Carbon market clearing mode evaluation method, device and equipment and storage medium

Abstract

The invention discloses an evaluation method, device and equipment for a clearing mode of a carbon market and a storage medium, and relates to the field of energy clearing modes and evaluation methods, and the method specifically comprises the steps: extracting high-frequency keywords for describing the clearing mode of the carbon market through a big data AI technology, and constructing an evaluation index system; performing translation correction on a zero value or a negative value in the index data, calculating a skewness coefficient after identifying and processing an abnormal value, correcting data skewness by adopting Box-Cox transformation when the skewness coefficient is greater than or equal to 0.5, and eliminating dimensional difference in combination with Z-score standardization; the subjective weight of the expert and the objective weight generated by the entropy weight method are obtained in combination with the Delphi method, and the comprehensive weight is generated through linear weighting; and constructing a weighted standardized matrix, calculating the distance and proximity between each scheme and the positive or negative ideal, and determining the optimal clearing scheme according to the proximity sequence. The technical problem of large data deviation in market clearing mode evaluation is solved, and the market resource allocation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon market trading, and particularly relates to an evaluation method, device, equipment and storage medium for a carbon market clearing method. Background Art

[0002] Carbon market clearing is to determine the equilibrium price and allocation volume of carbon emission allowances (CEAs) through supply-demand matching, and its core goal is to minimize the social emission reduction cost. The main processes include: Quota allocation: Initially, free allocation is mainly adopted (based on the industry benchmark method or historical emission method), and paid auctions are gradually introduced. For example, China currently adopts the benchmark line method for the power industry, covering approximately 5.1 billion tons of carbon emissions. Market trading: The centralized bidding mode is adopted, which is divided into listed agreement trading and block agreement trading, and the latter accounts for up to 83%. The clearing algorithm is mostly based on economic dispatch principles (such as the security-constrained unit commitment model), giving priority to meeting the power device balance demand rather than physical contract tracking. Compliance supervision: Enterprises need to submit allowances or nationally certified voluntary emission reductions (CCERs) that match the actual emissions within the compliance cycle, and those who fail to do so need to pay fines or purchase supplementary allowances.

[0003] The coupling mechanism between the carbon market and the power market promotes low-carbon development through the interaction of carbon quota constraints and price signals; the cap-and-trade mechanism sets a total carbon emission limit and allows quota trading to incentivize emission reduction; technologies such as dynamic carbon trading curves and two-layer optimization models accurately reflect supply-demand relationships and optimize market clearing; and the new energy consumption incentive mechanism promotes the utilization of clean energy and reduces carbon emissions. These technologies provide a scientific basis and technical support for carbon market clearing, and promote the efficient operation and sustainable development of the carbon market.

[0004] Although the current carbon market clearing and evaluation technologies have made progress in the institutional framework and basic tools, they still face core bottlenecks such as poor data quality. For example, Chinese Patent Publication No. CN 118396336 A discloses a multi-dimensional fine evaluation method for distributed energy systems based on the combined weighting method. Although a relatively objective multi-dimensional accuracy evaluation scheme is established, in step two, the energy efficiency, gross profit, economic efficiency, and plant electricity rate in the secondary index "economy" are further analyzed through the system clustering method. Before performing the clustering analysis, the Z-score method is used to standardize the samples. The method of using Z-score alone for standardization can only retain the original distribution skewness in terms of distribution form, is vulnerable to extreme values in terms of outlier resistance for linear scaling, and is only applicable to linear models, unable to cope with non-linear or machine learning models, and can only handle policy changes statically, making it difficult to reflect policy mutations. Summary of the Invention

[0005] The technical problem to be solved by the present invention is how to solve the problems of poor data quality and inability to dynamically respond to parameter changes in the existing methods.

[0006] The present invention solves the above technical problems through the following technical means: An evaluation method for the clearing method of the carbon market, the method comprising the following steps:

[0007] Step S1, constructing evaluation indicators for the clearing method of the carbon market: extracting high-frequency keywords describing the clearing method of the carbon market through big data AI and performing frequency ranking;

[0008] Step S2, performing standardization processing on the constructed evaluation indicators: performing translation correction on zero or negative values in the evaluation indicator values, detecting and processing outliers through Tukey's Fences, calculating the skewness coefficient Skewness, if |Skewness|≥ the set value, correcting the skewness through Box-Cox transformation, and performing Z-score standardization on the transformed data to eliminate the dimension difference;

[0009] Step S3, combining subjective and objective weighting: obtaining the subjective weight of experts through the Delphi method, combining the objective weight calculated by the entropy weight method, and generating a comprehensive weight by linear weighting;

[0010] Step S4, TOPSIS comprehensive evaluation: constructing a weighted standardized matrix, calculating the distances and proximities of each scheme to the positive and negative ideals, and determining the optimal clearing method according to the proximity ranking.

[0011] The present invention extracts high-frequency keywords through big data AI and ranks them, which can quickly identify the core influencing factors of the clearing method of the carbon market, avoid the subjectivity and inefficiency of manual screening, and improve the objectivity and comprehensiveness of the index system.

[0012] Further, the specific standardization processing of the constructed evaluation indicators is as follows:

[0013] Performing translation correction on zero or negative values:

[0014]

[0015] where x ij is the j-th index value of the i-th sample, n is the total number of samples, m is the total number of indicators, x j is the index value of the j-th, and ∈ is a small positive number;

[0016] Tukey's Fences outlier detection: Defining the upper and lower limit thresholds of outliers based on the interquartile range:

[0017] Lower limit threshold: Lower = Q1 - 1.5 × IQR

[0018] Upper threshold: Upper = Q3 + 1.5 × IQR

[0019] Among them, the quartiles are the 25th percentile Q1 and the 75th percentile Q3 after sorting the data from small to large; the interquartile range IQR is the difference between Q3 and Q1, representing the range of the middle 50% of the data; IQR = Q3 - Q1; all data points less than Lower or greater than Upper are determined as outliers, replacing the values less than Lower with Lower, and replacing the values greater than Upper with Upper;

[0020] Calculate the skewness coefficient, and the formula is:

[0021]

[0022] Among them, is the mean value;

[0023] If |Skewness| ≥ the set value, perform the BOX-COX transformation:

[0024]

[0025] Among them, λ is the transformation parameter, and λ is solved by maximum likelihood estimation:

[0026]

[0027] Among them, is the estimated value of the variance;

[0028] If |Skewness| < the set value, then skip the steps of the BOX-COX transformation;

[0029] Z-score standardization eliminates the dimension difference:

[0030]

[0031] Among them, is the data after the jth index value of the ith sample in the original data is transformed by Box-Cox, and are the mean and standard deviation of the data after the Box-Cox transformation.

[0032] The present invention corrects zero values and negative values through translation, detects and processes outliers by Tukey's Fences to improve data robustness, corrects significant skewness through Box-Cox transformation to make the distribution symmetric, and eliminates dimension differences through Z-score standardization, ultimately ensuring reasonable data distribution and adaptation to statistical modeling, and improving the accuracy and reliability of analysis.

[0033] Furthermore, the specific method for obtaining the expert subjective weight through the Delphi method is:

[0034] Responsible for designing the first-round questionnaire, using the d-level Likert scale and open-ended suggestion columns to collect experts' scores on the importance of indicators, and conducting data preprocessing, removing extreme values, and calculating the mean and standard deviation;

[0035]

[0036] where s kj is the score of the k-th expert for the j-th indicator, and n1 is the total number of experts, represents the mean, represents the standard deviation;

[0037] Second-round questionnaire: Feedback the results of the first-round statistics, require experts to revise their scores, and calculate the coefficient of variation for consistency test:

[0038] Calculate the coefficient of variation Require

[0039] Third-round termination condition: If the convergence condition is met, then terminate the iteration weight determination and consistency test;

[0040] Use weighted average to calculate the weight:

[0041]

[0042] where C r,k is the authority coefficient of the k-th expert;

[0043] Consistency verification: Calculate the Kendall's coefficient of concordance W:

[0044]

[0045] where R j is the sum of the ranks given by experts for the j-th indicator, is the theoretical average rank sum, and when W≥0.5, it is considered that the experts' opinions are coordinated;

[0046] Through Monte Carlo simulation, perturb the experts' authority coefficients and scores, observe the weight change rate, and introduce a policy correction factor to correct the final subjective weight;

[0047] Monte Carlo simulation: Perturb the experts' authority coefficients and scores by ±10%, and if the observed weight change rate Δw≤5%, accept the result;

[0048] Final subjective weight: α is the policy-oriented adjustment coefficient.

[0049] Furthermore, the objective weight calculated by combining the entropy weight method is specifically:

[0050] Data preprocessing: For the data y after Z-score standardization ij Perform non-negative translation;

[0051] Index positive transformation: Transform the extremely small type index into an extremely large type index:

[0052] Negative index: x′ ij = max(x j ) - x ij

[0053] Positive index: x′ ij = x ij - min(x j )

[0054] Standardization processing:

[0055]

[0056] Calculate the entropy value;

[0057]

[0058] If p ij = 0, then define p ij lnp ij = 0;

[0059] Calculate the difference coefficient: d j = 1 - E j ;

[0060] Determine the objective weight:

[0061] Furthermore, the specific method for generating the comprehensive weight by linear weighting is as follows:

[0062] Adopt the linear weighting combination method, and the formula is:

[0063]

[0064] Among them, β is the subjective weight coefficient (0 ≤ β ≤ 1), which can be set according to requirements.

[0065] The present invention combines the Delphi method and the entropy weight method to assign weights, taking into account both expert experience and data objectivity. The weights are dynamically adjusted through Monte Carlo simulation and policy correction factors, which not only improves the scientificity of the evaluation but also enhances the adaptability of the model to policy guidance and market changes. Multiple rounds of feedback and coordination degree tests further ensure the stability and consistency of the weight results.

[0066] Furthermore, the specific TOPSIS comprehensive evaluation is as follows:

[0067] Perform range standardization on the original data. For positive indicators:

[0068]

[0069] For negative indicators:

[0070]

[0071] Apply the comprehensive weight to the standardized data:

[0072]

[0073] Among them,

[0074] Determine the positive and negative ideal solutions:

[0075] Positive ideal solution V+: For each indicator, take the optimal value. For positive indicators, take the maximum value, and for negative indicators, take the minimum value;

[0076] V + =(max(v i1 ), min(v i2 ), max(v i3 ), …)

[0077] Negative ideal V-: For each indicator, take the worst value. For positive indicators, take the minimum value, and for negative indicators, take the maximum value;

[0078] V - =(min(v i1 ), max(v i2 ), min(v i3 ), …)

[0079] Calculate the distance and proximity. Distance to the positive ideal solution:

[0080]

[0081] Distance to the negative ideal solution:

[0082]

[0083] Proximity:

[0084]

[0085] Sort in descending order according to the C i value. The larger the C i value, the better the clearance method.

[0086] The present invention quantifies the comprehensive advantages and disadvantages of each clearing method through the distance calculation of the TOPSIS method combined with the positive and negative ideal solutions, provides an intuitive sorting result, and supports multi-dimensional decision-making. By dynamically setting the weight adjustment coefficient β, the proportion of subjective and objective factors can be flexibly adjusted to meet the customized requirements of different scenarios.

[0087] The present invention also provides an evaluation device for the carbon market clearing method, and the device includes the following modules:

[0088] Evaluation index construction module: used to extract high-frequency keywords describing the carbon market clearing method through big data AI and perform frequency ranking;

[0089] Index standardization processing module: used to perform translation correction on zero or negative values in the evaluation index values, detect and process outliers through Tukey's Fences, calculate the skewness coefficient Skewness, and if |Skewness|≥ the set value, correct the skewness through Box-Cox transformation, and perform Z-score standardization on the transformed data to eliminate the dimension difference;

[0090] Subjective and objective combined weighting module: used to obtain the expert subjective weight through the Delphi method, combine the objective weight calculated by the entropy weight method, and generate a comprehensive weight by linear weighting;

[0091] TOPSIS comprehensive evaluation module: used to construct a weighted standardized matrix, calculate the distances and proximities of each scheme to the positive and negative ideals, and determine the optimal clearing method according to the proximity ranking.

[0092] Further, the specific standardization processing of the constructed evaluation index is as follows:

[0093] Translation correction for zero or negative values:

[0094]

[0095] where x ij is the j-th index value of the i-th sample, n is the total number of samples, m is the total number of indexes, x j is the j-th index value, and ∈ is a small positive number;

[0096] Tukey's Fences outlier detection: Define the upper and lower limit thresholds of outliers based on the interquartile range:

[0097] Lower limit threshold: Lower = Q1 - 1.5×IQR

[0098] Upper limit threshold: Upper = Q3 + 1.5×IQR

[0099] Among them, the quartiles are the 25th percentile Q1 and the 75th percentile Q3 after sorting the data from smallest to largest; the interquartile range IQR is the difference between Q3 and Q1, representing the range of the middle 50% of the data; IQR = Q3 - Q1; all data points less than Lower or greater than Upper are determined to be outliers, and the values less than Lower are replaced with Lower, and the values greater than Upper are replaced with Upper;

[0100] Calculate the skewness coefficient, and the formula is:

[0101]

[0102] Among them, is the mean value;

[0103] If |Skewness| ≥ the set value, perform the BOX-COX transformation:

[0104]

[0105] Among them, λ is the transformation parameter, and λ is solved by maximum likelihood estimation:

[0106]

[0107] Among them, is the estimated value of the variance;

[0108] If |Skewness| < the set value, then skip the steps of the BOX-COX transformation;

[0109] Z-score standardization eliminates the dimension difference

[0110]

[0111] Among them, is the data after the j-th index value of the i-th sample in the original data is transformed by Box-Cox, and are the mean and standard deviation of the data after the Box-Cox transformation.

[0112] The present invention also provides a processing device, including at least one processor and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor can execute the above method steps by calling the program instructions.

[0113] The present invention also provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the above method steps.

[0114] The advantages of the present invention are as follows: By using big data AI to extract high-frequency keywords, the objectivity and efficiency of the evaluation system are ensured; By translating and correcting zero or negative values in the data, Tukey's Fences is used to detect and process outliers, and Box-Cox transformation combined with Z-score standardization is adopted to eliminate skewness and dimensional differences, improving data reliability; Delphi method and entropy weight method are integrated, combining expert experience and data objectivity to dynamically optimize weights; Based on TOPSIS to quantify the pros and cons of the scheme for ranking, flexibly adjusting subjective and objective factors to support precise decision-making. It combines technological innovation and application value, providing an efficient, robust, and quantifiable decision-making framework for the design of carbon market mechanisms. Description of the Drawings

[0115] Figure 1 It is a flowchart of an evaluation method for a carbon market clearing method in Embodiment 1 of the present invention;

[0116] Figure 2 It is a flowchart for standardizing the constructed evaluation indicators of the present invention. Detailed Embodiments

[0117] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0118] Embodiment 1

[0119] As Figure 1 shown is a flowchart of an evaluation method for a carbon market clearing method, including the following steps:

[0120] Step S1: Construct evaluation indicators for the carbon market clearing method: Extract high-frequency keywords describing the carbon market clearing method through big data AI and rank them by frequency;

[0121] Take the top X high-frequency related words as the evaluation indicators of the carbon market clearing method evaluation system, and at the same time, X should be greater than (at least equal to) the number of evaluation points included in the evaluation index system.

[0122] The constructed carbon market evaluation index system consists of market operation efficiency, system stability, trading activity, social benefits, and clean energy contribution.

[0123] Among them, market operation efficiency includes some or all of the evaluation points: market concentration (reflecting the control power of leading enterprises in the industry), barrier strength (measuring the policy and technical obstacles faced by new entrants), supply and demand elasticity (reflecting the ability to dynamically adjust market supply and demand) and corporate competitiveness (quantifying the vitality of market entities through bidding and tendering winning rates).

[0124] System stability includes some or all of the evaluation points of price volatility (reflecting the level of market risk), energy supply availability (quantifying the sustainable carbon supply capacity), and service reliability (including the comprehensive operational stability of software and hardware systems).

[0125] Trading activity: some or all of the evaluation points in the total trading volume (representing the scale of capital flow), turnover rate (reflecting market heat through trading frequency), and price difference range (measuring market trading efficiency).

[0126] Social benefits include: cost optimization for energy purchasers, spillover effects of carbon technology innovation, and some or all evaluation points in the regional employment promotion index.

[0127] Clean energy contribution includes: some or all evaluation points in energy station energy supply optimization, etc.

[0128] By extracting high-frequency words and matching them with an indicator system, a carbon market clearing evaluation framework was constructed, encompassing five major modules, including market efficiency. However, the dimensionality of each indicator varies significantly—for example, there's an order of magnitude gap between total trading volume (in the billions of yuan) and price spread (in percentages). Direct aggregate analysis would distort the results.

[0129] Step S2: Standardize the constructed evaluation index: use translation correction for zero or negative values in the evaluation index value, detect and process outliers through Tukey's Fences, calculate the skewness coefficient, and correct the skewness through Box-Cox transformation if |Skewness|≥ the set value. Perform Z-score standardization on the transformed data to eliminate dimensional differences. In this embodiment, the set value is generally 0.5, such as Figure 2 As shown in .

[0130] like Figure 2 The figure shows a flow chart for standardizing the constructed evaluation indicators; specifically:

[0131] Apply translation correction to zero or negative values:

[0132]

[0133] Among them, x ij is the jth index value of the i-th sample, n is the total number of samples, m is the total number of indicators, x j is the jth index value, ∈ is a small positive number;

[0134] Tukey's Fences Outlier Detection: Define the upper and lower threshold values of outliers based on the interquartile range:

[0135] Lower threshold: Lower = Q1 - 1.5 × IQR

[0136] Upper threshold: Upper = Q3 + 1.5 × IQR

[0137] Among them, the quartiles are the 25th percentile Q1 and the 75th percentile Q3 after sorting the data from smallest to largest; the interquartile range IQR is the difference between Q3 and Q1, representing the range of the middle 50% of the data; IQR = Q3 - Q1; all data points less than Lower or greater than Upper are determined as outliers, replace the value less than Lower with Lower, and replace the value greater than Upper with Upper;

[0138] Calculate the skewness coefficient, the formula is:

[0139]

[0140] Among them, is the mean;

[0141] If |Skewness| ≥ 0.5, perform the BOX-COX transformation:

[0142]

[0143] According to statistical experience rules, the threshold division of the absolute skewness (|Skewness|) is as follows:

[0144] |Skewness| < 0.5: The data is approximately symmetric and no transformation is required;

[0145] 0.5 ≤ |Skewness| < 1: Moderate skewness, need to adjust through the Box-Cox transformation;

[0146] |Skewness| ≥ 1: Severe skewness, data transformation must be performed.

[0147] This method selects 0.5 as the skewness threshold to ensure effective correction of moderately skewed and above data and improve the robustness of subsequent analysis.

[0148] Among them, λ is the transformation parameter, and λ is solved by maximum likelihood estimation:

[0149]

[0150] Among them, is the estimated value of the variance;

[0151] If |Skewness| < 0.5, skip the steps of Box-Cox transformation;

[0152] Z-score standardization eliminates the dimension difference:

[0153]

[0154] where is the data after Box-Cox transformation of the j-th index value of the i-th sample in the original data, and are the mean and standard deviation of the data after Box-Cox transformation.

[0155] Step S3, subjective and objective combined weighting: Obtain the expert subjective weight through the Delphi method, combine it with the objective weight calculated by the entropy weight method, and use linear weighting to generate the comprehensive weight;

[0156] The specific method of obtaining the expert subjective weight through the Delphi method is as follows:

[0157] Responsible for designing the first-round questionnaire, using the d-level Likert scale and the open-ended suggestion column to collect the experts' scores on the importance of the indicators, and conducting data preprocessing, removing extreme values, and calculating the mean and standard deviation; here d is taken as 9 because in the Delphi method, the 9-level Likert scale (1 = extremely unimportant, 9 = extremely important) is widely used for the following reasons:

[0158] High discrimination: Compared with the 5-level or 7-level scale, the 9-level can more finely distinguish the subtle differences in the importance of indicators by experts.

[0159] Statistical requirements: The odd-numbered grading allows experts to choose the middle value (such as the 5-level), avoiding forced bias, and at the same time the 9-level can meet the requirements of high-order statistical models (such as the entropy weight method) for data distribution.

[0160] Industry convention: The 9-level scale is commonly used in energy field research, which is convenient for horizontal comparison.

[0161] Questionnaire structure: Use the 9-level Likert scale (1 = extremely unimportant, 9 = extremely important) to score the importance of the indicators and set an open-ended suggestion column

[0162] Example question:

[0163] "Please evaluate the impact degree of 'clearing price volatility' on the stability of the carbon market: □1□2□3□4□5□6□7□8□9"

[0164] Data preprocessing:

[0165] Remove extreme values (such as data exceeding the mean ± 3 times the standard deviation), and calculate the mean and standard deviation of the scores of each indicator:

[0166]

[0167] Among them, s kj is the score given by the k-th expert for the j-th index, and n1 is the total number of experts. represents the mean value, represents the standard deviation;

[0168] Second-round questionnaire: Feedback the statistical results of the first round, require experts to revise their scores, and calculate the coefficient of variation for consistency test:

[0169] Calculate the coefficient of variation Require

[0170] Anonymously feedback the statistical results (mean value, standard deviation) of the first round and the initial scores of the experts to the experts. The experts adjust their scores according to the group opinions (for example, if their own scores are quite different from the mean value, they need to explain the reasons or make corrections).

[0171] Termination condition for the third round: If the convergence condition is met then terminate the determination of iterative weights and consistency test;

[0172] Use weighted average to calculate the weights:

[0173]

[0174] Among them, C r,k is the authority coefficient of the k-th expert;

[0175] Consistency verification: Calculate the Kendall concordance coefficient W:

[0176]

[0177] Among them, R j is the sum of the ranks given by the experts for the j-th index, is the theoretical average rank sum. When W≥0.5, it is considered that the experts' opinions are coordinated;

[0178] Through Monte Carlo simulation, perturb the experts' authority coefficients and scores, observe the change rate of weights, and introduce a policy correction factor to correct the weights;

[0179] Monte Carlo simulation: Perturb the experts' authority coefficients and scores by ±10%, and accept the results if the change rate of weights Δw≤5%;

[0180] Final subjective weight: α is the policy-oriented adjustment coefficient.

[0181] After three rounds of expert consultations, the coefficient of variation (CV) of each index weight converged to less than 0.3, and the Kendall's coefficient of concordance W reached 0.68, indicating that significant consensus was reached among experts. However, pure subjective weighting may overlook the implicit laws in market operation.

[0182] Coefficient of variation (CV): CV = standard deviation / mean. The requirement that CV ≤ 0.3 is to ensure the convergence of expert scores. This threshold refers to the "Delphi Method Operating Specification" (ISO / TR 19478). When CV ≤ 0.3, the opinions of the expert group tend to be consistent.

[0183] Kendall's coefficient of concordance W = 0.68: Calculated after three rounds of Delphi method iteration. W ≥ 0.5 indicates significant consistency among expert opinions (p < 0.05). 0.68 indicates moderate or above coordination, meeting the statistical requirements.

[0184] The objective weight calculated by combining with the entropy weight method is specifically as follows:

[0185] Data preprocessing: For the data y after Z - score standardization ij Perform non - negative translation;

[0186] Index positive - orientation: Convert the extremely small - type index into an extremely large - type index:

[0187] Negative - oriented index: x′ ij = max(x j ) - x ij

[0188] Positive - oriented index: x′ ij = x ij - min(x j )

[0189] Standardization processing:

[0190]

[0191] Calculate the entropy value:

[0192]

[0193] If p ij = 0, then define p ij lnp ij = 0;

[0194] Calculate the difference coefficient: d j = 1 - E j ;

[0195] Determine the objective weight:

[0196] The linear weighting method for generating the comprehensive weight is specifically:

[0197] The linear weighted combination method is used, and the formula is:

[0198]

[0199] Among them, β is the subjective weight coefficient (0≤β≤1), which can be set according to needs.

[0200] Step S4, TOPSIS comprehensive evaluation: construct a weighted standardized matrix, calculate the distance and proximity of each plan to the positive and negative ideals, and determine the optimal clearing method based on the proximity ranking.

[0201] If the original data is normalized by range, the positive indicator is:

[0202]

[0203] Negative indicators:

[0204]

[0205] Apply composite weights to the normalized data:

[0206]

[0207] in,

[0208] Determine the positive and negative ideal solutions:

[0209] Positive ideal solution V+: Each indicator takes the optimal value, the positive indicator takes the maximum value, and the negative indicator takes the minimum value;

[0210] V + =(max(v h1 ),min(v i2 ),max(v i3 ),…)

[0211] Negative ideal V-: each indicator takes the worst value, the positive indicator takes the minimum value, and the negative indicator takes the maximum value;

[0212] V - =(min(v i1 ),max(v i2 ),min(v u3 ),…)

[0213] Compute distance and proximity, distance to a positive ideal solution:

[0214]

[0215] Distance to the negative ideal solution:

[0216]

[0217] Proximity:

[0218]

[0219] Press C i Sort the values from high to low, C i The larger the value of C, the better the clearing method.

[0220] Table 1 is the classification of indicator directions

[0221]

[0222] Example 2

[0223] Suppose there are 3 carbon market clearing schemes (A, B, C) and 5 evaluation indicators: Indicator 1: Market concentration (positive)

[0224] Indicator 2: Price volatility (negative)

[0225] Indicator 3: Total trading volume (positive)

[0226] Indicator 4: Optimization of the cost for energy purchasers (positive)

[0227] Indicator 5: Optimization of energy supply at the energy station (positive)

[0228] Table 2 is the original data

[0229] Solution Index 1 Index 2 Index 3 Index 4 Index 5 A 80 15 5000 70 90 B 70 20 6000 80 80 C 60 18 4500 65 85

[0230] The original data are all positive values and do not require translation correction, and all data are within a reasonable range without outliers; the skewness coefficients of all indicators are < 0.5 and no transformation is required.

[0231] Perform Z-score standardization on each indicator for different schemes:

[0232]

[0233] Calculate the mean and standard deviation of each indicator and perform standardization:

[0234] Indicator 1 (Market concentration):

[0235] Mean = 70, Standard deviation ≈ 8.164

[0236] Standardized value: A = 1.225, B = 0, C = -1.225

[0237] Indicator 2 (Price volatility):

[0238] Mean ≈ 17.67, Standard deviation ≈ 2.055

[0239] Normalized values: A ≈ -1.299, B ≈ 1.133, C ≈ 0.162

[0240] Indicator 3 (Total trading volume):

[0241] Mean ≈ 5166.67, Standard deviation ≈ 623.61

[0242] Normalized values: A ≈ -0.267, B ≈ 1.337, C ≈ -1.070

[0243] Indicator 4 (Optimization of the energy purchaser's cost):

[0244] Mean ≈ 71.67, Standard deviation ≈ 6.236

[0245] Normalized values: A ≈ -0.267, B ≈ 1.336, C ≈ -1.070

[0246] Indicator 5 (Optimization of the energy station's energy supply):

[0247] Mean = 85, Standard deviation ≈ 4.082

[0248] Normalized values: A ≈ 1.225, B ≈ -1.225, C = 0

[0249] Table 3 shows the normalized results of each indicator for different scenarios

[0250] Solution Index 1 Index 2 Index 3 Index 4 Index 5 A 1.225 -1.299 -0.267 -0.267 1.225 B 0 1.133 1.337 1.336 -1.225 C -1.225 0.162 -1.070 -1.070 0

[0251] Step 2: Calculate the objective weights using the entropy weight method

[0252] Non - negative translation and positive normalization:

[0253] Indicator 2 (negative): x' ij = max(x j ) - x uh

[0254] max(x2) = 1.133

[0255] Other indicators (positive): x' ij = x ij - min(x j )

[0256] min(x1) = 1.225, min(x3) = -1.070, min(x4) = -1.070, min(x5) = -1.225 Table 4 shows the data after positive normalization

[0257] Solution Index 1 Index 2 Index 3 Index 4 Index 5 A 2.450 2.432 0.803 0.803 2.450 B 1.225 0 2.407 2.406 0 C 0 0.971 0 0 1.225

[0258] Table 5 shows the normalized data after positive processing

[0259] Solution Index 1 Index 2 Index 3 Index 4 Index 5 A 0.667 0.715 0.250 0.250 0.667 B 0.333 0 0.750 0.750 0 C 0 0.285 0 0 0.333

[0260] After normalizing each index, the entropy value (E j ) and the coefficient of variation (d j ) are calculated

[0261]

[0262] d j = 1 - E j

[0263] Index 1: E1 = -0.579, d1 = 0.421

[0264] Index 2: E2 = 0.545, d2 = 0.455

[0265] Index 3: E3 = 0.512, d3 = 0.488

[0266] Index 4: E4 = 0.512, d4 = 0.488

[0267] Index 5: E5 = 0.579, d5 = 0.421

[0268] Total coefficient of variation = 0.421 + 0.455 + 0.488 + 0.488 + 0.450 = 2.273

[0269]

[0270] Objective weight:

[0271] Weight ≈ [0.185, 0.200, 0.214, 0.214, 0.185]

[0272] The Delphi method assumes subjective weight

[0273] Since the expert group's scores were not truly obtained, assumptions were used. Assume the subjective weights of the experts are: [0.2, 0.15, 0.25, 0.3, 0.1]

[0274] Composite weight (β = 0.5)

[0275] Composite weight = 0.5 × subjective weight + 0.5 × objective weight:

[0276] Composite weight ≈ [0.1925, 0.175, 0.232, 0.257, 0.1425]

[0277] TOPSIS evaluation

[0278] Range normalization:

[0279] Positive index:

[0280] Negative index:

[0281] Table 6 is the weighted normalization matrix

[0282] Solution Index 1 Index 2 Index 3 Index 4 Index 5 A 0.1925 0.175 0.077 0.086 0.1425 B 0.096 0 0.232 0.257 0 C 0 0.07 0 0 0.071

[0283] Table 7 Positive ideal solution and negative ideal solution of each index

[0284] Index Direction Positive Ideal Solution (V+) Negative Ideal Solution (V) Index 1 Positive 0.1925 0 Index 2 Negative 0 0.175 Index 3 Positive 0.232 0 Index 4 Positive 0.257 0 Index 5 Positive 0.1425 0

[0285] Calculate the distance and closeness, distance to the positive ideal solution:

[0286]

[0287] Distance to the negative ideal solution:

[0288]

[0289] Closeness:

[0290]

[0291] Closeness (C i ):

[0292] Scheme A: C A ≈0.479

[0293] Scheme B: C B ≈0.699

[0294] Scheme C: C C ≈"0.237"

[0295] Ranking result: B > A > C, optimal scheme: B (highest closeness)

[0296] Verified by three clearing schemes A, B, and C in a certain pilot carbon market, this method successfully outputs the quantitative evaluation results with C i values of 0.479, 0.699, and 0.237 respectively. Sensitivity analysis shows that when the adjustment coefficient β fluctuates in the range of [0.4, 0.6], the scheme ranking remains stable, verifying the reliability of the method. The finally constructed "index construction - data standardization - combined weighting - TOPSIS decision-making" technical chain provides an innovative method for the quantitative optimization of carbon market clearing methods.

[0297] Example 3

[0298] The present invention also provides an evaluation device for a carbon market clearing method, and the device includes the following modules:

[0299] Evaluation index construction module: used to extract high-frequency keywords describing the carbon market clearing method through big data AI and perform frequency ranking;

[0300] Index standardization processing module: used to perform translation correction on zero or negative values in the evaluation index values, detect and process outliers through Tukey's Fences, calculate the skewness coefficient Skewness, and if |Skewness| ≥ the set value, correct the skewness through Box-Cox transformation, and perform Z-score standardization on the transformed data to eliminate the dimension difference;

[0301] Subjective and objective combined weighting module: used to obtain the expert subjective weight through the Delphi method, combine the objective weight calculated by the entropy weight method, and generate the comprehensive weight by linear weighting;

[0302] TOPSIS comprehensive evaluation module: used to construct a weighted standardized matrix, calculate the distances and proximities of each scheme to the positive and negative ideals, and determine the optimal clearing method according to the proximity ranking.

[0303] Furthermore, the specific standardization processing of the constructed evaluation index is as follows:

[0304] Perform translation correction on zero or negative values:

[0305]

[0306] where \(x\) ij is the \(j\)-th index value of the \(i\)-th sample, \(n\) is the total number of samples, \(m\) is the total number of indexes, \(x\) j is the \(j\)-th index value, and \(\epsilon\) is a small positive number;

[0307] Tukey's Fences outlier detection: Define the upper and lower threshold values of outliers based on the interquartile range:

[0308] Lower threshold: Lower = Q1 - 1.5 × IQR

[0309] Upper threshold: Upper = Q3 + 1.5 × IQR

[0310] where the quartiles are the 25% quantile Q1 and the 75% quantile Q3 after sorting the data from small to large; the interquartile range IQR is the difference between Q3 and Q1, representing the range of the middle 50% of the data; IQR = Q3 - Q1; all data points less than Lower or greater than Upper are determined as outliers, replace the values less than Lower with Lower, and replace the values greater than Upper with Upper;

[0311] Calculate the skewness coefficient, and the formula is:

[0312]

[0313] Among them, is the mean value;

[0314] If |Skewness| ≥ set value, perform the BOX-COX transformation:

[0315]

[0316] Among them, λ is the transformation parameter, and λ is solved by maximum likelihood estimation:

[0317]

[0318] Among them, is the estimated value of the variance;

[0319] If |Skewness| < set value, skip the steps of the BOX-COX transformation;

[0320] Z-score standardization eliminates the dimension difference

[0321]

[0322] Among them, is the data after the j-th index value of the i-th sample in the original data is transformed by Box-Cox, and are the mean value and standard deviation of the data after the Box-Cox transformation.

[0323] Example 4

[0324] Based on Example 1, Example 4 of the present invention further provides a processing device, which is characterized in that it includes at least one processor and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor can execute the method steps described in Example 1 by calling the program instructions.

[0325] Example 5

[0326] Based on Example 1, Example 5 of the present invention further provides a computer-readable storage medium, which is characterized in that the computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the method steps described in Example 1.

[0327] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An evaluation method for a carbon market clearing method, characterized in that: The following steps are involved: Step S1: Constructing carbon market clearing method evaluation indicators: extracting high-frequency keywords describing carbon market clearing methods through big data AI and ranking them by frequency; Step S2: Standardize the constructed evaluation index: apply translation correction to zero or negative values in the evaluation index, detect and process outliers through Tukey's Fences, calculate the skewness coefficient, and correct the skewness through Box-Cox transformation if |Skewness| ≥ the set value. Perform Z-score standardization on the transformed data to eliminate dimensional differences. Step S3: subjective and objective combined weighting: The subjective weights of experts are obtained through the Delphi method, combined with the objective weights calculated by the entropy weight method, and a comprehensive weight is generated using linear weighting; Step S4, TOPSIS comprehensive evaluation: construct a weighted standardized matrix, calculate the distance and proximity of each plan to the positive and negative ideals, and determine the optimal clearing method based on the proximity ranking.

2. The evaluation method of a carbon market clearing method according to claim 1, characterized in that: The standardization process of the constructed evaluation indicators is specifically as follows: Apply translation correction to zero or negative values: where x ij is the j-th index value of the i-th sample, n is the total number of samples, m is the total number of indices, x j is the j-th index value, and ∈ is a tiny positive number; Tukey's Fences outlier detection: defines the upper and lower thresholds of outliers based on the interquartile range: Lower threshold: Lower = Q1 - 1.5 × IQR Upper threshold: Upper = Q3 + 1.5 × IQR Among them, the quartiles are the 25% quantile Q1 and the 75% quantile Q3 after sorting the data from small to large; the interquartile range IQR is the difference between Q3 and Q1, representing the range of the middle 50% of the data; IQR = Q3-Q1; all data points less than the Lower or greater than the Upper are determined to be outliers, and values less than the Lower are replaced with the Lower value, and values greater than the Upper are replaced with the Upper value; Calculate the skewness coefficient using the formula: Among them, is the mean value; If |Skewness|≥ the set value, perform BOX-COX transformation: Among them, λ is the transformation parameter, and λ is solved by maximum likelihood estimation: Among them, is the estimated value of the variance; If |Skewness| < the set value, skip the BOX-COX transformation step; Z-score standardization eliminates dimensional differences: wherein, is the data after Box-Cox transformation of the j-th index value of the i-th sample in the original data, and are the mean and standard deviation of the data after Box-Cox transformation.

3. The evaluation method of a carbon market clearing method according to claim 1, characterized in that: The specific method of obtaining the subjective weight of experts through the Delphi method is as follows: Responsible for designing the first round of questionnaires, using a D-level Likert scale and an open-ended suggestion column to collect experts' ratings on the importance of indicators, and performing data preprocessing, removing extreme values, and calculating the mean and standard deviation; where s kj is the score given by the k-th expert to the j-th indicator, and n1 is the total number of experts. represents the mean value, and represents the standard deviation. Second round of questionnaires: Feedback on the first round of statistical results, asking experts to revise their scores, and calculate the coefficient of variation for coordination test: Calculate the coefficient of variation Requirements The termination condition for the third round: If the convergence condition is satisfied then terminate the determination of the iterative weights and the consistency check; The weights are calculated using weighted average: Among them, C r,k is the authority coefficient of the k-th expert; Coordination verification: Calculate the Kendall coordination coefficient W: where R j is the sum of the ranks given by the j-th index expert, is the theoretical average rank sum, and when W≥0.5, it is considered that the experts' opinions are coordinated; The expert authority coefficient and score were disturbed through Monte Carlo simulation, the weight change rate was observed, and the policy correction factor was introduced to correct the final subjective weight; Monte Carlo simulation: perturb the expert authority coefficient and score by ±10%, and accept the result if the weight change rate Δw ≤ 5%; Final subjective weight: α is the policy-oriented adjustment coefficient.

4. The evaluation method of a carbon market clearing method according to claim 2, characterized in that: The objective weight calculated by combining the entropy weight method is specifically: Data preprocessing: The data y after Z-score standardization ij Perform non-negative translation; Indicator Positive: Converting extremely small indicators into extremely large indicators: Negative indicator: x' ij = max(x j ) - x ij Positive index: x' ij = x ij - min(x j ) Standardization: Calculate entropy; If p ij = 0, then define p ij ln p ij = 0; Calculate the coefficient of variation: d j = 1 - e j ; Determine the objective weight:

5. The evaluation method of a carbon market clearing method according to claim 1, characterized in that: The specific method of generating the comprehensive weight by linear weighting is as follows: The linear weighting combination method is adopted, and the formula is: Among them, β is the subjective weight coefficient (0 ≤ β ≤ 1), which can be set according to requirements.

6. The evaluation method of a carbon market clearing method according to claim 1, characterized in that: The specific TOPSIS comprehensive evaluation is as follows: Perform range standardization on the original data. For positive indicators: For negative indicators: Apply the comprehensive weight to the standardized data: Among them, Determine the positive and negative ideal solutions: Positive Ideal Solution V + : Each index takes the optimal value, the positive index takes the maximum value, and the negative index takes the minimum value; V + =(max(v i1 ),min(v i2 ),max(v i3 ),…) Negative ideal solution V - : For each index, take the worst value. For positive indices, take the minimum value, and for negative indices, take the maximum value; V - =(min(v i1 ),max(v i2 ),min(v u3 ),…) Calculate the distance and closeness. The distance to the positive ideal solution: The distance to the negative ideal solution: Closeness: Press C i Sort the values from high to low, C i The larger the value, the better the clearing method.

7. An evaluation device for a carbon market clearing method, characterized in that: The device includes the following modules: Evaluation index construction module: used to extract high-frequency keywords describing the carbon market clearing method through big data AI and perform frequency ranking; Index standardization processing module: used to perform translation correction on zero or negative values in the evaluation index values, detect and process outliers through Tukey's Fences, calculate the skewness coefficient Skewness, and if |Skewness| ≥ the set value, correct the skewness through Box-Cox transformation, and perform Z-score standardization on the transformed data to eliminate the dimension difference; Subjective and objective combined weighting module: used to obtain the expert subjective weight through the Delphi method, combine the objective weight calculated by the entropy weight method, and generate the comprehensive weight by linear weighting; TOPSIS comprehensive evaluation module: used to construct a weighted standardized matrix, calculate the distance and closeness of each scheme to the positive and negative ideals, and determine the optimal clearing method according to the closeness ranking.

8. An evaluation device for a carbon market clearing method according to claim 7, characterized in that: The specific method of standardizing the constructed evaluation index is as follows: Perform translation correction on zero or negative values: where x ij is the j-th index value of the i-th sample, n is the total number of samples, m is the total number of indices, and x j is the j-th index value, and ∈ is a tiny positive number; Tukey's Fences outlier detection: Define the upper and lower limit thresholds of outliers based on the interquartile range: Lower limit threshold: Lower = Q1 - 1.5 × IQR Upper limit threshold: Upper = Q3 + 1.5 × IQR Among them, the quartiles are the 25% quantile Q1 and the 75% quantile Q3 after sorting the data from small to large; the interquartile range IQR is the difference between Q3 and Q1, representing the range of the middle 50% of the data; IQR = Q3 - Q1; all data points less than Lower or greater than Upper are determined as outliers, replace the values less than Lower with Lower, and replace the values greater than Upper with Upper; Calculate the skewness coefficient, and the formula is: Among them, is the mean value; If |Skewness| ≥ the set value, perform BOX-COX transformation: Among them, λ is the transformation parameter, and λ is solved by maximum likelihood estimation: Among them, is the estimated value of the variance; If |Skewness| < the set value, skip the steps of BOX-COX transformation; Z-score standardization to eliminate the dimension difference where is the data after Box-Cox transformation of the j-th index value of the i-th sample in the original data, and are the mean and standard deviation of the data after Box-Cox transformation.

9. A processing device, characterized in that, Includes at least one processor and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor can execute the method according to any one of claims 1 to 6 by calling the program instructions.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the method according to any one of claims 1 to 6.

Images