Clean low-carbon evaluation method and system for electric power system based on RAM (Random Access Memory)

Abstract

The invention relates to the field of power system clean low-carbon evaluation, and discloses an RAM-based power system clean low-carbon evaluation method and system. The method comprises the following steps: collecting multi-source heterogeneous data in real time and carrying out exception cleaning and structured processing to construct a standardized data set; dynamically adjusting the weight based on an entropy weight method and expert experience, calculating an efficiency value by using an RAM model, and optimizing a constraint condition; recognizing input redundancy and output insufficiency by decomposing slack variables, and generating a high-value improvement list in combination with principal component analysis and analytic hierarchy process; and finally, matching the optimization strategy library to form a decision report and realizing closed-loop updating of the index library. The system can dynamically evaluate the clean low-carbon level of the electric power system, accurately identify energy efficiency improvement key points, realize evaluation-optimization-feedback full-process automatic management, and significantly improve the accuracy of an evaluation result and the effectiveness of decision support.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power system clean low carbon evaluation, and particularly relates to a RAM-based power system clean low carbon evaluation method and system. BACKGROUND

[0002] In the field of power system clean low carbon evaluation, the existing scheme usually evaluates the low carbon performance of the system based on static index analysis or a single efficiency evaluation model, and has limitations such as insufficient identification granularity of input redundancy, weak correlation of causes of insufficient output, and lack of dynamic adjustment capability. The existing method directly generates an efficiency value sequence based on a traditional efficiency calculation framework, which is prone to problems such as incomplete decomposition of slack variables, broken mapping of redundant items and insufficient output characteristics in the scene of multi-energy coupling and dynamic change of operating parameters of the power system, and it is difficult to meet the needs of accurate tracing and improvement path generation in clean low carbon evaluation. For the joint processing of slack variables and insufficient output characteristics in efficiency analysis data, the existing technology generally lacks a cooperative mechanism for mutual correlation analysis and principal component contribution calculation, and it is difficult to form a continuous link of data decomposition-feature extraction-contribution determination in the whole process of input redundancy identification-cause correlation analysis-improvement strategy output, resulting in low accuracy of the generation of the list of redundant indicators and the cause table of improvement points. SUMMARY

[0003] The present application provides a RAM-based power system clean low carbon evaluation method and system to solve the problem of how to accurately identify input redundancy and insufficient output in power system clean low carbon evaluation based on the reliable efficiency value sequence output by the RAM model efficiency calculation and analysis module through slack variable decomposition, mutual correlation analysis and principal component contribution calculation.

[0004] To solve the above technical problems, the present application provides a RAM-based power system clean low carbon evaluation method, comprising:

[0005] Real-time acquisition of energy input and output data from a power system multi-source heterogeneous data acquisition device, the data including energy input data, renewable energy generation and carbon management related data, performing abnormal cleaning based on statistical methods and rule engines, combining missing value filling, multi-source data redundancy verification and time series inference technology and time stamp alignment using interpolation and resampling technology, and generating a structured data set;

[0006] Based on the structured data set, the original indicators are extracted, the normalization processing is performed using interval scaling method to map to the standardized interval, the dynamic weight adjustment is based on entropy weight method and expert experience to dynamically update the weight coefficient, and the dimension verification is performed to check the integrity of the indicators and verify the consistency of the data, and a verified data set is generated;

[0007] The RAM model input matrix maps the column vectors of input indicators and output indicators, the efficiency value is calculated using a linear programming method, the model adaptability is optimized and adjusted, the constraint conditions and parameter configuration are checked based on statistical methods and machine learning techniques, and the output is a sequence of reliable efficiency values;

[0008] The relaxation variables of the decomposed reliable efficiency value are based on the linear programming solution of the RAM model, the input redundancy is identified to generate a redundancy indicator list and output deficiency, the cause correlation analysis is completed to extract the output deficiency characteristics and use cross-correlation analysis and principal component contribution calculation and priority sorting to generate a high-value improvement point list based on the analytic hierarchy process;

[0009] According to the high-value improvement point matching optimization strategy library, an optimization scheme is generated and feasibility evaluation and multi-objective optimization are performed to form a final optimization decision report;

[0010] Extract the indicator update signal in the decision report, perform indicator library version comparison, candidate indicator screening and weight redistribution, complete compatibility testing and closed-loop verification.

[0011] Further, the process of extracting original indicators based on structured data sets includes:

[0012] Through the configured data extraction interface, the original indicator data that meets the requirements of the indicator system is screened from the structured data set according to the pre-defined indicator classification rules;

[0013] The indicator classification rules are based on indicator names, data types and timestamp information, and support a processing mode that combines batch screening and single verification.

[0014] Further, the process of performing normalization processing specifically includes:

[0015] The interval scaling method is used to map the values of indicators with different dimensions and scales to a unified standardized interval;

[0016] According to the nature of the indicators, the expected output indicators and the non-expected output indicators are distinguished, and the maximum and minimum value normalization and reverse normalization strategies are used respectively;

[0017] The implementation of normalization processing is based on the sliding window technology, combined with the statistical characteristics of real-time data flow to dynamically adjust the normalization parameters.

[0018] Further, the dynamic weight adjustment includes:

[0019] The dynamic update of the weight coefficient is realized by integrating a multi-factor analysis model, which combines the objective weight calculation of the entropy weight method and the subjective weight correction of expert experience, supplemented by machine learning algorithm optimization weight adjustment strategy;

[0020] The updating process of the dynamic weight parameter has a triggering mechanism, which supports both timing updating and triggering updating based on data anomalies or system events.

[0021] Further, the dimension verification includes:

[0022] Performing index integrity checking, data consistency verification, and abnormal dimension elimination;

[0023] The dimension verification is based on the preset index system rules, automatically identifies missing or abnormal dimensions, and uses a combination of rule engines and data verification algorithms.

[0024] For the detected missing or abnormal dimensions, the system automatically generates a data maintenance request through a feedback mechanism.

[0025] Further, the process of decomposing the slack variable of the credible efficiency value based on the linear programming solution of the RAM model includes:

[0026] Standardize the multi-dimensional index data according to the input and output classification, and organize them into corresponding input and output vectors;

[0027] Through the data mapping rule engine, the rule engine automatically classifies and labels the index data according to the index name, type, and weight information.

[0028] Further, the calculation of the efficiency value includes:

[0029] Utilize the interval adjustment feature of the RAM model to dynamically adjust the weight distribution of inputs and outputs;

[0030] The efficiency value calculation process is implemented through an iterative optimization algorithm, using the interior point method or simplex method to solve linear programming problems;

[0031] The system monitors the calculation state in real time, automatically marks and recalculates abnormal calculation results.

[0032] Further, the sequence of credible efficiency values includes:

[0033] Map the slack variable vector to the specific dimensions of each input and output index, and normalize each dimension slack variable by combining index weight and dynamic adjustment parameters;

[0034] Filter noise and small fluctuations by setting threshold rules to select significant slack variables;

[0035] Generate a list of redundant indicators, detailing the redundant indicator name, redundancy size, and corresponding time period.

[0036] Further, the process of generating a list of high-value improvement points includes:

[0037] The slack variable is normalized as follows:

[0038]

[0039] wherein, is the normalized slack variable, is the original slack variable, μ i is the historical mean of the i a th input indicator of the a i th category, σ a is the standard deviation, i is the index of the input indicator, i 1 is the index of the a

[0040] Further, the system uses an improved Shannon entropy model to screen significant redundant items and calculate the information contribution degree of the redundant indicators:

[0041]

[0042] wherein, H 1 is the improved Shannon entropy, used to measure the significance of redundancy; is the normalized redundancy probability distribution; ∈ is a smoothing factor; n is the total number of input indicators.

[0043] is generated and then H 1 is calculated 1 When H 1 <θ 2 is a significant redundancy determination threshold.

[0044] Further, the time series data of the redundant items to be analyzed and the output indicator sequence are cross-correlated to calculate the time lag correlation coefficient of the redundant input and output deficiency:

[0045]

[0046] wherein, r b (τ) is the time lag correlation coefficient; t is the index of the time point; is the redundancy at the t th time point; is the arithmetic mean within the time window T; is the j b th output deficiency indicator; is the j a th output deficiency indicator at the t+τ th time point; τ is the time lag parameter, T is the length of the time window;

[0047] Further, a principal component analysis model is constructed to extract key cause features and principal component contribution degrees:

[0048]

[0049] in, The contribution of the k-th principal component; For the k-th principal component pair i a The loading coefficient of the redundancy index; k is the principal component index; m is the number of redundancy index categories.

[0050] Furthermore, the analytic hierarchy process (AHP) is used to calculate the overall priority of each improvement point, and a weight matrix is ​​defined:

[0051]

[0052] Among them, W 4 This is the weight matrix; The m represents the influence weight of the p-th causal feature on the q-th improvement point. ′ To improve the total number of points;

[0053] Furthermore, the objective function is constructed by optimizing the priority allocation using a linear programming model:

[0054]

[0055] Where: α 5 and β 5 These are the weighting coefficients. Let q be the decision variable for the q-th improvement point; The contribution of the qth principal component.

[0056] Furthermore, a RAM-based clean and low-carbon power system evaluation system, applied to any of the methods described above, includes:

[0057] The parameter limiting module is used to receive and parse installation parameters;

[0058] The data acquisition module is used to acquire unit output data and carbon emission intensity data;

[0059] The preprocessing module is used to generate normalized output sequences and load-related emission factors;

[0060] The spatiotemporal alignment module is used to construct spatiotemporally coupled datasets;

[0061] The arbitration decision module is used to output unit dispatch instructions;

[0062] The control execution module is used to adjust the output curve of the grid-connected generating unit;

[0063] The status update module is used to maintain dynamic evaluation benchmarks.

[0064] The key innovations of this invention include:

[0065] (1) By decomposing slack variables in efficiency analysis data, input redundancy is identified, a list of redundant indicators is generated, and a complete link from data decomposition to indicator generation is formed.

[0066] (2) Extract the features of insufficient output from the redundant items to be analyzed, perform causal correlation analysis, generate a table of causes of improvement points, and construct a collaborative mechanism for feature extraction and causal analysis.

[0067] (3) By combining cross-correlation analysis and principal component contribution calculation, a continuous process for causal correlation analysis and improvement strategy output is established to enhance the coherence and accuracy of data processing.

[0068] The following are its main beneficial effects:

[0069] (1) By decomposing the slack variables in the efficiency analysis data, the resulting list of redundant indicators can accurately identify the input redundancy in the power system, ensuring that the identification of redundant items is combined with the accuracy of the system's low-carbon evaluation, and is applicable to scenarios with multi-energy coupling and dynamic operating parameters.

[0070] (2) Extract the output deficiency features from the redundant items to be analyzed and perform causal correlation analysis. The generated improvement point causal table realizes the accurate correlation analysis of the causes of output deficiency, ensuring the effective extraction of causal data and the accuracy of causal analysis. It is suitable for generating improvement paths for clean and low-carbon evaluation of power systems.

[0071] (3) By combining cross-correlation analysis and principal component contribution calculation, the established causal correlation analysis and improvement strategy output process enhances the coherence and accuracy of data processing, solves the problem of redundant terms and insufficient output feature mapping in the background technology, and is suitable for improving the accuracy and operability of power system clean and low carbon evaluation. Attached Figure Description

[0072] Figure 1 A schematic flowchart of a RAM-based clean and low-carbon evaluation method for power systems provided in this application embodiment;

[0073] Figure 2 This is a structural block diagram of a RAM-based clean and low-carbon evaluation system for power systems, provided as an embodiment of this application. Detailed Implementation

[0074] Example 1: Refer to Figure 1 This is a schematic flowchart of a RAM-based clean and low-carbon evaluation method for power systems provided in an embodiment of the present invention. The process may include at least steps S100-S600:

[0075] S100. Real-time acquisition of energy input and output data from the multi-source heterogeneous data acquisition device of the power system. The data includes energy input data, renewable energy power generation and carbon governance related data. Anomaly cleaning is performed based on statistical methods and rule engine, missing value imputation combined with multi-source data redundancy verification and time series inference technology, and timestamp alignment using interpolation and resampling technology to generate a structured dataset.

[0076] S200: Extract original indicators based on structured datasets, perform normalization processing, map to standardized intervals using interval scaling method, dynamically adjust weights based on entropy weight method and expert experience, dynamically update weight coefficients and dimensions, perform indicator integrity checks and data consistency verification, and generate a verified dataset.

[0077] S300. Construct a RAM model input matrix that maps input and output indicators into column vectors. Calculate efficiency values ​​using linear programming methods. Perform adaptive optimization of the model, adjusting constraints and parameter configurations, and verifying confidence levels. Based on statistical methods and machine learning techniques, output a sequence of reliable efficiency values.

[0078] S400, decompose the slack variables of the credible efficiency value, solve the linear programming results of the RAM model, identify input redundancy, generate a list of redundant indicators and output deficiency, complete the causal correlation analysis to extract the characteristics of output deficiency, and use cross-correlation analysis and principal component contribution calculation and priority ranking to calculate the comprehensive priority based on the analytic hierarchy process, and generate a list of high-value improvement points.

[0079] S500: Based on the high-value improvement points, match the optimization strategy library, generate optimization solutions, conduct feasibility assessments and multi-objective optimizations, and form a final optimization decision report;

[0080] S600: Extract indicator update signals from the decision report, perform indicator library version comparison, candidate indicator screening and weight reallocation, and complete compatibility testing and closed-loop verification.

[0081] Step S100 includes at least steps S110-S130:

[0082] S110. Acquire multi-source heterogeneous data from the power system, perform outlier cleaning, and obtain the cleaned data stream.

[0083] Specifically, raw data is acquired from multi-source heterogeneous data acquisition devices in the power system. This multi-source heterogeneous data includes, but is not limited to, energy input data, renewable energy power generation, carbon governance-related data, and related output indicators. Specifically, by configuring a data acquisition interface, the multi-source heterogeneous data is connected to the data processing platform in real time. This interface supports multiple data formats and communication protocols, enabling unified access to structured, semi-structured, and unstructured data. Further, outlier cleaning is performed on the acquired raw data. This outlier cleaning is based on a combination of statistical methods and a rule engine. First, values ​​significantly exceeding reasonable ranges are initially screened by setting threshold rules. Then, a time-series anomaly detection algorithm based on historical data is used to identify potential abnormal fluctuations. For detected abnormal data, the system automatically marks and performs missing value imputation or anomaly removal. Imputation methods include linear interpolation and weighted averaging based on adjacent time windows. Anomaly removal is logged and data maintenance personnel are notified. All operations during the cleaning process are recorded in detail in a cleaning log database for easy subsequent traceability and auditing. Finally, the cleaned data is converted into a unified structured data stream, which includes fields such as timestamp, indicator name, value and data source. This data stream is then passed to the "data to be standardized" field of the dynamic indicator standardization processing module S210 for further processing.

[0084] S120. Extract energy input indicators from the cleaned data stream, fill in missing values, and generate a complete data sequence.

[0085] In step S120, the cleaned data stream is used as input, specifically extracting energy input indicator data, covering key indicators such as coal consumption, natural gas usage, renewable energy input, and carbon control-related inputs in power production. For the extracted energy input indicators, the system further performs missing value imputation processing. This imputation combines multi-source data redundancy verification and time series inference technology. First, linear or nonlinear interpolation is performed using historical data from adjacent time points. If the imputation effect does not meet the preset confidence level, redundant data sources for the relevant indicators are called for supplementary verification. Specifically, the system improves the accuracy and stability of missing value imputation by dynamically adjusting the interpolation window size and weight allocation. After processing, a complete data sequence is generated. This complete data sequence contains energy input indicator data without missing values ​​within a continuous time period, and anomalies and noise have been removed. This complete data sequence is used as the output field of this step and passed to the "raw indicator data" of the dynamic indicator standardization processing module S220 for subsequent normalization and weight update processing.

[0086] S130. Perform timestamp alignment on the complete data sequence to generate a structured dataset;

[0087] Specifically, timestamp alignment is performed on the complete data sequence to ensure consistency and synchronization of multi-source heterogeneous data in the time dimension. Timestamp alignment is achieved by defining a unified time granularity and time window, and employing interpolation and resampling techniques to adjust data from different sources and with different sampling frequencies to a unified time axis. Specifically, the system first identifies the timestamp format and missing time points of each data stream, then resamples the data based on the set time granularity, filling in missing time points using a time series completion algorithm to ensure the continuity and integrity of the time series. Further, the aligned data is categorized and summarized according to indicator types to form a structured dataset. This structured dataset contains multi-dimensional indicator data under a unified time axis, facilitating subsequent standardization processing. This structured dataset serves as the output field of this step and is passed to the "input data" of the dynamic indicator standardization processing module S210. The accuracy and completeness of the product of this step provide a solid data foundation for the analysis and calculation of subsequent modules such as S200, S300, and S400.

[0088] Step S200 includes at least steps S210-S230:

[0089] S210. Extract the original indicator data from the structured dataset, perform normalization processing, and obtain a standardized data matrix.

[0090] Specifically, the system extracts raw indicator data from the structured dataset, covering multi-dimensional data on energy input, renewable energy input, carbon control input, and expected and unexpected output indicators. Firstly, the system uses a configured data extraction interface to filter raw indicator data from the structured dataset according to predefined indicator classification rules, ensuring that the data meets the requirements of the indicator system. These indicator classification rules are based on indicator name, data type, and timestamp information, supporting a combination of batch filtering and single-item verification to ensure the completeness and accuracy of the extracted data. Further, the system performs normalization on the extracted raw indicator data, using an interval scaling method to map indicator values ​​of different dimensions and scales to a unified standardized interval, typically 0 to 1. Specifically, the system distinguishes between expected and unexpected output indicators based on their nature, employing maximum-minimum normalization and reverse normalization strategies respectively to ensure the comparability and reasonableness of the normalized data. During normalization, the system automatically identifies extreme values ​​and adjusts for abnormal fluctuations using a smoothing algorithm to prevent data distortion. The normalization process is implemented using a sliding window technique, dynamically adjusting the normalization parameters based on the statistical characteristics of the real-time data stream to adapt to the time-varying nature of the data. All calculation steps and parameter adjustments during the normalization process are recorded in a normalization log database for easy auditing and backtracking. After normalization, a standardized data matrix is ​​generated, containing multi-dimensional indicator data within a unified interval, maintaining the integrity of the time series and the correspondence between indicators. This standardized data matrix serves as the output field of this step and is passed to the "model input matrix" of the RAM model efficiency calculation and analysis module S310 for subsequent efficiency calculations. Furthermore, the accuracy and completeness of the standardized data matrix significantly impact the performance of the RAM model and the objectivity of the evaluation results.

[0091] S220. Extract dynamic weight parameters from the original indicator data, update the weight coefficients, and generate a dynamic weight table.

[0092] Step S220 uses the original indicator data as input to specifically extract dynamic weight parameters. These dynamic weight parameters dynamically adjust the weight coefficients based on real-time data trends and historical performance. Specifically, the system achieves dynamic updates of the weight coefficients by integrating a multi-factor analysis model. This model combines objective weight calculation using the entropy weight method with subjective weight correction based on expert experience, supplemented by machine learning algorithms to optimize the weight adjustment strategy. First, the system performs statistical feature analysis on the original indicator data, including correlation analysis between indicators, volatility assessment, and contribution calculation, forming basic data for weight adjustment. Subsequently, based on this basic data, the system calculates the objective weights of each indicator using the entropy weight method, reflecting the magnitude of the indicator's information content. Further, by combining weight correction rules from the expert experience database, the objective weights are adjusted to reflect domain knowledge and policy guidance. The weight correction rules automatically match indicator features through a rule engine and perform weighting or subtraction operations. Combined with machine learning algorithms, the system optimizes the parameters of the weight adjustment model based on historical evaluation results and actual operational feedback, improving the accuracy and adaptability of the dynamic weights. The update process of the dynamic weight parameters has a trigger mechanism, supporting both scheduled updates and updates triggered by data anomalies or system events. After the update is complete, a dynamic weight table is generated. This table reflects the weight distribution of the current evaluation index system and includes fields such as index name, weight value, and update time. This dynamic weight table, as an output field of this step, is passed to the "weight parameters" of the RAM model efficiency calculation and analysis module S320 to support the dynamic adaptability of efficiency calculation. The operation logs and adjustment records of the entire weight update process are stored in the weight management database for easy tracking and verification later.

[0093] S230. Perform dimensional verification on the standardized data matrix to generate a verified dataset.

[0094] In step S230, the standardized data matrix is ​​used as input, and the system further performs dimensional verification processing on it, specifically performing indicator integrity checks, data consistency verification, and abnormal dimension removal. The dimensional verification is based on preset indicator system rules, automatically identifying missing or abnormal dimensions. It employs a combination of a rule engine and a data verification algorithm. First, it verifies the integrity of each indicator column in the standardized data matrix to determine if any indicators have missing or invalid values. For detected missing or abnormal dimensions, the system automatically generates a data maintenance request through a feedback mechanism, notifying relevant personnel to supplement or correct the data. Further, the system performs data consistency verification, detecting the logical relationships and historical trends between indicators, and identifying abnormal fluctuations or data anomalies. Abnormal dimensions are marked using an anomaly detection algorithm, and whether to remove them is determined based on a preset threshold. The removal operation follows the principle of indicator system integrity to avoid data structure disorder caused by removal. During the dimensional verification process, the system dynamically adjusts the indicator selection strategy, balancing data quality and indicator coverage. After verification, a verified dataset is generated. This dataset ensures the integrity and standardization of the input data, containing standardized indicator data that have passed verification and their corresponding time series. This validation uses the dataset as the output field for this step, passing it to the "Validated Data" of the RAM model efficiency calculation and analysis module S310, providing effective assurance for subsequent efficiency calculations. Detailed results and processing records of the dimensional validation are stored in the data quality monitoring platform, supporting continuous data quality management and improvement.

[0095] Step S300 includes at least steps S310-S330:

[0096] S310. Construct a RAM model based on the model input matrix, calculate the efficiency value, and generate a preliminary efficiency value.

[0097] Specifically, the standardized data matrix is ​​first mapped to the input matrix of the RAM (Range Adjusted Measure) model. This mapping process includes structural transformation and format standardization of the indicator data. According to the preset model input specifications, the system organizes the multi-dimensional standardized indicator data into corresponding input and output vectors based on input and output categories. Specifically, the model input matrix construction module first identifies input indicators such as energy input, renewable energy input, and carbon control input, as well as expected and unexpected output indicators in the standardized data matrix. Following the requirements of the RAM model, the input indicators are arranged as column vectors of the input matrix, and the output indicators are arranged as column vectors of the output matrix. This process is implemented through a data mapping rule engine, which automatically classifies and labels the indicator data based on indicator name, type, and weight information to ensure the integrity and accuracy of the input matrix. Furthermore, the system performs data consistency checks on the input matrix, including missing value detection, outlier screening, and format validation. For detected outliers, preset anomaly handling strategies are used for correction or removal. All processing procedures are recorded in detail in the model input log database for easy subsequent traceability and auditing.

[0098] Based on the model input matrix, the RAM model calculation module initiates the efficiency value calculation process. Specifically, the RAM model employs a linear programming method to construct the objective function and constraints, calculating the efficiency value of each evaluation unit. During the efficiency value calculation process, the system automatically identifies potential areas of input redundancy and output insufficiency, and calculates the relative efficiency score of each evaluation unit through analysis of the relative relationship between input and output data. Specifically, for each evaluation unit, the system utilizes the interval adjustment characteristic of the RAM model to dynamically adjust the weight distribution of input and output, avoiding the bias of subjective weight setting in traditional DEA (Data Envelopment Analysis) models. The efficiency value calculation process is implemented through an iterative optimization algorithm, using the interior-point method or simplex method to solve the linear programming problem, ensuring the convergence and stability of the calculation process. During the calculation process, the system monitors the calculation status in real time, automatically marking and recalculating abnormal calculation results to ensure the reliability of the efficiency values. After completing the efficiency value calculation, the system generates a preliminary efficiency value dataset, containing the efficiency score of each evaluation unit and its corresponding input-output index mapping relationship. The preliminary efficiency value is passed to the "efficiency value to be optimized" in step S320 as the output field of this step, and is consumed by the RAM model efficiency calculation and analysis module in subsequent steps.

[0099] S320. Extract boundary condition parameters from the efficiency value to be optimized, perform adaptive optimization of the model, and generate the optimized efficiency value.

[0100] Step S320 uses the preliminary efficiency value output in step S310 as input, specifically extracting boundary condition parameters from the efficiency value to be optimized, and performing model adaptive optimization. This optimization process addresses potential boundary effects and insufficient model adaptability in the preliminary efficiency value calculation. The system improves the accuracy and applicability of the efficiency value by adjusting the constraints and parameter configurations of the RAM model. Specifically, the system first analyzes the distribution characteristics of the preliminary efficiency value, identifies efficiency value clusters and boundary units, and uses statistical analysis methods to estimate the confidence interval of the efficiency value to determine potential boundary condition parameters. Subsequently, based on the boundary condition parameters, the system dynamically adjusts the slack variable constraints and weight adjustment mechanisms in the RAM model, reconstructs the optimization objective function, and performs local optimization on the boundary units. This process is implemented through multiple iterations, each iteration including model parameter updates, linear programming solutions, and result evaluation, until the optimization objective reaches a preset convergence criterion or an upper limit on the number of iterations. Furthermore, the system automatically detects and corrects abnormal data and computational anomalies generated during the optimization process to ensure the stable operation of the model adaptive optimization. After optimization, an optimized efficiency value dataset is generated, containing the adjusted efficiency score and the corresponding model parameter configuration. The optimized efficiency value is used as the output field of this step and passed to the "final efficiency value" in step S330, and is consumed by the subsequent confidence verification processing module.

[0101] S330. Perform confidence verification on the final efficiency value to generate a sequence of reliable efficiency values;

[0102] Step S330 takes the optimized efficiency value output from step S320 as input and specifically performs confidence verification processing on the final efficiency value to generate a reliable efficiency value sequence. This confidence verification process is based on statistical methods and machine learning techniques to evaluate the reliability and stability of each efficiency value. Specifically, the system first performs distribution analysis on the final efficiency value sequence, calculating the mean, variance, and confidence interval, and identifying abnormal fluctuation points and extreme values. Further, the system combines historical efficiency data and operating environment parameters, using a supervised learning model to predict and verify the efficiency values, determining the confidence level of the efficiency values. During the confidence verification process, the system employs a multiple verification strategy, including cross-validation, time series analysis, and anomaly detection algorithms, to ensure the accuracy and consistency of the efficiency values. For efficiency values ​​with low confidence, the system automatically marks them and generates an anomaly report for subsequent focused analysis by the slack variable precise identification module. After verification, the system generates a reliable efficiency value sequence, including efficiency scores and their confidence indices that have been filtered and adjusted for confidence. The sequence of reliable efficiency values ​​is used as the output field of this step and is passed to the "efficiency analysis data" of the slack variable accurate identification module S410 to provide a data basis for the accurate identification of input redundancy and output insufficiency.

[0103] Step S400 includes at least steps S410-S430:

[0104] S410. Decompose slack variables from efficiency analysis data, identify input redundancy, and generate a list of redundant indicators.

[0105] Specifically, the system first decomposes the efficiency analysis data into slack variables. This decomposition process is based on the output of the RAM (Range Adjusted Measure) model, using the slack variable values ​​obtained during the linear programming solution process to correspond to the redundancy and insufficiency of each input and output indicator. Specifically, the system identifies redundant parts in the input dimension and insufficient parts in the output dimension by analyzing the slack variable data of each evaluation unit in the reliable efficiency value sequence. The decomposition process combines matrix operations with constraint analysis. First, the slack variable vector is mapped to the specific dimensions of each input and output indicator. Then, the slack variables in each dimension are normalized by combining indicator weights and dynamic adjustment parameters, facilitating subsequent accurate identification and comparison. Furthermore, the system filters significant slack variables by setting threshold rules, filtering noise and minor fluctuations to ensure that the identified redundant inputs and insufficient outputs have practical improvement value. The filtering rules are dynamically adjusted based on historical operating data and expert experience databases, supporting real-time adaptation to different power system operating environments. Specifically, for the selected significant input redundancy items, the system generates a list of redundancy indicators, which records the name of the redundancy indicator, the amount of redundancy, and the corresponding time period. This list is used as the output field "Redundancy Indicator List" for this step and is passed to the "Redundancy Items to be Analyzed" of the slack variable precise identification module S420 for subsequent use in the extraction of insufficient output features and the analysis of causes.

[0106] S420. Extract the characteristics of insufficient output from the redundant items to be analyzed, perform causal correlation analysis, and generate a table of causes of improvement points.

[0107] Based on the "redundant items to be analyzed" output from step S410 as input, specifically, the system extracts the characteristic information of insufficient output from the list of redundant indicators and conducts causal correlation analysis. This analysis process first performs data correlation matching on the output indicators corresponding to the redundant indicators, and constructs a causal relationship model between input redundancy and insufficient output by combining the power system's operation logs and environmental parameters. Specifically, the system uses multi-dimensional association rule mining technology and causal inference algorithms to analyze the statistical correlation and time series synchronicity between redundant input and insufficient output, identifying potential causal paths. Further, the system combines power system equipment status monitoring data, operation and maintenance records, and external environmental influencing factors, utilizing knowledge graphs and rule engines to perform causal knowledge base matching, enhancing the accuracy and comprehensiveness of the causal analysis. During the causal correlation analysis, the system dynamically adjusts the analysis model parameters, supporting multi-factor and multi-level comprehensive analysis, and can reveal complex internal relationships within the system. For the identified main causes, the system generates an improvement point causal table, recording in detail the specific manifestations of insufficient output, associated redundant input indicators, cause categories, and cause intensity scores. The improvement point cause table, as the output field of this step, is passed to the "cause data" of the slack variable accurate identification module S430, providing a basis for subsequent priority sorting and improvement point screening.

[0108] S430. Prioritize the causal data and generate a list of high-value improvement points.

[0109] Using the "causal data" output in step S420 as input, the system specifically prioritizes the improvement point causal table. This prioritization process is based on a multi-indicator comprehensive evaluation model, combining causal intensity scores, indicator influence ranges, improvement cost estimates, and potential benefit assessments. It employs a combination of the Analytic Hierarchy Process (AHP) and weighted scoring to calculate the comprehensive priority score for each improvement point. Specifically, the system first normalizes each factor to eliminate the influence of dimensions, then assigns weights to each factor according to a preset weight system. This weight system supports dynamic adjustment to adapt to the strategic priorities and operating environments of different power systems. Furthermore, the system uses a machine learning model to correct the priority ranking results by incorporating historical improvement project effect data, enhancing the scientific rigor and practicality of the ranking. During the ranking process, the system automatically identifies improvement points with similar priorities and performs cluster analysis to avoid redundant improvements and resource waste. The ranking results are generated in list form, including the improvement point name, comprehensive priority score, suggested improvement measures, and corresponding input-output indicator correlation information. The system filters high-value improvement points from the sorted list to form a list of high-value improvement points, which is used as the output field of this step and passed to the "optimization input data" of the clean and low-carbon evaluation and decision generation module S510 for subsequent optimization scheme formulation and decision support.

[0110] In another embodiment, in S410, slack variables are decomposed from the efficiency analysis data to identify input redundancy and generate a list of redundant indicators. Using the reliable efficiency value sequence output by the RAM model efficiency calculation and analysis module S330 as input, the system first parses the slack variable components in the efficiency analysis data. Specifically, based on the linear programming solution results of the RAM model, the system extracts the input slack variable vector for each evaluation unit, denoted as... Where i a Let a represent the input indicator of type a. Formula ① involves constructing slack variables and normalizing them:

[0111]

[0112] in, Normalized slack variables, μ is the original slack variable. i For the i-th a Historical average of similar input indicators, σ i Let i be the standard deviation, and i be the index of the input indicator. a : Index of input indicators of type a.

[0113] Extract the "redundancy field" from the "credible efficiency value sequence" and denote it as: Generated after normalization

[0114] Furthermore, the system employs an improved Shannon entropy model to screen for significantly redundant terms. Formula ② calculates the information contribution of the redundancy index:

[0115]

[0116] Among them, H 1 An improved Shannon entropy, used to measure the significance of redundancy; denoted as the normalized redundancy probability distribution; ∈ is the smoothing factor; and n is the total number of input indicators.

[0117] From formula ① generate Then calculate H 1 When H 1 <θ 1 When the threshold is derived from the policy table, it is considered significantly redundant, where θ 1 The threshold for determining significant redundancy is set. The system generates a list of redundancy indicators, including indicator name, redundancy amount, and timestamp, which is output as the "Redundancy Indicator List" field and consumed by the "Redundancy Items to be Analyzed" field of S420.

[0118] Furthermore, in S420, using the list of redundant indicators output by S410 as input, the system first extracts the time-series data of the redundant items to be analyzed and performs cross-correlation analysis with the output indicator sequence. Formula ③ calculates the time-delay correlation coefficient between redundant input and insufficient output:

[0119]

[0120] Where: r 2 (τ) represents the time-delay correlation coefficient; t: time point index; Let be the redundancy at time t; for The arithmetic mean within the time window T; For the jth b Indicators of insufficient output; For the j-th time at time t+τ b Indicator of insufficient output; τ is the time delay parameter, and T is the time window length; The arithmetic mean within the time window T.

[0121] Extracted from the 'Redundant Indicator List' Extracted from 'raw indicator data' Calculate r 2 (τ).

[0122] Based on the results of formula ③, the system constructs a principal component analysis (PCA) model to extract key causal features. Formula ④ defines the contribution of principal components:

[0123]

[0124] in, The contribution of the k-th principal component; For the k-th principal component pair i a The loading coefficient of the redundancy index; k is the principal component index; m is the number of redundancy index categories.

[0125] From formula ③ r 2 (τ) Construct the covariance matrix and generate it after PCA decomposition. filter The principal components (thresholds derived from the policy table) are used as causal features, where θ 2 The principal component contribution threshold is used to generate an improvement point cause table based on equipment status data. This table includes cause type, correlation strength, and time series matching degree. It is output as the "improvement point cause table" and consumed by the "cause data" of S430.

[0126] In S430, using the improvement point cause table output from S420 as input, the system employs the Analytic Hierarchy Process (AHP) to calculate the overall priority of each improvement point. Formula ⑤ defines the weight matrix:

[0127]

[0128] Among them, W 4 This is the weight matrix; The m represents the influence weight of the p-th causal feature on the q-th improvement point. ′ To improve the total number of points.

[0129] Extract the intensity of causal features from the "Cause Table of Improvement Points" and combine it with formula ④. Generate W 4 .

[0130] Furthermore, the system optimizes priority allocation using a linear programming model. Equation ⑥ constructs the objective function:

[0131]

[0132] Where: α 5 and β 5 These are the weighting coefficients. Let q be the decision variable for the q-th improvement point; The contribution of the qth principal component.

[0133] From formula ⑤ W 4 And formula ④ Input, and solve to generate the optimal filter The improvement points are used to form a list of high-value improvement points, which includes improvement measures, priority scores and resource requirements. This list is then passed as an output field to the "optimized input data" of the Clean and Low-Carbon Assessment and Decision Generation Module S510.

[0134] The technical effects of this section are as follows: By using slack variable decomposition, cross-correlation analysis, and principal component contribution calculation, a precise correlation between input redundancy and output insufficiency is achieved; by combining the analytic hierarchy process (AHP) and linear programming optimization, a list of high-value improvement points is generated, providing structured input for subsequent optimization decisions.

[0135] Step 500 includes at least steps S510-S530:

[0136] S510. Based on the optimized input data matching strategy library, generate an optimization scheme and generate a preliminary decision scheme;

[0137] Specifically, the system first receives the list of high-value improvement points, which includes the name of each improvement point, its comprehensive priority score, suggested improvement measures, and corresponding input-output indicator association information. Specifically, the system accesses a pre-defined optimization strategy library through a configured strategy library interface. This library contains predefined optimization strategy templates and execution schemes for different types of improvement points. Based on the characteristics and attributes of the high-value improvement points, the system matches the corresponding strategy templates, uses a rule engine to analyze the input redundancy and output insufficiency characteristics of the improvement points, and dynamically generates targeted optimization schemes by combining historical optimization cases and an expert knowledge base. Specifically, the optimization scheme generation process includes multi-stage processing. First, a detailed analysis of the input and output indicators of the improvement points is performed to identify optimization space and potential improvement paths. Then, based on the scheme templates in the strategy library, combined with the current power system operating status and resource constraints, customized operation steps, resource configuration, and expected effect descriptions are generated. The system further performs consistency verification on the generated preliminary decision scheme to ensure that the scheme content complies with power system operating specifications and safety standards. During the scheme generation process, the system supports multiple scheme alternatives and automatically selects the optimal scheme based on the priority of the improvement points and the applicability of the strategy. All generation processes and scheme versions are recorded in detail in the optimization scheme management database for easy auditing and scheme tracking. The preliminary decision scheme, as the output field of this step, is passed to the "Scheme to be Verified" of the clean and low-carbon evaluation and decision generation module S520 for subsequent feasibility analysis. At the same time, this output supports data interaction with the dynamic indicator system update module S600, promoting the dynamic improvement of the indicator library and the closed-loop feedback of optimization decisions.

[0138] S520. Extract feasibility parameters from the proposed solution, conduct a cost-benefit analysis, and generate a feasibility assessment report.

[0139] Step S520 uses the preliminary decision scheme output by the clean and low-carbon evaluation and decision generation module S510 as input. Specifically, the system extracts feasibility parameters from the preliminary decision scheme, covering multiple dimensions such as resource input costs, implementation period, technical feasibility, environmental impact, and expected economic benefits. Specifically, the system integrates a cost-benefit analysis module and uses a multi-indicator evaluation model to comprehensively evaluate the decision scheme. This model combines quantitative analysis and qualitative judgment. First, it performs a detailed calculation of resource input costs, including equipment investment, operation and maintenance costs, and human resource expenditures. The calculation process is based on real-time market price data and a historical cost database. Subsequently, the system evaluates technical feasibility, considering the technological maturity, implementation risks, and compatibility of the scheme, and uses an expert system and technical evaluation model for judgment. The environmental impact assessment combines power system carbon emission data and environmental monitoring information, using a life cycle analysis method to quantify the environmental benefits of the scheme. Economic benefit prediction is based on market demand, policy support, and electricity price trends, and is predicted through a dynamic simulation model. The system further weights and summarizes the various evaluation indicators to form a comprehensive feasibility score. During the assessment process, the system automatically identifies potential risk points, generates risk warning reports, and proposes improvement suggestions for solutions with higher risks. All assessment data, analysis processes, and results are stored in the feasibility assessment database, supporting subsequent review and optimization. The feasibility assessment report, as an output field of this step, is transmitted to the "Assessment Data" of the Clean and Low-Carbon Evaluation and Decision Generation Module S530 for multi-objective optimization processing. Simultaneously, this report provides an assessment basis for the Dynamic Indicator System Update Module S600, promoting the dynamic adjustment and optimization of the indicator system.

[0140] S530. Perform multi-objective optimization processing on the evaluation data and generate a final optimization decision report;

[0141] Step S530 uses the feasibility assessment report output by the clean and low-carbon evaluation and decision generation module S520 as input. Specifically, the system performs multi-objective optimization processing on the assessment data, comprehensively considering multiple objectives such as improving clean and low-carbon efficiency, cost control, environmental benefits, and implementation risks, to construct a multi-objective optimization model. This model employs advanced heuristic algorithms and mathematical programming methods. Specifically, the system first normalizes each objective function to eliminate the influence of different dimensions, and then dynamically adjusts the priority of each objective according to a preset weight system, supporting real-time response to strategy adjustments and the operating environment. During the optimization process, the system uses iterative search and local improvement strategies to find the optimal or near-optimal solution that meets the constraints, including resource limitations, technical specifications, and environmental standards. The system further integrates with the decision support module to simulate the implementation effects of different optimization schemes and evaluate the feasibility and stability of the schemes. For conflicting objectives that arise during the optimization process, the system uses the Pareto front analysis method to generate a multi-scheme selection set, supporting decision-makers to select schemes according to actual needs. The optimization results are formatted and output through the decision report generation module. The report includes detailed information on the final optimization decision, execution steps, expected effects, and risk assessment. All optimization process data and results are recorded in the optimization decision database for easy tracking and adjustment. The final optimization decision report, as the output field of this step, is transmitted to the "decision input data" of the dynamic indicator system update module S610, realizing closed-loop feedback between evaluation results and the indicator system, and supporting continuous optimization and dynamic adjustment of the clean and low-carbon development of the power system.

[0142] Step S600 includes at least steps S610-S630:

[0143] S610. Extract indicator system update signals from decision input data, compare indicator library versions, and generate indicator items to be updated.

[0144] Specifically, the system first extracts the indicator system update signal from the decision input data. This signal contains information on changes in indicator performance and potential improvement needs reflected in the latest optimization decision results. Specifically, the system accesses key information fields in the final optimization decision report through a configured indicator update interface, including the evaluation of the optimization scheme's implementation effect, changes in resource utilization, adjustments to carbon emissions, and changes in related environmental indicators. Further, the system performs version comparison processing on the indicator system update signal. This version comparison is based on pre-stored indicator library version information. Specifically, the system calls the indicator version management module to automatically retrieve the current indicator library's version number, indicator definition, and weight distribution, and compares the latest indicator update signal with historical version data item by item. During the comparison process, the system uses a difference detection algorithm to identify changes in indicator names, weight adjustments, indicator additions or deletions, and changes in indicator attributes. The difference detection combines text matching, numerical comparison, and semantic analysis techniques to ensure the comprehensiveness and accuracy of the version comparison. For detected indicator changes, the system performs a change impact assessment, analyzing the potential impact of the changes on the stability of the evaluation system and the calculation model. Combining historical data trends and an expert rule base, the system determines the rationality and necessity of the changes. All operation records for version comparison and impact assessment are stored in detail in the indicator version comparison log library for easy auditing and traceability. After the above processing is completed, the system generates indicator items to be updated. These indicator items include a list of newly added indicators, indicators whose weights need to be adjusted, and indicators to be deleted, with detailed records of indicator name, change type, change magnitude, and change time. This indicator item to be updated serves as the output field of this step and is passed to the "update candidate set" of the dynamic indicator system update module S620, providing basic data for subsequent indicator screening and weight redistribution. At the same time, this output is indirectly linked to the dynamic weight parameters of the dynamic indicator standardization processing module S220, realizing cross-module data feedback and closed-loop management.

[0145] S620. Select highly relevant indicators from the updated candidate set, perform weight redistribution, and generate a new version of the indicator library.

[0146] Based on the update candidate set output by the dynamic indicator system update module S610 as input, the system first performs correlation screening on the various indicators included in the update candidate set. This screening process combines statistical analysis and machine learning techniques. Specifically, the system calculates the correlation between candidate indicators and indicators in the existing indicator library, using methods such as Pearson correlation coefficient, mutual information, and principal component analysis to evaluate the information redundancy and complementarity between indicators. Further, the system combines power system operating environment data and historical evaluation results, using feature selection algorithms such as Recursive Feature Elimination (RFE) and LASSO regression to screen indicators with high contribution and strong correlation to clean and low-carbon evaluation. During the screening process, the system dynamically adjusts the correlation threshold and contribution weight to support flexible adaptation to different operating stages and strategic objectives. For the screened highly correlated indicators, the system performs weight redistribution processing. Specifically, based on the updated indicator set, the system recalculates the information entropy of each indicator using the entropy weight method to reflect its information content and discriminative power. Simultaneously, combined with the weight adjustment rules in the expert knowledge base, weights are weighted or corrected for specific indicator categories or policy orientations. The weight redistribution process is automatically matched and executed through a rules engine, supporting multiple rounds of iterative optimization. It combines historical weight adjustment records and feedback data to adjust the weight distribution, improving the dynamic adaptability and scientific rigor of the evaluation model. The system automatically detects and corrects abnormal fluctuations and extreme weight values ​​during the weight adjustment process to prevent weight imbalance. After weight redistribution, a new version of the indicator library is generated. This new version of the indicator library contains a complete list of indicators, updated weight distribution, and version information, supporting dynamic expansion and management of indicator attributes. This new version of the indicator library, as an output field of this step, is passed to the "Dynamic Weight Parameters" of the dynamic indicator standardization processing module S220 for subsequent weight updates. Simultaneously, this output is used by the dynamic indicator system update module S630 for compatibility testing and closed-loop verification. The operation logs, filtering results, and weight adjustment records of the entire process are stored in the indicator management database, providing data support for the continuous optimization of the indicator system.

[0147] S630. Perform compatibility testing on the new version of the indicator library and generate closed-loop verification results.

[0148] Using the new version of the indicator library output by the dynamic indicator system update module S620 as input, the system specifically performs compatibility testing on the new version of the indicator library. This testing process first uses a predefined set of indicator compatibility rules to automatically verify the consistency of indicator definitions, the standardization of data formats, and the rationality of weight distribution for each indicator in the indicator library. Specifically, the system performs rule matching on indicator name standardization, indicator classification accuracy, and logical relationships between indicators to identify potential naming conflicts, classification errors, and logical inconsistencies. Further, the system applies the new version of the indicator library to a historical dataset in a simulated test environment, performing simulated calculations and evaluations to detect the impact of the indicator library update on the evaluation results. The simulated test includes verification of indicator data integrity, analysis of weight application effects, and stability testing of evaluation results, using statistical analysis and anomaly detection algorithms to identify abnormal fluctuations and unreasonable results. For problems found during the test, the system automatically generates a compatibility anomaly report, recording the abnormal indicators, anomaly types, and suggested corrective measures in detail, while triggering a feedback mechanism to notify indicator maintenance personnel. The system supports iterative correction and retesting of the indicator library based on the feedback results, forming a closed-loop verification process. During compatibility testing, the system comprehensively records test data, results, and operational processes, storing them in a compatibility test log library for easy tracking and quality control. Upon completion of the compatibility test, the system generates a closed-loop verification result. This result includes a test pass identifier, an anomaly report summary, and correction suggestions, serving as the output fields for this step. It is then passed to the "weight update instruction" of the dynamic indicator standardization processing module S210 to guide weight adjustments and indicator updates. Simultaneously, it supports version management and release control of the dynamic indicator system update module. The application of this closed-loop verification result enables the dynamic improvement and continuous optimization of the indicator system, promoting the scientific rigor and practicality of clean and low-carbon evaluation methods for power systems.

[0149] Example 2: Figure 2 A structural block diagram of a RAM-based clean and low-carbon power system evaluation system according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include:

[0150] The parameter constraint module 01 is used to receive and parse installation parameters. Specifically, it receives installation parameters from the power system initialization configuration, parses key fields in the installation parameters through the configured parameter interface, including unit type, rated power, emission factor, etc., to form a parameter configuration table. The parameter configuration table is recorded as a parameter log file and maintains a consistent association with the system initialization state; the parameter log file is transferred to the data acquisition module for parameter input, while parameter logs are retained for subsequent traceability.

[0151] Data acquisition module 02 is used to acquire unit output data and carbon emission intensity data. Specifically, it receives parameter log files from the parameter limiting module and data streams from real-time power system monitoring equipment. Through the configured data acquisition interface, it acquires unit output data and carbon emission intensity data, performs data format conversion, protocol matching, and outlier filtering to form a structured dataset. The structured dataset is then passed to the preprocessing module as acquired data, and the corresponding time information is registered in the data buffer for subsequent modules to read.

[0152] Preprocessing module 03 is used to generate a normalized output sequence and a load-related emission factor. Specifically, it receives acquired data from the data acquisition module, performs data normalization processing and emission factor calculation in conjunction with the system load curve, and generates a normalized output sequence and a load-related emission factor. The normalized output sequence and load-related emission factor are then used as input data by the spatiotemporal alignment module, and their correspondence with the parameter configuration is recorded in the preprocessing record library.

[0153] The spatiotemporal alignment module 04 is used to construct a spatiotemporal coupled dataset. Specifically, based on the normalized output sequence and load-related emission factor from the preprocessing module, it completes temporal alignment and spatial mapping to generate a spatiotemporal coupled dataset. The spatiotemporal coupled dataset is output to the arbitration decision module as decision input data and the alignment status is returned to the preprocessing module for registration.

[0154] Arbitration decision module 05 is used to output unit scheduling instructions. Specifically, it receives the spatiotemporally coupled dataset from the spatiotemporal alignment module, performs scheduling optimization and instruction generation through a decision algorithm, and obtains the unit scheduling instructions. The unit scheduling instructions are then output to the control execution module, maintaining an index relationship consistent with the decision time.

[0155] The control execution module 06 is used to adjust the output curves of grid-connected generating units. Specifically, it receives unit dispatch instructions from the arbitration decision module, combines them with the real-time status of the power system, executes output curve adjustments, and generates adjusted output configurations. The adjusted output configurations are then returned to the status update module for status maintenance and to update the adjustment logs.

[0156] The status update module 07 is used to maintain the dynamic evaluation benchmark. Specifically, it receives the adjusted output configuration from the control execution module, performs an update operation on the system evaluation benchmark, and forms a new evaluation benchmark. The evaluation benchmark is then returned to the parameter limiting module for parameter backfeeding and to complete the update of the status index.

Claims

1. A RAM-based power system cleaning low-carbon evaluation method, characterized in that, Comprise: Real-time acquisition of energy input and output data from power system multi-source heterogeneous data acquisition devices, including energy input data, renewable energy generation and carbon management related data, based on statistical methods and rule engine, missing value filling, combination of multi-source data redundancy verification and time series inference technology and time stamp alignment, interpolation and resampling technology, to generate structured data sets; Based on the structured data set, the original index is extracted, the normalization processing is carried out, the interval scaling method is used to map to the standardized interval, the dynamic weight adjustment is based on the entropy weight method and the dynamic update of the weight coefficient based on the expert experience, and the dimension verification is carried out to check the integrity of the index and verify the consistency of the data, to generate a verified data set; Build RAM model input matrix to map input index and output index column vector, calculate efficiency value using linear programming method, optimize model adaptability, adjust constraint conditions and parameter configuration, and check confidence based on statistical methods and machine learning technology, output reliable efficiency value sequence; Decompose the slack variable of the reliable efficiency value based on the linear programming solution of the RAM model, identify input redundancy to generate a redundant index list and output deficiency, complete cause correlation analysis to extract output deficiency characteristics and use cross-correlation analysis and principal component contribution calculation and priority sorting based on AHP to calculate comprehensive priority, generate a high-value improvement point list; According to the high-value improvement point matching optimization strategy library, generate optimization scheme and perform feasibility evaluation and multi-objective optimization, form the final optimization decision report; Extract the index update signal in the decision report, perform index library version comparison, candidate index screening and weight redistribution, complete compatibility test and closed loop verification.

2. The RAM-based low carbon evaluation method of power system cleaning according to claim 1, characterized in that, The process of extracting original indexes based on structured data sets includes: Through the configured data extraction interface, according to the pre-defined index classification rules, the original index data meeting the requirements of the index system is filtered out from the structured data set; The index classification rules are based on index name, data type and timestamp information, supporting batch filtering combined with single verification processing mode.

3. The RAM-based power system cleaning low carbon evaluation method according to claim 1, characterized by, The process of executing normalization processing specifically includes: Using interval scaling method to map different dimension and scale index values to a unified standardized interval; According to the nature of the index, distinguish between expected output index and non-expected output index, and use maximum and minimum value normalization and reverse normalization strategy respectively; The realization of normalization processing is based on sliding window technology, combined with the statistical characteristics of real-time data stream to dynamically adjust the normalization parameters.

4. The RAM-based low carbon evaluation method of power system cleaning according to claim 1, characterized in that, The dynamic weight adjustment includes: Through the integration of multi-factor analysis model to realize the dynamic update of weight coefficient, the model combines the objective weight calculation of entropy weight method and the subjective weight correction of expert experience, supplemented by machine learning algorithm to optimize weight adjustment strategy; The update process of dynamic weight parameter has a trigger mechanism, which supports both timed update and trigger update based on data anomaly or system event.

5. The RAM-based power system cleaning low carbon evaluation method according to claim 1, characterized by, The dimension verification includes: Performing index integrity check, data consistency verification and abnormal dimension elimination; The dimension verification is based on pre-defined index system rules, automatically identifies missing or abnormal dimensions, and uses rule engine and data verification algorithm in combination; For the detected missing or abnormal dimensions, the system automatically generates data maintenance requests through a feedback mechanism.

6. The RAM-based low carbon evaluation method of power system cleaning according to claim 1, wherein, The process of decomposing the relaxation variable of the credible efficiency value based on the linear programming solution result of the RAM model includes: The multi-dimensional standardized index data is classified into corresponding input vectors and output vectors according to input and output classification; Through the data mapping rule engine, the rule engine automatically classifies and labels the index data according to the index name, type and weight information.

7. The RAM-based low carbon evaluation method of power system cleaning according to claim 1, wherein, The calculation of the efficiency value includes: Using the interval adjustment characteristics of the RAM model, the weight distribution of input and output is dynamically adjusted; The efficiency value calculation process is realized by an iterative optimization algorithm, which solves the linear programming problem by using the interior point method or the simplex method; The system monitors the calculation state in real time, and automatically marks and recalculates abnormal calculation results.

8. The RAM-based power system cleaning low carbon evaluation method according to claim 1, characterized by, The sequence of credible efficiency values includes: Map the relaxation variable vector to the specific dimensions of each input and output index, and normalize each dimension relaxation variable by combining the index weight and dynamic adjustment parameter; By setting threshold rules to filter significant relaxation variables, noise and small fluctuations are filtered out; Generate a list of redundant indicators, detailing the redundant indicator name, redundant size and corresponding time period.

9. The RAM-based power system cleaning low carbon evaluation method according to claim 1, wherein, The process of generating a list of high-value improvement points includes: Constructing relaxation variable normalization processing: wherein, is the normalized slack variable, is the original slack variable, μ i is the historical mean of the i a th input indicator of the a i th input indicator, i is the index of the input indicator, i a is the index of the a th input indicator. Further, the system uses an improved Shannon entropy model to filter significant redundant items and calculates the information contribution of redundant indicators: where H 1 is the improved Shannon entropy, which is used to measure the significance of redundancy; is the normalized redundancy probability distribution; ∈ is the smoothing factor; n is the total number of input indicators. generated by generated by H is further calculated 1 when H 1 <θ 1 is determined to be significant redundancy, wherein θ 1 is a significant redundancy determination threshold. Further, the time series data of the redundant items to be analyzed and the output indicator sequence are cross-correlated to calculate the time lag correlation coefficient of redundant input and output deficiency: wherein r 2 (τ) is a time-lag dependent coefficient; t is a time point index; is the redundancy at time t; is the redundancy at time t+τ; is the arithmetic mean over the time window T; is the j b class underproduction indicator; is the j b class underproduction indicator at time t+τ; τ is a time-lag parameter and T is the length of the time window; is the arithmetic mean over the time window T; Further, a principal component analysis model is constructed to extract key cause features, and the principal component contribution is calculated: wherein, contribution of the kth principal component; contribution of the kth principal component to the ith a load coefficient of the redundancy index; k is the principal component index; m is the number of redundancy index categories. Further, the analytic hierarchy process is used to calculate the comprehensive priority of each improvement point, and the weight matrix is defined: wherein W 4 is a weight matrix; represents the influence weight of the pth cause feature on the qth improvement point, m ′ is the total number of improvement points; Further, the priority allocation is optimized through a linear programming model to construct the objective function: wherein: a 5 and β 5 are weight coefficients, is the decision variable of the qth improvement point; is the contribution of the qth principal component.

10. A RAM-based low carbon evaluation system for power system cleaning, applied to the method of any one of claims 1-9, characterized in that, It includes: Parameter limiting module, used to receive and analyze installation parameters; Data acquisition module, used to obtain unit output data and carbon emission intensity data; Preprocessing module, used to generate normalized output sequence and load-related emission factor; Space-time alignment module, used to construct space-time coupled data set; Arbitration decision module, used to output unit scheduling instructions; Control execution module, used to adjust the grid-connected unit output curve; State update module, used to maintain dynamic evaluation benchmarks.

Images