to data fluctuations, providing a better starting point compared to random initialization.

[0046] Furthermore, the step of dynamically updating the control parameters includes: The standard deviation of each time scale component at the reference time is calculated using a sliding window. The step size correction is obtained by summing the standard deviations of each time scale component at the reference time, dividing by the total number of time scales, and then multiplying by the step size adjustment factor. The updated control parameters are obtained by subtracting the product of the current iteration number and the difference between the initial and final control parameter values, divided by the maximum iteration number, and then adding the step size correction.

[0047] Specifically, after population initialization, position updates are required during the iterative optimization process. Traditional whale optimization algorithms lack the ability to adapt to the strong fluctuations in renewable energy output and load data. This invention achieves adaptive search by dynamically updating control parameters. Control parameters In traditional algorithms, the step size decreases linearly from 2 to 0 to balance global exploration and local exploitation. This invention introduces a step size correction, allowing the control parameter to be dynamically adjusted based on data fluctuation characteristics.

[0048] The standard deviation of each time scale component at the reference time was calculated using the sliding window technique. This reflects the degree of data fluctuation at each scale. The sum of the standard deviations of each time scale component at the reference time is divided by the total number of time scales. The mean standard deviation is obtained, and then adjusted with the step size factor. Multiply to obtain the step size correction. The calculation formula is: ; in To adjust the degree of influence of data volatility on step size correction, the parameter is preferably set to 1.2 in this invention. Then, the initial value of the control parameter is... Subtract the current iteration number With control parameter initial and final values Difference divided by the maximum number of iterations The product, plus the step size correction. The updated control parameters are obtained. The updated formula is: ; in Select setting to 2, The preferred setting is 0. This indicates a parameter update operation. This dynamic update mechanism enables the algorithm to adaptively adjust its global and local search tendencies during iteration, enhancing its responsiveness to data fluctuations.

[0049] Based on updated control parameters and random vectors Calculate the first coefficient vector Second coefficient vector The calculation formula is: ; ; in and All A random vector within a certain range. Controlling the movement direction of an individual whale promotes global exploration with positive values ​​and local development with negative values. Controlling the movement distance of individual whales enhances the randomness of the search.

[0050] Furthermore, the step of selecting a search strategy includes: Calculate the L2 norm of the distance between each individual whale's location and its historical best location, and then average them to obtain the population average distance. The negative of the product of the average population distance and the search balance coefficient is used as the input of the exponential function. The global search probability is obtained by adding 1 to the value of the exponential function and taking the reciprocal. A random number is generated and compared with the global search probability. When the random number is less than the global search probability, the coefficient vector is calculated based on the updated control parameters and the random vector, and the whale individual position is updated by selecting either the encirclement predation strategy or the spiral update strategy. Otherwise, the gradient of the objective function at the current individual position is calculated by the finite difference method and multiplied by the local learning rate before updating the whale individual position.

[0051] Specifically, during the location update process, a dynamic search strategy needs to be selected to balance global exploration and local development. This invention calculates the global search probability based on the average distance between the current population and the historical best position. First, the position of each individual whale is calculated. Best historical location L2 norm of distance Then calculate the average to obtain the average population distance. The calculation method is as follows: ; in The L2 norm is the Euclidean norm. The average distance of this population characterizes the diversity of the population; the greater the distance, the more dispersed the population distribution.

[0052] Population average distance With search balance coefficient The negative of the product is used as the input to the exponential function to obtain... ,in The natural exponential function is shown. To adjust the influence of the average distance on the search probability, it is preferably set to 0.8 in this invention. The global search probability is obtained by adding 1 to this exponential function value and taking its reciprocal. The calculation formula is: ; This global search probability representation algorithm tends to favor global exploration; the closer the value is to 1, the more it favors global search. When the population distribution is dispersed... Larger The population is relatively large, and the algorithm tends to explore globally. As the population gradually converges... Smaller The algorithm is relatively small, so it is being developed locally.

[0053] When updating at a specific execution location, generate random numbers within the range and the global search probability Compare them. When random numbers Less than the global search probability A global search strategy is then executed. This is based on the coefficient vector. and and the current best individual position Update the individual whale's location. Specifically, generate another one. The selection of either an encirclement-prey strategy or a spiral update strategy is based on random numbers within a given range. The update formula for the encirclement-prey strategy is: ; The update formula for the spiral update strategy is: ; in Spiral shape constant, for Random numbers within a range Represents the natural constant.

[0054] When random number Greater than or equal to the global search probability A local search strategy is then employed. The objective function is then calculated using the finite difference method. At the current individual location gradient approximation at The gradient represents the direction of change of the objective function at that point. The gradient is approximated by the local learning rate. After multiplying, add the result to the current position to update the individual whale's position. The update formula is: ; in The step size for gradient updates is preferably set to 0.05 in this invention. Introducing gradient information can improve search accuracy and convergence speed during the local search phase.

[0055] After the location is updated, runtime constraints need to be addressed to ensure the feasibility of the optimization results. For the updated location... Boundary projection correction is performed, projecting components that exceed the upper and lower bounds of the optimization variables back into the feasible region. The correction formula is as follows: ; in This indicates the operation of finding the maximum value. This indicates the minimum value operation. If a new solution is found... If the operating constraints of the energy storage system are violated, the penalty function method is used to adjust the objective function value, and the degree of violation of the constraints is added as a penalty term to the objective function.

[0056] In iterative optimization, it is necessary to determine whether convergence has occurred to decide when to terminate and output the results. Traditional methods using a fixed number of iterations or a single threshold can easily lead to premature stopping or excessive iterations, failing to accurately identify whether the optimization process has truly converged. This invention uses adaptive convergence judgment based on the multi-scale stability of the objective function value and population diversity. It compares the objective function value of the current iteration by backtracking. Compared with the previous one The objective function value of the next iteration Calculate the relative change ,in This is the backtracking window size. Simultaneously, the average population distance is monitored. To assess the degree of population concentration.

[0057] When the relative change in the objective function value is less than the convergence threshold And the average distance between populations is less than the convergence threshold. When the optimization process has converged, it is determined that the following conditions are met: ; At this point, output the current best individual position. This is the optimized energy storage capacity configuration result. The result includes the optimal energy storage capacity. and rated power of energy storage It can achieve optimized system operation while meeting multiple objectives such as renewable energy consumption, load guarantee, and energy storage lifespan. If any convergence condition is not met, the subsequent iteration process continues until convergence or the maximum number of iterations is reached.

[0058] By implementing the improved whale optimization algorithm, intelligent optimization configuration of energy storage capacity is achieved. This method can adapt to the multi-scale fluctuation characteristics of new energy output and load data, ensuring optimization efficiency while avoiding getting trapped in local optima, and providing an economical and efficient energy storage capacity configuration solution for different new energy penetration scenarios.

[0059] Example 2: This example uses a real-world scenario for simulation experiments. Through comparative analysis, it demonstrates the technical advantages of the energy storage capacity configuration system based on multi-scale data and adaptive optimization of this invention in processing the normalization of new energy output data and optimizing energy storage capacity configuration. The energy storage capacity configuration system based on multi-scale data and adaptive optimization, such as... Figure 2 As shown, it includes: The data acquisition module is used to collect historical renewable energy output data and historical load data and perform cleaning processing. The multi-scale normalization module is used to decompose the cleaned data into short-term, medium-term and long-term components according to time scale. It calculates the local mean of each time scale component through a sliding window, calculates the difference between each time scale component and the corresponding local mean, and normalizes it by combining the weight decay coefficient through an exponential function to obtain the adaptive weight coefficient. Based on the extreme values ​​within the time window of each time scale, it scales the corresponding scale component and combines it with the corresponding adaptive weight coefficient to obtain multi-scale normalized data. The objective function construction module is used to calculate the load power shortage, renewable energy curtailment, and energy storage lifetime loss factor based on multi-scale normalized data, and to construct a comprehensive objective function by weighting and combining the three and defining the operating constraints of the energy storage system. The adaptive optimization module initializes the whale population position using the adaptive weight coefficients, local mean, and standard deviation of each time scale component at the reference time. It dynamically updates the control parameters by calculating the step size correction amount through the standard deviation of each time scale component at the reference time. Based on the average distance between the current population and the historical best position, it calculates the global search probability through an exponential function and selects a search strategy. When the relative change of the objective function value and the average distance of the population are both less than the threshold, it outputs the best position as the energy storage capacity configuration result.

[0060] Specifically, this embodiment compares it with the conventional minimum-maximum normalization method and the Z-score normalization method. Figure 4The paper presents a comparison of the results of processing the first day's renewable energy output data using three normalization methods. The conventional min-maximum normalization method exhibits over-smoothing when handling rapid fluctuations, severely weakening local abrupt changes. While the conventional Z-score normalization method retains some fluctuation characteristics, it produces unreasonable numerical jumps when the data changes significantly. In contrast, the method of this invention, through multi-scale decomposition and adaptive dynamic adjustment of weight coefficients, preserves long-term trend information while effectively capturing short-term fluctuation characteristics, achieving smooth transitions even at points of data abrupt change.

[0061] Figure 5 The normalized comparison results of the load data for Day 1 are presented. Conventional normalization methods also suffer from feature distortion, while the method of this invention can accurately characterize local changes while maintaining the overall trend of the data, providing higher quality input data for subsequent energy storage capacity optimization.

[0062] This embodiment designs 11 simulation scenarios with different new energy penetration rates ranging from 20% to 80%. Figure 6 This paper compares the optimal energy storage capacity determined by the method of this invention with that determined by conventional configuration methods. The horizontal axis represents the penetration rate of new energy sources, and the vertical axis represents the energy storage capacity, with units in megawatt-hours (MWh). As the penetration rate of new energy sources increases, the required energy storage capacity of both methods shows an upward trend. However, the energy storage capacity determined by the method of this invention is significantly lower than that of the conventional method in all scenarios, and the growth trend is more gradual. This indicates that the present invention, through multi-objective comprehensive optimization and the precise search of the improved algorithm, can more accurately match the actual needs of the system and avoid over-configuration problems.

[0063] Figure 7 The overall cost comparison of the two methods is shown, with the vertical axis representing tens of thousands of yuan per megawatt-hour. The overall cost of the method in this invention is significantly lower than that of conventional methods across all penetration rate scenarios, and the cost advantage becomes increasingly significant as the penetration rate of new energy sources increases. This advantage stems from the multi-objective optimization function constructed in this invention, which can comprehensively weigh various factors, as well as the powerful global search capability and adaptive adjustment mechanism of the improved whale optimization algorithm.

[0064] Figure 8 The probability density distribution of energy storage capacity savings rate is shown, with the horizontal axis representing the savings rate in percentage and the vertical axis representing the probability density. The kernel density estimation curve shows that the savings rate is mainly concentrated in the 20% to 40% range, with an average savings rate of 23.0%. This indicates that the method of this invention can achieve significant capacity savings in most scenarios, avoiding the financial waste caused by over-configuration.

[0065] Figure 9The probability density distribution of the overall cost savings rate is shown. The cost savings rate is mainly concentrated in the range of 15% to 30%, with an average savings rate of 21.8%. The cost savings not only come from the reduction in energy storage capacity, but also from the optimized configuration scheme that achieves better renewable energy consumption and load satisfaction, reduces curtailment losses and power shortage penalties, and extends the service life of energy storage equipment by comprehensively considering the energy storage life loss factor, thereby reducing replacement and maintenance costs.

[0066] The experimental results of this embodiment fully demonstrate that the system of the present invention can achieve precise configuration of energy storage capacity under different new energy penetration scenarios, effectively avoid over-configuration, significantly reduce the overall cost of energy storage system, and has strong economic efficiency, adaptability and promotion application value.

[0067] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for energy storage capacity configuration based on multi-scale data and adaptive optimization, characterized in that, The method comprises the following steps: S1: Collect historical new energy output data and historical load data and perform cleaning processing; S2: Decompose the cleaned data into short-term components, medium-term components and long-term components according to time scales, calculate the local mean values of the components of each time scale through a sliding window, calculate the difference between each time scale component and the corresponding local mean value, combine a weight decay coefficient and map through an exponential function to obtain an adaptive weight coefficient, scale the corresponding scale component based on the extreme value in the time window of each time scale and combine the corresponding adaptive weight coefficient to obtain multi-scale normalized data; S3: Calculate the load power shortage, new energy power curtailment and energy storage life loss factor based on the multi-scale normalized data, combine the three to construct a comprehensive objective function and define the energy storage system operation constraint; S4: Initialize the whale population position by using the adaptive weight coefficient, local mean value and standard deviation of each time scale component at the reference time, calculate the step correction amount by the standard deviation of each time scale component at the reference time to dynamically update the control parameter, calculate the global search probability by the exponential function based on the average distance between the current population and the historical best position and select the search strategy, and output the best position as the energy storage capacity configuration result when the relative change of the objective function value and the average distance of the population are less than the threshold value. 2.The energy storage capacity configuration method based on multi-scale data and adaptive optimization according to claim 1, wherein, The calculation steps of the adaptive weight coefficient comprise: Calculate the absolute difference between each time scale component and the corresponding local mean value; Take the product of the absolute difference and the weight decay coefficient as the input of the exponential function to obtain an exponential value; Divide the exponential value of each time scale component by the sum of the exponential values of all time scale components to obtain the adaptive weight coefficient of each time scale component at each time. 3.The energy storage capacity configuration method based on multi-scale data and adaptive optimization according to claim 1, characterized in that, The calculation steps of the multi-scale normalized data comprise: Obtain the maximum and minimum values of the corresponding scale component in each time scale time window; Divide the difference between each time scale component and the minimum value in the corresponding time window by the difference between the maximum value and the minimum value to obtain a scaling value; Multiply the scaling value and the adaptive weight coefficient at the corresponding time to obtain the multi-scale normalized data of each time scale. 4.The energy storage capacity configuration method based on multi-scale data and adaptive optimization according to claim 1, wherein, The calculation steps of the load power shortage comprise: Compare the difference between the load data in the multi-scale normalized data and the sum of the grid interaction power and the energy storage discharge power with zero, and take the larger value as the load power shortage.

5. The method for energy storage capacity configuration based on multi-scale data and adaptive optimization according to claim 1, characterized in that, The calculation steps of the new energy power curtailment comprise: Compare the difference between the new energy output data in the multi-scale normalized data and the sum of the energy storage charging power and the grid interaction power with zero, and take the larger value as the new energy power curtailment. 6.The energy storage capacity configuration method based on multi-scale data and adaptive optimization according to claim 1, wherein, The calculation steps of the energy storage life loss factor comprise: Calculate the absolute value of the difference between the energy storage charging power and the energy storage discharge power at each time and sum them up, and divide the sum by the energy storage capacity to obtain the energy storage life loss factor. 7.The energy storage capacity configuration method based on multi-scale data and adaptive optimization according to claim 1, wherein, The steps of initializing the whale population position comprise: Select the middle point of the data time range as the reference time; Calculate the product of the adaptive weight coefficient and the local mean value of each time scale component at the reference time and sum them up, and divide the sum by the maximum value of the local mean values of all time scale components at the reference time to obtain a normalization factor; The standard deviation of each time scale component at the reference time is obtained through a sliding window, the product of the standard deviation of each time scale component at the reference time and a random number is calculated and summed, and the result is multiplied by the population initialization adjustment rate to obtain an individual correction increment; The lower limit and upper limit of the optimization variable are multiplied by a normalization factor, and the lower limit of the optimization variable is added to obtain the initial position of each whale individual. 8.The energy storage capacity configuration method based on multi-scale data and adaptive optimization according to claim 1, wherein, The step of dynamically updating the control parameter comprises: The standard deviation of each time scale component at the reference time is calculated through a sliding window; The standard deviations of each time scale component at the reference time are summed and divided by the total number of time scales, and then multiplied by a step size adjustment coefficient to obtain a step size correction; The initial value of the control parameter is subtracted by the product of the current iteration number and the difference between the initial value and the final value of the control parameter divided by the maximum number of iterations, and then added to the step size correction to obtain the updated control parameter. 9.The energy storage capacity configuration method based on multi-scale data and adaptive optimization of claim 1, wherein, The step of selecting a search strategy comprises: The L2 norm of the distance between the position of each whale individual and the historical best position is calculated and averaged to obtain the population average distance; The product of the population average distance and the search balance coefficient is taken as the negative input of the exponential function, and the value of the exponential function is added to 1 to obtain the global search probability. A random number is generated and compared with the global search probability. When the random number is less than the global search probability, the coefficient vector is calculated based on the updated control parameter and the random vector, and the whale individual position is updated based on the surrounding predation or spiral update strategy. Otherwise, the gradient of the objective function at the current individual position is calculated by the finite difference method, multiplied by the local learning rate, and the whale individual position is updated.

10. A system for energy storage capacity configuration based on multi-scale data and adaptive optimization, characterized in that, It comprises: A data acquisition module for acquiring historical new energy output data and historical load data and performing cleaning processing; A multi-scale normalization module for decomposing the cleaned data into short-term components, medium-term components and long-term components according to time scales, calculating the local mean of each time scale component through a sliding window, calculating the difference between each time scale component and the corresponding local mean, and combining the weight decay coefficient to map through an exponential function to obtain adaptive weight coefficients. Based on the extreme value in each time scale time window, the corresponding scale component is scaled and combined with the corresponding adaptive weight coefficient to obtain multi-scale normalized data; A target function construction module for calculating load power shortage, new energy power abandonment and energy storage life loss factor based on multi-scale normalized data, and constructing a comprehensive objective function by weighting and combining the three, and defining the energy storage system operation constraint; An adaptive optimization module for initializing the whale population position using the adaptive weight coefficients, local means and standard deviations of each time scale component at the reference time, calculating the step size correction by the standard deviation of each time scale component at the reference time to dynamically update the control parameter, calculating the global search probability by the average distance between the current population and the historical best position through an exponential function, and selecting a search strategy. When the relative change of the objective function value and the population average distance are both less than the threshold value, the best position is output as the energy storage capacity configuration result.

Images