Energy storage comprehensive evaluation method for multiple scenes of side peak regulation and frequency modulation of power grid

Abstract

The invention provides an energy storage comprehensive evaluation method for multiple scenes of power grid lateral peak regulation and frequency modulation, and relates to the technical field of energy storage allocation decisions, in particular to objective weighting, baseline sorting, scene partial order modeling and robustness test under uncertainty. The method is suitable for energy storage technology selection, configuration and investment decision support under the background of a power grid side, source grid load storage integration and a novel power system. Applicability mask, objective empowerment, baseline sorting, scene partial order and Monte Carlo robustness test are adopted to form a unified link, and remarkable effects are achieved for the difficulties of cross-technology incomparability, empowerment redundancy, scene difference expression, uncertainty verification and the like.

Description

Technical Field

[0001] This invention relates to the field of energy storage technology for power systems. Background Technology

[0002] With a high proportion of renewable energy integrated into the grid, the demand for energy storage in power systems is increasing simultaneously in two typical scenarios: peak shaving and frequency regulation. Different technologies vary significantly in terms of continuous discharge duration, power response speed, round-trip efficiency, lifetime equivalent, construction period, land and terrain constraints, and total life-cycle cost. This leads to the possibility that the ranking of the same technology in different scenarios may be interchangeable or complementary. In engineering practice, there is an urgent need to conduct unified, traceable, and interpretable comprehensive evaluations under multiple indicators, heterogeneous samples, and multiple scenario constraints to support technology selection, capacity allocation, and investment decisions, and to ensure consistency with grid operation objectives.

[0003] However, existing multi-indicator decision-making methods often directly employ entropy weighting and analytic hierarchy process (AHP) to assign weights to multi-dimensional indicators and perform comprehensive ranking. When used for comparing energy storage technology allocation solutions, these methods generally suffer from the following shortcomings: First, the weighting process is highly subjective and lacks redundancy reduction for highly correlated indicators, easily leading to double-counting and bias. Second, cross-technology applicability differences are not explicitly characterized; incomparable dimensions are often handled by removing or simply filling in gaps, resulting in information loss and structural errors. Third, normalization and missing value management are sensitive to outliers, and the ranking results fluctuate excessively for individual anomalies, making engineering simulation and verification insufficient. Fourth, most methods conclude with a single global ranking, failing to reflect the differences in target preferences and constraint focuses between peak shaving and frequency regulation, and also failing to output a physically meaningful partial order structure of scenarios.

[0004] Therefore, it is necessary to propose an integrated evaluation method for multiple scenarios of peak shaving and frequency regulation, so as to provide a decision-making basis that can be directly implemented for the scenario-based selection and configuration of energy storage technologies. Summary of the Invention

[0005] To overcome the common shortcomings of existing indicator-based decision-making, this invention provides a comprehensive evaluation method for energy storage in multiple scenarios of grid-side peak shaving and frequency regulation.

[0006] The technical solution adopted by this invention to achieve the above objectives is: a comprehensive evaluation method for energy storage in multiple scenarios of grid-side peak shaving and frequency regulation, comprising the following steps:

[0007] Step S1: First, establish an evaluation index set and an original evaluation matrix to obtain multiple indicators for evaluating the comprehensive effect of energy storage, and arrange them in a regular pattern; including but not limited to vanadium redox flow battery (VRFB), lithium iron phosphate (LFP), pumped hydro storage (PHES), compressed air (CAES), flywheel, sodium sulfur (NaS), and hydrogen energy. This indicates at least m evaluation indicators (such as energy cost, power capacity, round-trip efficiency, lifespan equivalent, construction period, etc.).

[0008] To eliminate differences in dimensions and inconsistencies in statistical standards, unit conversion and standardization were first completed, and the data were quantile-trimmed and missing data were handled to obtain the processed sample set.

[0009] Step S2: Obtain the masked normalized matrix Then, an objective weight calculation method for indicators under complex information is adopted, namely the CRITIC method, which assigns objective weights to each indicator based on the contrast strength between dispersion and decorrelation, and removes indicators with excessive dispersion and low correlation.

[0010] Step S3: Perform baseline sorting on each evaluation index sequence to obtain an evaluation index sequence sorted according to the closeness coefficient;

[0011] Step S4: Perform scenario partial ordering on the input sorted evaluation index sequence to obtain two sets of comprehensive preference evaluation index sequences with physical meaning: peak modulation and frequency modulation.

[0012] Step S5: Perform a Monte Carlo test on the comprehensive preference evaluation index sequence to obtain the index sequence ranked according to the baseline or the index sequence ranked according to the partial order.

[0013] Specifically, the steps of step S1 are as follows:

[0014] Step S11: To suppress the amplifying effect of outliers on subsequent sorting, quantile pruning is used:

[0015]

[0016] In the formula, , As an indicator The lower and upper quantiles, Use an empirical value of 0.01 to 0.05. These are the cropped observations;

[0017] Step S12: When a numerical item is missing, perform bounded completion, that is, assign a value within the permitted range. The midpoint of the quantile is preferred. Imputation; and record data quality labels for each cell for audit traceability; then define To synthesize the effective values ​​after "trimming or completing": —Non-missing or —Complete missing information;

[0018] Step S13: To achieve monotonic direction consistency, set a direction factor. Corresponding to benefit-type and cost-type indicators respectively, we have:

[0019]

[0020] In the formula, This is the value after direction adjustment; , The endpoints of the quantile in the corresponding direction;

[0021] Step S14: Perform interval normalization based on consistent directions to obtain dimensionless and comparable standardized matrices. :

[0022]

[0023] In the formula, To avoid extreme situations Numerical stability term at time; And maintain a monotonic mapping with the original indication of superiority or inferiority;

[0024] Considering situations where technologies are not comparable or conditions are applicable, such as flow batteries where energy and power can be independently configured, and pumped storage being constrained by terrain and head, an applicability mask matrix is ​​introduced: when technologies are comparable in terms of metrics, otherwise, the standardized output is written as follows:

[0025] ;

[0026] Here, is a masked standardized matrix; subsequent weighting and aggregation are only performed on the comparable intersection of samples, thereby ensuring consistency in evaluation criteria while minimizing information loss.

[0027] Through the above processing, a comparable sample set with unified dimensions, consistent direction, and suppressed outliers can be formed, providing a consistent and robust data foundation for subsequent objective weighting, baseline ranking, and scenario partial ranking.

[0028] Specifically, the steps of S2 are as follows:

[0029] Step S21: Record the first The sample mean of each indicator on the comparable sample set is In the formula The sample standard deviation is:

[0030] ;

[0031] Standard deviation The ability to distinguish indicators within a comparable range is characterized by the greater the distinguishability, the more marginal information it provides for ranking.

[0032] Step S22: To measure decorrelation, between the two indicators... Common comparable sample set Calculate the Pearson correlation coefficient:

[0033] ;

[0034] In the formula and In the set respectively If the two columns are perfectly linear and in the same or opposite direction, then... If statistically independent or linearly independent, Approaching zero;

[0035] Step S23: Combine discriminant and decorrelation scores, and define the first... The comparative strength of the indicators:

[0036] ;

[0037] In the formula, when When it approaches 1, A value close to 0 indicates that this indicator is highly redundant with other indicators. It then contracts; when When the value is large and has low correlation with other indicators, The increase indicates a higher contribution to the differentiation scheme;

[0038] Step S24: Obtain the global objective weights by normalizing the contrast intensity:

[0039] ;

[0040] In the formula This refers to the numerically stable term under extreme conditions.

[0041] Considering that different energy storage technologies may be incomparable or lack data on some indicators, a "comparable intersection reweighting" mechanism needs to be introduced at the sample level. That is, for a candidate technology, it only forms a comparable intersection with other technologies on a subset of indicators with data or applicability. In this subset, the weights of each indicator are renormalized according to the relative proportion of the original objective weights, and incomparable (inapplicable) sample sets are removed, so that the sum of the weights of each indicator in this subset remains at 1. This avoids including indicators that are not applicable to the technology in the scoring, and also avoids the total weight from decreasing or the evaluation results from structural bias due to simply removing indicators.

[0042] Step S25: Record the sample The set of comparable metrics is Then the sample In indicators The effective weights on the t-axis are defined as:

[0043] ;

[0044] This ensures that the sum of the weights of each sample within its own comparable intersection is 1, guaranteeing consistent evaluation criteria and conservation of total weight.

[0045] Finally, the weighted matrix is ​​obtained.

[0046] ;

[0047] In the formula, This serves as the input for subsequent baseline ranking and scenario partial order calculations; The normalized value with a mask; Effective weights at the sample level.

[0048] Specifically, the steps of S3 are as follows:

[0049] Step S31: To measure the degree of approximation to the ideal, calculate the effective sample set for each indicator column. Define the ideal and antiideal solutions above:

[0050] ;

[0051] In the formula, Indicators The optimal weighted performance. Indicates the worst-weighted performance; both are determined solely by the valid data in that column and are not affected by missing or incomparable cells.

[0052] Step S32: For any sample In their comparable intersection The Euclidean distance between the ideal and antiideal solutions is calculated above:

[0053] ;

[0054] In the formula, The smaller the value, the closer the sample is to the optimal value in each column. A larger value indicates that the sample is farthest from the worst in each column; if an individual indicator is in the sample... If the data is not comparable, then the indicator will naturally not be included in the summation, thus avoiding penalties or rewards due to missing data.

[0055] Step S33: Define the closeness coefficient accordingly.

[0056] ;

[0057] In the formula This is a numerically stable term to prevent division by zero when both distances are simultaneously zero. The larger the value, the more likely it is to be a scheme. The closer to the ideal and the further from the anti-ideal, the more accurate the ranking of the global baseline. The distance based on comparable intersections and column endpoint calculations ensures that different energy storage technologies can still obtain a realistic global ranking under a fair caliber when incomparable dimensions exist. At the same time, it is consistent with the redistribution mechanism of the previous stage, realizing an integrated baseline evaluation process in which intra-column endpoints are determined according to effective samples, intra-row distances are accumulated according to comparable intersections, and intra-sample weights are conserved.

[0058] Specifically, the steps of step S4 are as follows:

[0059] Step S41: Obtain the global objective weights Weighting of intersection with in-sample comparable samples Based on this, scene multiplier vectors are introduced. Expression scenario The focus of attention, in the formula They represent peak shaving and frequency modulation, respectively.

[0060] Step S42: Recalibrate the global weights according to the scenario as follows:

[0061] ;

[0062] In the formula, For the scene Lowering the target Normalized weights; As an indicator Global objective weight; For the scene Subordination to indicators The focus is on the multiplier; the larger the value, the more important the indicator is in that scenario. Alternative options A set of applicable indicators with data. This is the pairwise comparable intersection of the two; For the scene The weighted normalization factor of all indicators in the pairwise comparable intersection set; For indexing indicators; when season This indicates that the indicator is not included in the plan. and Paired comparisons;

[0063] Step S43: Using the weighted values ​​obtained in step S3 For utility, define option a relative to option b in terms of metrics. The difference on:

[0064] ;

[0065] All indicators have already achieved directional consistency in step S1, therefore This indicates that scheme a is superior to scheme b on this indicator; using indifference intervals With preference threshold The V-shaped preference function yields the single-index preference:

[0066] ;

[0067] In the formula, This represents an interval where minute differences are negligible. The threshold is considered sufficiently good to be significantly better. Both of these thresholds can be set based on historical operation and engineering experience and vary depending on the scenario.

[0068] Step S44: Aggregate the preferences of each indicator according to the scenario weight to obtain the pairwise preference index:

[0069] ;

[0070] In the formula, the larger π is, the more significant the impact on the scene. The stronger the overall preference for option a relative to b, the better; for each option a, in the candidate set... Calculate positive and negative preferences:

[0071] ;

[0072] And define the overall preference degree as the criterion for scene partial order:

[0073] ;

[0074] when When, it indicates that solution a is suitable for the scenario. If b is better than b; if both are close to zero and their difference is not significant, then it reflects a partial order relation that cannot be judged; this can be determined by pairwise comparable intersections. With scene weight Through normalization, PROMETHEE can maintain a consistent comparison standard even when there are incomparable dimensions and missing data; combined with a threshold function Suppressing minute differences and amplifying significant differences can output two sets of comprehensive preference conclusions with physical meaning: peak modulation and frequency modulation.

[0075] Specifically, the steps of S5 are as follows:

[0076] Step S51: To evaluate the sensitivity of weighting and ranking to parameter uncertainty, let the nominal parameter vector be... Let represent energy investment, power investment, round-trip efficiency, lifetime equivalent conversion factor, and the indifference and preference threshold of the scenario preference function, respectively; then, each component is randomly perturbed according to the engineering interval to form the . Secondary sample : , , , Next, missing units are sampled within the permissible range of the indicator. Finally, for each sampling, steps S1 to S4 are completely recalculated to obtain the scheme. Baseline closeness Or scene net flow And generate rankings based on this. Let the total number of repetitions be . ;

[0077] Step S52: For the robustness output, first record the frequency and average rank of the first-ranked product, as shown in the following formula:

[0078] ;

[0079] In the formula For indicator functions, Representation scheme The proportion that ranks first among all disturbances. For the plan The average relative position.

[0080] Step S53: If the baseline ranking is used, then Depend on Obtained by decreasing; if using scenario ranking, then by Obtained by decreasing.

[0081] A comprehensive evaluation device for energy storage in multiple scenarios of grid-side peak shaving and frequency regulation includes:

[0082] The module for establishing the evaluation index set and the original evaluation matrix is ​​used to obtain multiple indicators for evaluating the comprehensive effect of energy storage and arrange them in a regular pattern.

[0083] The weight allocation module is used to obtain the masked normalized matrix. Then, the CRITIC method was used to assign objective weights to each indicator based on the contrast strength between dispersion and decorrelation, and indicators with excessive dispersion and low correlation were eliminated.

[0084] Baseline sorting module;

[0085] Contextual partial order module;

[0086] The Monte Carlo robustness test module outputs an index sequence ranked by baseline or ranked by partial order.

[0087] Specifically, the baseline sorting module includes:

[0088] Solve the defined unit, and define the effective sample set as the ideal solution and the anti-ideal solution;

[0089] The Euclidean distance calculation unit is used to calculate the Euclidean distance from any sample to the ideal solution and the antiideal solution.

[0090] Specifically, the scenario partial order module includes:

[0091] The scenario calibration unit is used to calibrate the global weights in both frequency modulation and peak modulation scenarios.

[0092] The preference index aggregation unit is used to aggregate the preference indices of each sample to obtain the preference degree, thereby obtaining the comprehensive preference degree.

[0093] A computer-readable medium includes a memory and a processor, wherein the memory stores a program for a comprehensive evaluation method of energy storage for multiple scenarios of grid-side peak shaving and frequency regulation, and the processor executes the steps of the above method.

[0094] The beneficial effects of this invention are as follows: First, the method of this invention ensures fair scoring across technologies by using an applicability mask and reweighting of comparable intersections under a unified caliber; it adopts CRITIC objective weighting based on discriminability and correlation structure, adaptively reducing redundancy of highly correlated indicators and minimizing subjectivity; and it generates a global baseline ranking based on the ideal solution proximity of TOPSIS, resulting in a clear caliber and transparent weighting. Second, after repeated evaluation with Monte Carlo under reasonable parameter perturbations, the baseline ranking and key conclusions show small overall fluctuations and high consistency, demonstrating robustness to common uncertainties. Third, based on the baseline ranking, a scenario multiplier and a threshold-based preference function are introduced, and aggregation on paired comparable intersections based on PROMETHEE yields a comprehensive scenario preference, forming a dual-track evidence of global ranking and scenario partial ranking. Fourth, the method simultaneously outputs column contribution and paired advantage matrices, clearly indicating that peak-shaving side emphasizes energy duration and frequency-regulating side emphasizes power and response, thereby providing a traceable and robust decision-making basis for energy storage technology selection in different scenarios. Attached Figure Description

[0095] Figure 1 This is a flowchart of the algorithm of the present invention. Detailed Implementation

[0096] The present invention will be further explained and described below with reference to the accompanying drawings and embodiments.

[0097] The comprehensive evaluation method for energy storage in multiple scenarios of grid-side peak shaving and frequency regulation according to the present invention includes the following steps:

[0098] Step S1: First, establish the evaluation index set and the original evaluation matrix. In the formula, Candidate energy storage technologies or solutions are indicated – including representative energy storage technologies such as vanadium redox flow batteries (VRFB), lithium iron phosphate (LFP), pumped hydro storage (PHES), compressed air storage (CAES), flywheel, sodium sulfur (NaS), and hydrogen. The evaluation indicators include safety and maturity, environmental factors (power density, energy density, environmental impact), economic factors (power cost, capacity cost, operation and maintenance cost), and technical factors (energy conversion efficiency, power level, response time, charge and discharge duration, cycle life, depth of charge and discharge).

[0099] To eliminate differences in dimensions and inconsistencies in statistical standards, unit conversion and standardization were first completed, and the data was quantified and missing data were handled to obtain a processed sample set. Text-type entries were converted into numerical values ​​through built-in code mapping: for example, maturity "commercial / demonstration project / R&D stage" was mapped to 4 / 3 / 2, safety "high / relatively high / medium / low" was mapped to 4 / 3 / 2 / 1, and environmental impact "no pollution / with residue / pollution residue / air pollution" was mapped to 4 / 3 / 3 / 2. Power level, response time, and duration were mapped into ordinal quantity tables according to the rules of "magnitude, time scale, and duration interval", and efficiency percentages were converted into numerical values.

[0100] Step S2: Obtain the masked normalized matrix Then, the CRITIC method was used to assign objective weights to each indicator based on the contrast strength between dispersion and decorrelation.

[0101] Considering that volumetric energy density and power density are not comparable to PHES, CAES and hydrogen energy storage, an applicability mask is constructed to mark these dimensions as incomparable in the corresponding technologies. Subsequent aggregation is only performed on the comparable intersection (excluding incomparable factors).

[0102] Step S3: Baseline Ranking: To ensure the comprehensiveness of the evaluation, the baseline ranking adopts the TOPSIS comprehensive evaluation method: a weighted matrix is ​​constructed by the column weights after in-sample reweighting and the standardized values. The ideal and anti-ideal endpoints are determined on the valid samples of each column. The Euclidean distance from the sample to the two endpoints is calculated only within their comparable intersection. The proximity coefficient is used as the basis for ranking the global baseline.

[0103] Step S4: Scenario Partial Order: The implementation process follows the principle of "one-time objective weighting and solidification—cross-scenario recalibration." First, directional consistency and interval normalization are completed, and all indicators are standardized to [0,1] according to the principle of "the larger the value, the better." On this basis, CRITIC is used for objective weighting: the standard deviation and correlation coefficient with other columns are calculated for each indicator column, using only the intersection of paired effective samples; the contrast intensity is synthesized according to "discrimination × redundancy removal," and a coverage penalty strategy is introduced to suppress columns with too few effective samples; the global weight vector w is obtained by normalization, and the weights are redistributed at the sample level according to "comparable intersection," ensuring that the sum of the weights of each sample in its own comparable dimension is 1. This weight set remains unchanged in all subsequent steps and is not repeatedly adjusted.

[0104] Step S5: Monte Carlo robustness test.

[0105] The evaluation results show that, under this configuration and data, LFP and VRFB ranked first and second in the frequency modulation scenario (with a comprehensive preference of approximately 0.168 and 0.118), while VRFB and LFP ranked first and second in the peak shaving scenario (with a comprehensive preference of approximately 0.142 and 0.114), indicating that the two have complementary advantages in terms of power response and energy duration.

[0106] Specifically, the steps of step S1 are as follows:

[0107] To suppress the amplifying effect of outliers on subsequent ranking, quantile pruning is employed:

[0108] ;

[0109] In the formula, , As an indicator The lower and upper quantiles, Use an empirical value of 0.01 to 0.05. These are the cropped observations;

[0110] When missing values ​​exist, bounded completion is performed, meaning values ​​are assigned within the permitted range. The midpoint of the quantile is preferred. Imputation; and record data quality labels for each cell for audit traceability; then define To synthesize the effective values ​​after "trimming or completing": —Non-missing or —Complete missing information;

[0111] To achieve monotonicity with consistent direction, let a direction factor be defined. Corresponding to benefit-type and cost-type indicators respectively, we have:

[0112] ;

[0113] In the formula, This is the value after direction adjustment; , The endpoints of the quantile in the corresponding direction;

[0114] Based on consistent direction, interval normalization is performed to obtain a dimensionless and comparable standardized matrix. :

[0115] ;

[0116] In the formula, To avoid extreme situations Numerical stability term at time; And it maintains a monotonic mapping with the original indication of superiority or inferiority.

[0117] S2 includes the following specific steps:

[0118] Record No. The sample mean of each indicator on the comparable sample set is In the formula The sample standard deviation is:

[0119] ;

[0120] Standard deviation The ability to distinguish indicators within a comparable range is characterized by the greater the distinguishability, the more marginal information it provides for ranking.

[0121] To measure decorrelation, in two indicators Common comparable sample set Calculate the Pearson correlation coefficient:

[0122] ;

[0123] In the formula and In the set respectively If the two columns are perfectly linear and in the same or opposite direction, then... If statistically independent or linearly independent, Approaching zero;

[0124] Combining discrimination and decorrelation, the definition of the first The comparative strength of the indicators:

[0125] ;

[0126] In the formula, when When it approaches 1, A value close to 0 indicates that this indicator is highly redundant with other indicators. It then contracts; when When the value is large and has low correlation with other indicators, The increase indicates a higher contribution to the differentiation scheme;

[0127] The global objective weights are obtained by normalizing the contrast intensity:

[0128] ;

[0129] In the formula This refers to the numerically stable term under extreme conditions.

[0130] For a candidate technology, only on a subset of indicators that have data or applicability can a comparable intersection with other technologies be formed, and the weights of each indicator in the subset are renormalized according to the relative proportion of the original objective weights, so that the sum of the weights of each indicator in the subset remains 1.

[0131] Record the sample The set of comparable metrics is Then the sample In indicators The effective weights on the t-axis are defined as:

[0132] ;

[0133] This ensures that the sum of the weights of each sample within its own comparable intersection is 1, guaranteeing consistent evaluation criteria and conservation of total weight.

[0134] Finally, the weighted matrix is ​​obtained.

[0135] ;

[0136] In the formula, This serves as the input for subsequent baseline ranking and scenario partial order calculations; The normalized value with a mask; Effective weights at the sample level.

[0137] The specific steps of S3 are as follows: To measure the degree of approximation to the ideal, in the effective sample set of each indicator column... Define the ideal and antiideal solutions above:

[0138] ;

[0139] In the formula, Indicators The optimal weighted performance. Indicates the worst-weighted performance; both are determined solely by the valid data in that column and are not affected by missing or incomparable cells.

[0140] For any sample In their comparable intersection The Euclidean distance between the ideal and antiideal solutions is calculated above:

[0141] ;

[0142] In the formula, The smaller the value, the closer the sample is to the optimal value in each column. A larger value indicates that the sample is farthest from the worst in each column; if an individual indicator is in the sample... If the data is not comparable, then the indicator will naturally not be included in the summation, thus avoiding penalties or rewards due to missing data.

[0143] Based on this, the closeness coefficient is defined as follows:

[0144] ;

[0145] In the formula This is a numerically stable term to prevent division by zero when both distances are simultaneously zero. The larger the value, the more likely it is to be a scheme. The closer to the ideal and the further away from the anti-ideal, the more it serves as the basis for ranking the global baseline.

[0146] S4 includes the following specific steps:

[0147] After obtaining the global objective weight Weighting of intersection with in-sample comparable samples Based on this, scene multiplier vectors are introduced. Expression scenario The focus of attention, in the formula In this context, `peak` and `freq` represent the peak-tuning sample set and the frequency-tuning sample set, respectively.

[0148] The global weights are then recalibrated according to the scenario as follows:

[0149] ;

[0150] In the formula, For the scene Lowering the target Normalized weights; As an indicator Global objective weight; For the scene Subordination to indicators The focus is on the multiplier; the larger the value, the more important the indicator is in that scenario. Alternative options A set of applicable indicators with data. This is the pairwise comparable intersection of the two; For the scene The weighted normalization factor of all indicators in the pairwise comparable intersection set; For indexing indicators; when season This indicates that the indicator is not included in the plan. and Paired comparisons;

[0151] The weighted value obtained in step S3 For utility, define option a relative to option b in terms of metrics. The difference on:

[0152] ;

[0153] All indicators have been aligned in direction in step S1, therefore This indicates that scheme a is superior to scheme b on this indicator; using indifference intervals With preference threshold The V-shaped preference function yields the single-index preference:

[0154] ;

[0155] In the formula, This represents an interval where minute differences are negligible. The thresholds represent what is considered significantly better, and both can be set based on historical operation and engineering experience and vary depending on the scenario; for peak shaving and frequency regulation, peak shaving places more emphasis on continuity and cost, while frequency regulation places more emphasis on response and power capability.

[0156] The pairwise preference index is obtained by aggregating the preferences of each indicator according to the scenario weight:

[0157] ;

[0158] In the formula, a larger value of π indicates a larger scene The stronger the overall preference for option a relative to b, the better; for each option a, in the candidate set... Calculate positive and negative preferences:

[0159] ;

[0160] And define the overall preference degree as the criterion for scene partial order:

[0161] ;

[0162] when At that time, in scenario a If b is better than b; if both are close to zero and their difference is not significant, then it reflects a partial order relation that cannot be judged; this can be determined by pairwise comparable intersections. With scene weight Through normalization, PROMETHEE can maintain a consistent comparison standard even when there are incomparable dimensions and missing data; combined with a threshold function Suppressing minute differences and amplifying significant differences can output two sets of comprehensive preference conclusions with physical meaning: peak modulation and frequency modulation.

[0163] S5 includes the following steps:

[0164] To assess the sensitivity of weighting and ranking to parameter uncertainty, let a nominal parameter vector be defined. , representing energy investment, power investment, round-trip efficiency, lifetime equivalent conversion factor, and the indifference and preference threshold of the scenario preference function, respectively; each component is randomly perturbed according to the engineering interval to form the first... Secondary sample : , , , Missing units are sampled within the permissible range of the indicator. Each sampling is performed by completely recalculating steps S1 to S4 to obtain the scheme. Baseline closeness Or scene net flow And generate rankings based on this. ;

[0165] Let the total number of repetitions be . ;

[0166] Robust output first records the frequency of ranking first and the average ranking:

[0167] ;

[0168] In the formula For indicator functions, Representation scheme The proportion that ranks first among all disturbances. For the plan The average relative position; if the baseline ranking is used, then Depend on Obtained by decreasing; if using scenario ranking, then by Obtained by decreasing.

[0169] This embodiment selects the actual annual operation data of a provincial power grid as the background, and selects vanadium redox flow batteries, lithium iron phosphate batteries, flywheel energy storage, pumped hydro storage, sodium-sulfur batteries, compressed air, and hydrogen energy storage as candidate technologies. First, a unified multi-index evaluation system is established, with the index set covering four primary dimensions: safety and maturity, environmental impact, technicality, and economics.

[0170] The implementation process first uses the CRITIC method for objective weighting, determining the weights of each level by calculating the comparative strength of each indicator. Data shows that the technical dimension has the highest weight (0.4489), followed by economic efficiency (0.1953), environmental impact (0.1794), and safety (0.1764). Among the secondary indicators, response time (0.1043), environmental impact (0.0937), safety (0.0922), and depth of charge / discharge (0.0885) dominate, demonstrating the high level of attention grid-side energy storage places on rapid response capabilities and safety and environmental protection characteristics (see Table 1 for specific weight distribution). Based on this, the TOPSIS method combined with an applicability redistribution mechanism is used for baseline ranking. The results show that vanadium redox flow batteries rank first with a comprehensive score of 0.625380, followed closely by lithium iron phosphate batteries with 0.608583. To verify the reliability of the ranking, 500 Monte Carlo parameter perturbation tests were conducted. Statistical results showed that vanadium redox flow batteries ranked first with a frequency as high as 0.826 and an average ranking of 1.236, demonstrating extremely strong robustness; lithium iron phosphate batteries ranked second with a first-place frequency of 0.146 and an average ranking of 1.966 (see Table 2 for specific ranking and robustness data).

[0171] Further, scenario multipliers were introduced, and the PROMETHEE method was used to calculate the net flow and partial order relationship under two typical scenarios: peak shaving and frequency regulation, to reveal the advantages of different technologies under specific needs. The analysis results show that in the peak shaving scenario, which emphasizes energy-related indicators, vanadium redox flow batteries, with their long lifespan and large capacity, achieve a net flow of 0.141759, ranking first, while lithium iron phosphate batteries rank second with a net flow of 0.114720. However, in the frequency regulation scenario, which emphasizes power-related indicators, lithium iron phosphate batteries surpass vanadium redox flow batteries with their high power density and fast response characteristics, achieving a net flow of 0.168501, significantly better than vanadium redox flow batteries' 0.117990 (see Table 3 for specific scenario ranking data).

[0172] In this embodiment, the full-process output includes: CRITIC global weight and weight percentages summarized by first-level dimension (as shown in Table 1), TOPSIS baseline ranking and Monte Carlo robustness statistics (Top-1 frequency, average ranking) (as shown in Table 2), and positive, negative and net currents and scenario rankings for PROMETHEE peaking / frequency modulation (as shown in Table 3).

[0173] Table 1

[0174]

[0175] Table 2

[0176]

[0177] Table 3

[0178]

[0179] Therefore, this implementation plan, without changing the methodological structure, completes unified modeling, one-time weighting and solidification, cross-scenario partial order and quantitative verification, ensuring that the evaluation link is fair, objective and has engineering usability, and can directly support the selection, configuration and strategy comparison of energy storage technologies in peak shaving and frequency regulation scenarios.

[0180] In summary, this method constructs a comprehensive evaluation system for multiple energy storage technologies and multiple application scenarios based on a unified evaluation index system and an objective weighting method.

[0181] The evaluation effect and significant features of the comprehensive evaluation method of this invention are mainly reflected in three aspects: First, it provides a comprehensive score and ranking of various energy storage technologies in a global dimension to reflect their overall technical economy and operational performance level; second, it forms corresponding scenario net current, partial order relationship and substitutable chain for different application scenarios such as peak shaving and frequency regulation to reveal the relative advantages and applicable scope of each technology under different demand focuses; third, it obtains robustness indicators such as ranking distribution and frequency of first occurrence through uncertainty analysis to test the stability of the evaluation conclusions under parameter disturbance conditions.

[0182] This invention has been described through embodiments. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of this invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of this invention.

Claims

1. A comprehensive evaluation method for energy storage in multiple scenarios of grid-side peak shaving and frequency regulation, characterized in that, Includes the following steps: First, an evaluation index set and an original evaluation matrix are established to obtain multiple indicators for evaluating the overall effect of energy storage, and these indicators are arranged in a regular pattern. In obtaining the masked normalized matrix Then, the objective weight calculation method for indicators under complex information, CRITIC, is used to assign objective weights to each indicator based on the comparison strength between dispersion and decorrelation, and indicators with dispersion exceeding the first threshold and decorrelation exceeding the second threshold are removed according to the objective weights. Baseline sorting of each evaluation index sequence yields an evaluation index sequence sorted by proximity coefficient; The ordered evaluation index sequence is subjected to scenario partial ordering to obtain two sets of comprehensive preference evaluation index sequences with physical meaning: peak modulation and frequency modulation. A Monte Carlo test is performed on the comprehensive preference evaluation index sequence to obtain the index sequence ranked according to the baseline or the index sequence ranked according to the partial order.

2. The comprehensive energy storage evaluation method according to claim 1, characterized in that: Establishing the original evaluation matrix of the evaluation index set includes the following specific steps: To eliminate differences in dimensions and inconsistencies in statistical standards, unit conversion and standardization were first completed, and the data were quantile-trimmed and missing data were handled to obtain the processed sample. To suppress the amplifying effect of outliers on subsequent ranking, quantile pruning is employed: ； In the formula, , As an indicator The lower and upper quantiles, Use an empirical value of 0.01 to 0.

05. These are the cropped observations; When missing values ​​exist, bounded completion is performed, meaning values ​​are assigned within the permitted range. The midpoint of the quantile is preferred. Interpolation; And record data quality labels for each unit for audit traceability; Subsequently defined To synthesize the effective values ​​after "trimming or completing": —Non-missing or —Complete missing information; To achieve monotonicity with consistent direction, let a direction factor be defined. Corresponding to benefit-type and cost-type indicators respectively, we have: ； In the formula, This is the value after direction adjustment; , The endpoints of the quantiles in the corresponding directions; Based on consistent direction, interval normalization is performed to obtain a dimensionless and comparable standardized matrix. : ； In the formula, To avoid extreme situations Numerical stability term at time; And maintain a monotonic mapping with the original indication of superiority or inferiority; Considering cases where cross-technology incomparability or conditional applicability exist, an applicability mask matrix is ​​introduced: when technologies are comparable in terms of metrics, otherwise, the standardized output is written as follows: ； Here, is a masked standardized matrix; subsequent weighting and aggregation are only performed on the comparable intersection of samples, thereby ensuring consistent evaluation criteria while minimizing information loss; after the above processing, a consistent and robust data foundation with unified dimensions, consistent direction, and suppressed outliers can be formed, providing a consistent and robust data foundation for subsequent objective weighting, baseline ranking, and scenario partial ordering.

3. The comprehensive energy storage evaluation method according to claim 1, characterized in that: The specific steps for assigning objective weights to each indicator are as follows: Record No. The sample mean of each indicator on the comparable sample set is In the formula ; The sample standard deviation is: ； Standard deviation It is used to represent the discrimination of an indicator within a comparable range. The greater the discrimination, the higher the amount of marginal information provided for ranking. To measure decorrelation, in two indicators Common comparable sample set Calculate the Pearson correlation coefficient: ； In the formula, and In the set respectively If the two columns are perfectly linear and in the same or opposite direction, then... If statistically independent or linearly independent, Approaching zero; Combining discrimination and decorrelation, the first definition is... The comparative strength of the indicators: ； In the formula, when When it approaches 1, A value close to 0 indicates that this indicator is highly redundant with other indicators. It then contracts; when When the value is large and has low correlation with other indicators, The increase indicates that it contributes more to distinguishing technical solutions; The global objective weights are obtained by normalizing the contrast intensity: ； In the formula, This refers to the numerically stable term under extreme conditions. For a candidate technology, only on a subset of indicators that have data or applicability can a comparable intersection with other technologies be formed, and the weights of each indicator in the subset are renormalized according to the relative proportion of the original objective weights, so that the sum of the weights of each indicator in the subset remains 1. Record the sample The set of comparable metrics is Then the sample In terms of indicators The effective weights on the t-axis are defined as: ； This ensures that the sum of the weights of each sample within its own comparable intersection is 1, guaranteeing consistent evaluation criteria and conservation of total weight. Finally, the weighted matrix is ​​obtained. ； In the formula, This serves as the input for subsequent baseline ranking and scenario partial order calculations; The normalized value with a mask; Effective weights at the sample level.

4. The comprehensive evaluation method for energy storage according to claim 1, characterized in that: The specific steps of baseline sorting are as follows: To measure the degree of approximation to the ideal, the effective sample set of each indicator sequence is used. Define the ideal and antiideal solutions above: ； In the formula, Indicators The optimal weighted performance. Indicates the worst weighted performance; Both are determined solely by the valid data in that column and are unaffected by missing or incomparable cells; For any sample In their comparable intersection The Euclidean distance between the ideal and antiideal solutions is calculated above: ； In the formula, The smaller the value, the closer the sample is to the optimal value in each column. A larger value indicates that the sample is farthest from the worst in each column; if an individual indicator is in the sample... If the data is not comparable, then the indicator will naturally not be included in the summation, thus avoiding penalties or rewards due to missing data. Based on this, the closeness coefficient is defined as follows: ； In the formula This is a numerically stable term to prevent division by zero when both distances are simultaneously zero. The larger the value, the better. The closer to the ideal and the further away from the anti-ideal, the more it serves as the basis for ranking the global baseline.

5. The comprehensive energy storage evaluation method according to claim 1, characterized in that: The specific steps for partial ordering of scenarios are as follows: After obtaining the global objective weight Redistribution weights with in-sample comparable intersection Based on this, scene multiplier vectors are introduced. Expression scenario The focus of attention, in the formula They represent peak shaving and frequency modulation, respectively. The global weights are then recalibrated according to the scenario as follows: ； In the formula, For the scene Lowering the target Normalized weights; As an indicator Global objective weight; For the scene Subordination to indicators The focus is on the multiplier; the larger the value, the more important the indicator is in that scenario. Alternative options A set of applicable indicators with data. This is the pairwise comparable intersection of the two; For the scene The weighted normalization factor of all indicators in the pairwise comparable intersection set; where For indexing indicators; when season This indicates that the indicator is not included in the plan. and Paired comparisons; Weighted values ​​obtained from the baseline sorting step For utility, define option a relative to option b in terms of metrics. The difference on: ； All indicators have been aligned in direction after the steps of establishing the evaluation indicator set and the original evaluation matrix were completed. This indicates that scheme a is superior to scheme b on this indicator; using indifference intervals With preference threshold The V-shaped preference function yields the single-index preference: ； In the formula, This represents an interval where minute differences are negligible. The threshold is considered sufficiently good to be significantly better; both can be set based on historical operation and engineering experience and vary depending on the scenario. The pairwise preference index is obtained by aggregating the preferences of each indicator according to the scenario weight: ； In the formula, a larger value of π indicates a larger scene The stronger the overall preference for option a relative to option b; For each solution a, in the candidate set Calculate positive and negative preferences: ； And define the overall preference degree as the criterion for scene partial order: ； when At that time, in scenario a If b is better than b; if both are close to zero and their difference is not significant, then it reflects a partial order relation that cannot be judged; this can be determined by pairwise comparable intersections. With scene weight Through normalization, PROMETHEE can maintain a consistent comparison standard even when there are incomparable dimensions and missing data; combined with a threshold function Suppressing minute differences and amplifying significant differences can output two sets of comprehensive preference conclusions with physical meaning: peak modulation and frequency modulation.

6. The comprehensive evaluation method for energy storage according to claim 1, characterized in that: The specific steps of the Monte Carlo robustness test are as follows: To assess the sensitivity of weighting and ranking to parameter uncertainty, let a nominal parameter vector be defined. , respectively represent energy investment, power investment, round-trip efficiency, lifetime equivalent conversion factor, and indifference and preference threshold of the scenario preference function; Randomly perturb each component according to the engineering interval to form the first... Secondary sample : , , , Missing units are sampled within the permissible range of the indicator. Each sampling is performed by completely recalculating steps S1 to S4 to obtain the scheme. Baseline closeness Or scene net flow And generate rankings based on this. ; Let the total number of repetitions be... ; Robust output first records the frequency of ranking first and the average ranking: ； In the formula For indicator functions, Representation scheme The proportion that ranks first among all disturbances. For the plan The average relative position; If the baseline ranking is used, then Depend on Obtained by decreasing; if scenario ranking is used, then by Obtained by decreasing.

7. A comprehensive evaluation device for energy storage in multiple scenarios of grid-side peak shaving and frequency regulation, characterized in that, include: The evaluation index set and original evaluation matrix establishment module is used to establish the evaluation index set and the original evaluation matrix, which are used to obtain multiple indicators for evaluating the comprehensive effect of energy storage and arrange them in a regular pattern. The weight allocation module is used to obtain a masked normalized matrix. Then, the objective weight calculation method for indicators under complex information, CRITIC, is used to assign objective weights to each indicator based on the comparison strength between dispersion and decorrelation, and indicators with dispersion exceeding the first threshold and decorrelation exceeding the second threshold are removed according to the objective weights. The baseline sorting module is used to perform baseline sorting on each evaluation index sequence to obtain an evaluation index sequence sorted according to the closeness coefficient. The scenario partial ordering module is used to perform scenario partial ordering on the sorted evaluation index sequence to obtain two sets of comprehensive preference evaluation index sequences with physical meaning: peak adjustment and frequency adjustment. The testing module is used to perform Monte Carlo tests on the comprehensive preference evaluation index sequence to obtain the index sequence ranked according to the baseline or ranked according to the partial order.

8. The comprehensive evaluation device for energy storage in multiple scenarios of grid-side peak shaving and frequency regulation according to claim 7, characterized in that, The baseline sorting module includes: Solve the defined unit, and define the effective sample set as the ideal solution and the anti-ideal solution; The Euclidean distance calculation unit is used to calculate the Euclidean distance from any sample to the ideal solution and the antiideal solution.

9. The comprehensive evaluation device for energy storage in multiple scenarios of grid-side peak shaving and frequency regulation according to claim 7, characterized in that, The scenario partial order module includes: The scenario calibration unit is used to calibrate the global weights in both frequency modulation and peak modulation scenarios. The preference index aggregation unit is used to aggregate the preference indices of each sample to obtain the preference degree, thereby obtaining the comprehensive preference degree.

10. A computer-readable medium comprising a memory and a processor, wherein the memory stores the method of claims 1-6, and the processor performs the steps of the method of claims 1-6.

Images