Industrial adjustable load measurement method based on multi-dimensional load classification and scene self-adaptation

Abstract

The present application belongs to the technical field of power system demand side management, and particularly relates to an industrial adjustable load calculation method based on multi-dimensional load classification and scene self-adaptation, which comprises the following steps: establishing a multi-dimensional refined load classification system, dividing industrial load into four dimensions; constructing a multi-dimensional constraint parameter system, establishing a constraint evaluation matrix about constraints and equipment; obtaining the comprehensive weight of each type of constraint and calculating the constraint correction coefficient of a single device; establishing a progressive calculation model of single device, subdivided category, load category and total load, and outputting the total adjustable load; using the self-adaptive Monte Carlo simulation method to output the uncertainty calculation result; training the XGBoost regression model to predict the deviation correction value of the current calculation result, and online correcting the load adjustable proportion, the industry adjustment coefficient and the comprehensive weight of each type of constraint, and outputting the final calculation result after closed-loop optimization. The present application can realize accurate calculation and self-adaptive update of industrial adjustable load.

Description

Technical Field

[0001] This invention belongs to the field of power system demand-side management technology, specifically involving a method for calculating industrial adjustable load based on multi-dimensional load classification and scenario adaptation. Background Technology

[0002] With the continuous advancement of the construction of new power systems, the high proportion of renewable energy grid integration has led to a significant increase in the pressure on power grid peak shaving, voltage regulation, and power balance. Industrial adjustable load, as the largest and most flexible resource on the demand side with the strongest adjustment potential, has become a crucial support for ensuring the safe and stable operation of the power grid. Traditional methods for calculating industrial adjustable load often focus on single industries and single equipment, relying on fixed parameters and empirical coefficients for estimation. They lack a unified and standardized calculation framework, resulting in poor cross-industry applicability and scenario adaptability. Furthermore, existing methods often employ subjective assignment or a single entropy weight method when determining constraint weights, failing to consider both objective data patterns and actual dispatch requirements. This leads to significant weight distortion and substantial deviations in the calculation of adjustable potential.

[0003] Furthermore, traditional models generally employ static load classification and fixed parameter calculation methods, failing to consider dynamic changes in production safety, environmental compliance, and equipment characteristic constraints. They also lack differentiated adaptation for different scheduling scenarios such as minute-level, hourly-level, and day-ahead-level scheduling, resulting in a significant disconnect between the calculated results and the actual on-site adjustable capacity. Most methods output only a single, deterministic value, lacking uncertainty analysis and risk quantification, thus failing to provide reliable boundaries for scheduling decisions. Simultaneously, the models lack dynamic correction capabilities, cannot utilize historical response data for iterative optimization, and their long-term operational accuracy continues to decline. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this invention is to provide an industrial adjustable load calculation method based on multi-dimensional load classification and scenario adaptation. By constructing a unified calculation framework, a multi-dimensional constraint correction system, a hierarchical recursive calculation structure and a closed-loop dynamic optimization mechanism, the method can achieve accurate calculation and adaptive updating of industrial adjustable load.

[0005] To achieve the above objectives, this invention provides a method for calculating industrial adjustable loads based on multidimensional load classification and scenario adaptation, comprising the following steps: S1. For industrial loads, acquire raw data including static basic data and dynamic operating data. After data cleaning, establish a multi-dimensional refined load classification system, dividing industrial loads into four dimensions: major load categories, sub-categories, single equipment levels, and time scales. S2. Construct a multi-dimensional constraint parameter system that includes production safety constraints, environmental compliance constraints, and equipment characteristic constraints; formulate standardized value rules for the correction coefficients of various constraint items; collect single-equipment constraint parameters for the target measurement object; and establish a constraint evaluation matrix for constraints and equipment. S3. Normalize the constraint evaluation matrix and combine it with the entropy weight method and the analytic hierarchy process to obtain the comprehensive weight of various constraints, and calculate the constraint correction coefficient of a single device. S4. Establish a progressive calculation model for single equipment, sub-categories, load categories, and total load. Using a single equipment as the smallest unit of load calculation, the model sequentially summarizes the sub-categories, load categories, and total load, outputs the total adjustable load, and calculates the load adjustable ratio and industry adjustment coefficient. S5. Identify key random parameters that significantly affect the measurement results, set the probability distribution corresponding to the key random parameters, use the adaptive Monte Carlo simulation method with convergence judgment mechanism to generate random parameter samples based on the set probability distribution, substitute them into the progressive measurement model to complete the batch calculation of the load adjustable ratio, perform convergence judgment, perform statistical analysis on all sample results after convergence, and output the uncertainty measurement results. S6. Train the XGBoost regression model to predict the deviation correction value of the current calculation result, and make online corrections to the comprehensive weights of the load adjustable ratio, industry adjustment coefficient and various constraints, and output the final calculation result after closed-loop optimization.

[0006] As a preferred embodiment of the present invention, in S1, the static basic data includes the rated parameters of industrial enterprise equipment, production process attributes, load category of equipment, and grid connection point information; the dynamic operation data includes the real-time operating power of equipment, production condition data, grid dispatch instruction data, meteorological data, environmental compliance constraint data, and equipment operation and maintenance monitoring data. After obtaining the raw data, data cleaning is performed to remove invalid duplicate data and abnormal data that does not conform to physical boundaries.

[0007] As a preferred embodiment of the present invention, in S1, the multidimensional refined load classification system divides industrial load into four dimensions, specifically: The first dimension is the load category, which includes three load categories: core production load, auxiliary production load, and non-production load. The second dimension is the subcategories, which break down the three major load categories into core and optional categories, including six subcategories: core process load, non-core process load, necessary auxiliary load, optional auxiliary load, necessary non-production load, and optional non-production load. The third dimension is the single device level, which establishes a standardized parameter input interface for each device, covering rated power, device type, process stage, load category, and historical operating data. The fourth dimension is the time scale, which is divided into three time scales according to the core demand response scenarios of the power grid: minute-level emergency response, hour-level peak response, and day-level day-ahead response. For each sub-category, the upper limit of the allowable adjustment ratio and the constraint boundary under the corresponding time scale are matched.

[0008] As a preferred embodiment of the present invention, in S2, the standardized value rules for the various constraint sub-item correction coefficients are as follows: for production safety constraints, the value range for core non-shutdown equipment is [0.1, 0.3], and the value range for non-core shut-down equipment is [0.8, 1.0]; for environmental compliance constraints, the value range for environmental protection equipment that needs to operate synchronously with the main equipment is [0.2, 0.4], and the value range for other equipment is [0.9, 1.0]; for equipment characteristic constraints, the value range for equipment whose response time and adjustment range meet the scheduling requirements is [0.9, 1.0], and the value range for equipment that does not meet the scheduling requirements is [0.3, 0.5]. Constraint parameters of a single device are collected for the target measurement object, and a constraint evaluation matrix of m types of constraints and n devices is established. ,in Let be the partial correction factor for the i-th device under the j-th type of constraint.

[0009] As a preferred embodiment of the present invention, the process of obtaining the comprehensive weights of various constraints and calculating the constraint correction coefficients in S3 is as follows: S3.1 Perform Min-Max normalization on the constraint evaluation matrix to eliminate the dimensional differences between different constraint indicators and obtain a standardized matrix. , This represents the standardized component correction factor for the i-th device under the j-th type of constraint; S3.2 Calculate the information entropy of the j-th type of constraint. Based on the negative correlation between information entropy and weight, the objective weight of the j-th type of constraint is calculated. ; S3.3. Based on the analytic hierarchy process, obtain the subjective weights of the j-th type of constraint. ; S3.4. Based on the principle of minimum discriminative information, the comprehensive weights of various constraints are obtained: ; In the formula, This represents the overall weight of the j-th type of constraint; S3.5 Calculate the constraint correction factor for a single device: ; In the formula, represents the constraint correction coefficient of the i-th device.

[0010] As a preferred embodiment of the present invention, in step S4, the process of outputting the total adjustable load and calculating the load adjustable ratio and industry adjustment coefficient is as follows: S4.1 Calculate the actual adjustable load of a single device: ; In the formula, This represents the actual adjustable load of the i-th device; This represents the design rated power of the i-th device; This represents the allowable adjustment ratio of the i-th device in the target time scale scenario; S4.2 Calculate the total adjustable load of each subclass: ; In the formula, This represents the total adjustable load of the subclass "sub"; The number of devices in each sub-category; S4.3 Calculate the total adjustable load of the load category: ; In the formula, This represents the total adjustable load of load category c. ; This represents the number of subcategories under the c-th major load category; S4.4, Total adjustable output load: ; In the formula, This represents the total adjustable load of an industry or enterprise. S4.5, Calculate the adjustable load ratio : ; In the formula, Indicates the total load of an industry or enterprise; S4.6 Calculate the industry adjustment coefficient K: ; In the formula, This indicates the proportion of the total adjustable load of load category c to the total industry load; This represents the power-weighted average constraint correction factor for the c-th load category; This indicates the number of devices in the c-th load category; This indicates the total number of devices in an industry or enterprise.

[0011] As a preferred embodiment of the present invention, in S5, the key random parameters include the single equipment allowable adjustment ratio, the production safety constraint sub-item correction coefficient, the environmental compliance constraint sub-item correction coefficient, and the equipment characteristic constraint sub-item correction coefficient. The probability distribution corresponding to the key random parameters is set based on industry statistical data and production operation characteristics. In the adaptive Monte Carlo simulation method with convergence determination mechanism, the initial number of simulations is set. Maximum number of simulations Convergence threshold Iteration step size Random parameter samples are generated based on a set probability distribution and substituted into a progressive calculation model to complete batch calculations of the load adjustable ratio. The next sampling is performed to calculate the mean and coefficient of variation of the current adjustable load proportions for all samples. Perform a convergence check; if the absolute value of the difference between the coefficients of variation of two adjacent iterations is less than the convergence check threshold... Or the number of simulations reaches the maximum number of simulations. If the result is not found, the iteration stops; otherwise, the number of simulations is increased, and the sampling and calculation process is repeated. Perform statistical analysis on all converged sample results and output the uncertainty calculation results, including the mean, standard deviation and 95% confidence interval of the load adjustable ratio.

[0012] As a preferred embodiment of the present invention, in S6, the actual adjustable load, actual adjustment ratio, equipment operating parameters, constraint data, and scenario parameter data in the historical demand response events of the target industry or enterprise are collected to construct a historical dataset including feature set and label set; The historical dataset is divided into training and testing sets in a 7:3 ratio to train the XGBoost regression model. The input parameters calculated by the progressive calculation model are substituted into the trained XGBoost regression model to predict the deviation correction value of the current calculation result. Based on the deviation correction value, the comprehensive weights of the load adjustable ratio, industry adjustment coefficient and various constraints are corrected online. The calculation results of the progressive calculation model are updated based on the correction results, and the final calculation result after closed-loop optimization is output.

[0013] The beneficial effects of this invention are: This invention constructs a four-dimensional refined load classification system to match multi-time period scheduling scenarios. It adopts the entropy weight hierarchical analysis method to combine weighting and quantify the impact of multi-dimensional constraints. Relying on the hierarchical recursive calculation logic, it summarizes the adjustable capacity of loads with different characteristics layer by layer. This not only achieves unified and standardized calculation across industries, but also significantly reduces the calculation deviation caused by a single estimation mode. The calculation results are consistent with the actual production, environmental protection, and equipment operation constraints of enterprises. The engineering versatility and calculation accuracy are significantly better than the existing static calculation schemes.

[0014] This invention employs adaptive Monte Carlo simulation to quantify the uncertainty of adjustable potential, and can output the confidence interval of the calculation results to provide risk boundaries for power grid dispatch. At the same time, it introduces the XGBoost algorithm to build a data feedback closed-loop optimization mechanism, and continuously corrects the constraint weights and industry adjustment coefficients based on historical response data, solving the problem of accuracy decay of traditional models over long-term use. The whole method has deterministic calculation, risk assessment and dynamic self-updating capabilities, and can provide long-term stable and multi-dimensional data support for demand response and virtual power plant aggregation dispatch. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the principle of this invention; Figure 2 This is a technical architecture diagram of the calculation method of the present invention. Detailed Implementation

[0016] The embodiments of the present invention will be further described below with reference to the accompanying drawings: Example 1: As Figure 1 As shown, the industrial adjustable load calculation method based on multidimensional load classification and scenario adaptation includes the following steps: S1. For industrial loads, acquire raw data including static basic data and dynamic operating data. After data cleaning, establish a multi-dimensional refined load classification system, dividing industrial loads into four dimensions: major load categories, sub-categories, single equipment levels, and time scales. S2. Construct a multi-dimensional constraint parameter system that includes production safety constraints, environmental compliance constraints, and equipment characteristic constraints; formulate standardized value rules for the correction coefficients of various constraint items; collect single-equipment constraint parameters for the target measurement object; and establish a constraint evaluation matrix for constraints and equipment. S3. Normalize the constraint evaluation matrix and combine it with the entropy weight method and the analytic hierarchy process to obtain the comprehensive weight of various constraints, and calculate the constraint correction coefficient of a single device. S4. Establish a progressive calculation model for single equipment, sub-categories, load categories, and total load. Using a single equipment as the smallest unit of load calculation, the model sequentially summarizes the sub-categories, load categories, and total load, outputs the total adjustable load, and calculates the load adjustable ratio and industry adjustment coefficient. S5. Identify key random parameters that significantly affect the measurement results, set the probability distribution corresponding to the key random parameters, use the adaptive Monte Carlo simulation method with convergence judgment mechanism to generate random parameter samples based on the set probability distribution, substitute them into the progressive measurement model to complete the batch calculation of the load adjustable ratio, perform convergence judgment, perform statistical analysis on all sample results after convergence, and output the uncertainty measurement results. S6. Train the XGBoost regression model to predict the deviation correction value of the current calculation result, and make online corrections to the comprehensive weights of the load adjustable ratio, industry adjustment coefficient and various constraints, and output the final calculation result after closed-loop optimization.

[0017] In S1, static basic data includes resource attribute data such as rated parameters of industrial enterprise equipment, production process attributes, load category of equipment, and grid connection point information; dynamic operation data includes operating environment and constraint data such as real-time operating power of equipment, production condition data, grid dispatch instruction data, meteorological data, environmental compliance constraint data, and equipment operation and maintenance monitoring data. After obtaining the raw data, data cleaning is performed to remove invalid duplicate data and abnormal data that does not conform to physical boundaries.

[0018] Static basic data can be obtained through Enterprise Resource Planning (ERP) systems, Manufacturing Execution Systems (MES), equipment ledger management systems, and power grid GIS platforms, covering rated parameters of industrial enterprise equipment, production process attributes, equipment load categories, and power grid connection point information. Dynamic operating data can be collected by smart meters and power sensors deployed on the equipment to collect real-time operating power, extract production condition data from SCADA systems, receive power grid dispatch instructions via the dispatch data network, obtain meteorological data through meteorological department data interfaces or local meteorological monitoring stations, extract environmental compliance constraint data based on the environmental protection department's online monitoring platform and the enterprise's pollutant discharge permit implementation report, and obtain equipment operation and maintenance monitoring data using equipment condition monitoring and fault diagnosis systems.

[0019] The original data consisted of multi-source heterogeneous data, which was uniformly accessed and cleaned before entering the subsequent modeling process. The data cleaning stage first standardized the access and format of the multi-source heterogeneous data. Invalid and duplicate data were removed using a combination of hash-based deduplication algorithms and sliding window timestamp comparison to ensure that there were no redundant records for the same device in adjacent sampling periods. For anomalous data that did not conform to physical boundaries, a boundary judgment model was established based on physical rules such as the upper and lower limits of the device's rated power, a reasonable power factor range (0.3~1.0), the logical consistency of the device's start-up and shutdown states, the load change rate threshold (e.g., a change of no more than 30% of the rated power per minute), and the coupling constraints of ambient temperature and the device's heat dissipation capacity. Outliers exceeding the boundaries were handled using the Laida criterion. The data is identified using criteria or box plot quartile method, and after manual verification and confirmation in conjunction with equipment operation logs, it is either eliminated or interpolated to form a high-quality, non-redundant, standardized dataset that conforms to the physical laws of the industrial site.

[0020] In the multidimensional refined load classification system, industrial load is divided into four dimensions, specifically: The first dimension is the load category, which includes three load categories: core production load, auxiliary production load, and non-production load, covering all types of electricity loads of industrial enterprises; The second dimension is the sub-categories. Based on the equipment regulation characteristics and the importance of the production process, the three major load categories are divided into core categories and optional categories, including six sub-categories: core process load, non-core process load, necessary auxiliary load, optional auxiliary load, necessary non-production load, and optional non-production load. The regulation characteristic boundaries of each sub-category are clearly defined. Core process loads are those of core production equipment that cannot be shut down but can only be slightly reduced, such as cement rotary kilns, ferroalloy submerged arc furnaces, and chemical gasification furnaces; non-core process loads are those of production equipment that can be shut down temporarily and adjusted significantly, such as raw material mills and intermediate reactors; essential auxiliary loads are auxiliary equipment that operates synchronously with the main equipment, such as environmental dust collectors and core equipment cooling water pumps; optional auxiliary loads are auxiliary equipment that can be shut down independently, such as conveyor belts in non-core areas and auxiliary ventilation equipment; essential non-production loads are non-adjustable safety loads, such as fire-fighting equipment and emergency lighting; optional non-production loads are non-production loads that can be fully adjusted, including office central air conditioning and general lighting.

[0021] Table 1 shows an example of a refined load classification system consisting of the first and second dimensions.

[0022] Table 1 Refined Load Classification System

[0023] The third dimension is the single-device level, the smallest calculation unit, which establishes a standardized parameter input interface for each device, covering core parameters such as rated power, equipment type, process stage, load category, and historical operating data. The fourth dimension is the time scale, which is divided into three time scales according to the core demand response scenarios of the power grid: minute-level emergency response, hour-level peak response, and day-level day-ahead response. For each sub-category, the upper limit of the allowable adjustment ratio and the constraint boundary under the corresponding time scale are matched. Among them, the minute-level emergency response only opens the adjustment authority of optional auxiliary loads and optional non-production loads, the hour-level peak response simultaneously opens the adjustment authority of non-core process loads, and the day-ahead response opens the compliant adjustment authority of all types of loads, so as to achieve accurate adaptation of the progressive calculation model to the actual dispatching scenario of the power grid.

[0024] The core demand response scenario of the power grid refers to the typical control mode in which the power system sends price signals or dispatch instructions to the user side to cope with fluctuations in supply and demand balance, and guides industrial, commercial and other users to actively adjust their electricity consumption behavior in order to achieve peak shaving and valley filling and ensure the safe and stable operation of the power grid. It is divided into minute-level emergency response, hour-level peak response and day-ahead response.

[0025] By constructing the above-mentioned four-dimensional refined load classification system, standardized adaptation can be achieved for all categories of industrial industries such as cement, ferroalloys, chemicals, and machinery manufacturing. Load classification and calculation preparation for different industries can be completed without reconstructing the core structure of the model, thus solving the core problem of poor cross-industry adaptability of existing models.

[0026] After the underlying classification architecture is built, a multi-dimensional constraint parameter system and constraint correction core module are further constructed to accurately depict the actual impact of constraints on load regulation capacity throughout the entire industrial production process. This is the core guarantee for the accuracy of model calculations.

[0027] In S2, the standardized value rules for the correction coefficients of various constraints are as follows: For production safety constraints, the correction coefficient for core non-shutdown equipment is [0.1, 0.3], and the correction coefficient for non-core shut-down equipment is [0.8, 1.0]; for environmental compliance constraints, the correction coefficient for environmental protection equipment that needs to operate synchronously with the main equipment is [0.2, 0.4], and the correction coefficient for other equipment is [0.9, 1.0]; for equipment characteristic constraints, the correction coefficient for equipment whose response time and adjustment range meet the scheduling requirements is [0.9, 1.0], and the correction coefficient for equipment that does not meet the scheduling requirements is [0.3, 0.5]. The value rules are shown in Table 2. Table 2 Rules for Determining Correction Factors

[0028] Constraint parameters of a single device are collected for the target measurement object, and a constraint evaluation matrix of m types of constraints and n devices is established. ,in Let m be the partial correction coefficient for the i-th device under the j-th type of constraint, thus completing the construction of the model constraint input system. m=3 corresponds to the three types of constraints.

[0029] In S3, the process of obtaining the comprehensive weights of various constraints and calculating the constraint correction coefficients is as follows: S3.1 Perform Min-Max normalization on the constraint evaluation matrix to eliminate the dimensional differences between different constraint indices: ; In the formula, This represents the standardized component correction factor for the i-th device under the j-th type of constraint; The correction coefficient for the j-th type of constraint. The minimum value, for The maximum value, if ,but This indicates that this type of constraint has no effect on the adjustment capability of all devices.

[0030] Obtain the standardized matrix ; S3.2 Construct an objective weight calculation unit based on the entropy weight method to calculate the information entropy of the j-th type of constraint. : ; In the formula, the proportion coefficient ,like ,but The smaller the information entropy, the greater the dispersion of the partial correction coefficients for this type of constraint, and the more significant its impact on load regulation capability. Based on the negative correlation between information entropy and weight, the objective weight of the j-th type of constraint is calculated. : ; The formula satisfies Normative requirements.

[0031] S3.3. Simultaneously build a subjective weight calculation unit based on the analytic hierarchy process (AHP). Establish a constraint hierarchy structure for the target application scenario. Using the impact of constraints on load regulation capacity as a benchmark, construct pairwise comparison judgment matrices. Then, perform a consistency check on the judgment matrices and calculate the consistency ratio CR. When the judgment matrix passes the consistency test, the subjective weights of the j-th type of constraint are obtained based on the judgment matrix that passed the test and the analytic hierarchy process (AHP). ,satisfy ; S3.4 After obtaining the objective and subjective weights, a coupled calculation unit for the objective and subjective weights is constructed based on the principle of minimum discriminative information to obtain the comprehensive weights of various constraints: ; In the formula, Let represent the comprehensive weight of the j-th type of constraint; this formula retains the data-driven characteristics of objective weights while incorporating the scenario-adaptation requirements of subjective weights, achieving optimal coupling between objective and subjective weights, and satisfying . Normative requirements.

[0032] S3.5 Construct a single-equipment constraint correction coefficient calculation unit to comprehensively reflect the combined impact of multi-dimensional constraints on the adjustment capability of a single piece of equipment. The calculation method for the constraint correction coefficient of a single piece of equipment is as follows: ; In the formula, denoted as the constraint correction coefficient for the i-th device, with a value range of [0, 1]. The closer the value is to 1, the more relaxed the constraint on the device and the higher the degree of realization of the adjustment capability. The closer the value is to 0, the more stringent the constraint on the device and the more limited the actual adjustment capability.

[0033] By constructing a multi-dimensional constraint parameter system and a core module for constraint correction, the core defects of traditional weighting methods, such as large human bias and poor scenario adaptability, are solved, providing a reliable basis for constraint correction for the accurate calculation of subsequent adjustable loads.

[0034] The next step is to build the core calculation model, constructing a four-layer progressive calculation logic of single equipment, sub-categories, load categories, and total load. This enables accurate hierarchical aggregation of adjustable loads from single equipment to the entire industry, avoiding cross-interference between equipment with different adjustment characteristics and improving calculation accuracy.

[0035] In S4, the process of outputting the total adjustable load and calculating the load adjustable ratio and industry adjustment coefficient is as follows: S4.1 Construct a single-device adjustable load calculation unit as the basic calculation node for the entire calculation model. Its value is the product of the device's rated power, the scenario adjustment ratio, and the constraint correction coefficient. The calculation method for the actual adjustable load of a single device is as follows: ; In the formula, This represents the actual adjustable load of the i-th device, in kW, consistent with the unit of rated power. This represents the design rated power of the i-th device; This represents the allowable adjustment ratio of the i-th device in the target time scale scenario, with a value range of [0, 1]; when or When this occurs, it indicates that the equipment has no adjustable load.

[0036] S4.2. Based on the calculation results of a single device, a sub-category adjustable load aggregation unit is constructed to aggregate the adjustable loads of devices with similar adjustment characteristics within the same sub-category. The calculation method for the total adjustable load of each sub-category is as follows: ; In the formula, This represents the total adjustable load of the subclass "sub"; This unit represents the number of devices in each sub-category; by summarizing these sub-categories, the total adjustable capacity of each type of load characteristic can be precisely quantified.

[0037] S4.3 Calculate the total adjustable load of the load category: ; In the formula, This represents the total adjustable load of load category c. These correspond to core production load, auxiliary production load, and non-production load, respectively. The adjustable capacity of each of these three types of loads can be obtained through this unit. This represents the number of subcategories under the c-th major load category; S4.4 Finally, construct the total adjustable load and adjustable ratio calculation unit, and output the core calculation results. The total adjustable load is the sum of the adjustable loads of the three main load categories. The calculation method for the total adjustable load is as follows: ; In the formula, This represents the total adjustable load of an industry or enterprise; S4.5, Calculate the adjustable load ratio : ; In the formula, This represents the total load of an industry or enterprise, which is the sum of the rated power of all equipment.

[0038] S4.6. Simultaneously establish an industry-specific adjustment coefficient K-value calculation unit to quantify the comprehensive potential of load adjustment in different industries and achieve horizontal comparison of cross-industry adjustment capabilities. The calculation method for the industry adjustment coefficient K is as follows: ; In the formula, This indicates the proportion of the total adjustable load of load category c to the total load of the industry; This represents the power-weighted average constraint correction factor for the c-th load category; This indicates the number of devices in the c-th load category; This represents the total number of devices in an industry or enterprise. The value of K ranges from 0 to 1. A higher value indicates a greater load adjustment potential for the industry. Through the construction of this core calculation model, standardized and accurate calculation of the adjustable load of the entire industry has been achieved.

[0039] This can be obtained by weighting the individual equipment constraint correction coefficients of all equipment under load category c according to their respective rated power. That is, the constraint correction coefficient of equipment with higher rated power contributes more to the category average, while the contribution of equipment with lower rated power is relatively smaller, thus ensuring... It can accurately reflect the degree of overall constraint leniency of this load category in actual operation.

[0040] This progressive calculation logic is bidirectionally adaptable. It can be applied to a single enterprise to assess its potential for participating in demand response adjustments, and it can be extended to the entire industry to achieve aggregated management and horizontal comparison of cross-enterprise load resource pools.

[0041] An adaptive Monte Carlo uncertainty analysis module was built: In S5, key random parameters include the allowable adjustment ratio of a single device, the correction coefficient for production safety constraints, the correction coefficient for environmental compliance constraints, and the correction coefficient for equipment characteristic constraints. The probability distributions corresponding to the key random parameters were set based on industry statistical data and production operation characteristics. For example, regarding the adjustment ratio of a single piece of equipment, based on the statistical data of the actual adjustment ratio of similar equipment in historical demand response events in the target industry (such as the average load reduction ratio of rotary kilns in the cement industry being about 15%, with a standard deviation of about 3%), and combined with production and operation characteristics such as equipment type, technological maturity, and years of operation, it is set as a normal distribution. , where the mean Take the design adjustment ratio of the device in the target scenario, and its standard deviation. The adjustment coefficients are set based on the industry's statistical fluctuation range (generally 10% to 20% of the average). For the three categories of constraints—production safety, environmental compliance, and equipment characteristics—the adjustment coefficients are set uniformly based on the proportion of constraint attributes for each type of equipment in the industry's equipment ledger (e.g., the proportion of core, non-shutdownable equipment, the proportion of environmental protection equipment requiring simultaneous operation), historical constraint execution records of similar equipment, and production and operation characteristics such as equipment aging, maintenance cycles, and the operational rate of environmental protection facilities. Where the lower limit 'a' and the upper limit 'b' correspond to the most lenient and most stringent boundary values ​​of this type of constraint, respectively (e.g., the core equipment value for production safety constraints). Non-core equipment This ensures that the probability distribution reflects both the overall statistical patterns of the industry and the actual operating conditions of specific equipment.

[0042] An adaptive Monte Carlo simulation unit with a convergence determination mechanism is constructed. In the adaptive Monte Carlo simulation method with a convergence determination mechanism, the initial number of simulations is set. Maximum number of simulations Convergence threshold Iteration step size Random parameter samples are generated based on a set probability distribution and substituted into a progressive calculation model to complete batch calculations of the load adjustable ratio. The next sampling is performed to calculate the mean and coefficient of variation of the current adjustable load proportions for all samples. Perform a convergence check; if the absolute value of the difference between the coefficients of variation of two adjacent iterations is less than the convergence check threshold... Or the number of simulations reaches the maximum number of simulations. If the result is not found, the iteration stops; otherwise, the number of simulations is increased, and the sampling and calculation process is repeated. A statistical analysis unit was built to perform statistical analysis on all converged sample results, outputting uncertainty calculation results, including the mean, standard deviation, and 95% confidence interval of the load adjustment ratio. The mean reflects the expected level of the load adjustment ratio, the standard deviation reflects the dispersion of the calculation results (a larger standard deviation indicates higher uncertainty), and the 95% confidence interval is calculated using the 2.5% and 97.5% quantiles after sample sorting, reflecting the possible range of values ​​for the load adjustment ratio at a 95% confidence level.

[0043] By constructing an adaptive Monte Carlo uncertainty analysis module, we can accurately quantify the risks and provide a clear risk boundary and safety margin reference for power grid dispatching decisions.

[0044] In S6, a historical dataset construction unit is built to collect actual adjustable load, actual adjustment ratio, equipment operating parameters, constraint data, and scenario parameter data from historical demand response events of the target industry or enterprise. This results in the construction of a historical dataset that includes a feature set and a label set. The feature set includes three categories: equipment rated parameters, constraint parameters, and scenario feature parameters. The label set consists of the deviation between the actual adjustable load ratio and the static calculation value of the progressive calculation model. A bias prediction model training unit was constructed, and the historical dataset was divided into training and test sets in a 7:3 ratio. The XGBoost regression model was trained with the goal of minimizing the bias value. Hyperparameter optimization was performed using a grid search method, and the coefficient of determination R was used. 2 The mean absolute error (MAE) is used as an evaluation metric for model fit, when the test set... When the fit is deemed satisfactory, it is determined that the fit is satisfactory. The input parameters calculated by the progressive calculation model are substituted into the trained XGBoost regression model to predict the deviation correction value of the current calculation result. Based on the deviation correction value, the comprehensive weights of the load adjustable ratio, industry adjustment coefficient and various constraints are corrected online. The calculation result of the progressive calculation model is updated based on the correction result, and the final calculation result after closed-loop optimization is output.

[0045] An exemplary online correction process is as follows: First, the bias correction value output by the XGBoost regression model is... Defined as the difference between the calculated load adjustable ratio output by the current progressive calculation model and the actual adjustable ratio, this is used as the total target deviation for this correction; a deviation allocation coefficient is pre-set. These correspond to the deviation bearing ratios of three correction paths: direct calibration of the load adjustable ratio, adjustment of the constraint comprehensive weight, and adjustment of the industry adjustment coefficient, respectively, to meet the requirements. .

[0046] Then, online corrections for the three types of objects are performed simultaneously: Perform a result-level calibration on the load adjustable ratio; the corrected load adjustable ratio is equal to the original measured value and... The sum, after calibration, is limited to a reasonable physical range of [0, 1]. The parameters of the comprehensive weights of various constraints are adjusted, and the... The differential coefficients are decomposed into three types of constraints based on the historical deviation attribution statistics. ,satisfy The weight adjustment shares corresponding to production safety constraints, environmental compliance constraints, and equipment characteristic constraints are respectively used to calculate the differentiated weight scaling factor for the j-th type of constraint. / Calculated value of the original load adjustable ratio For the differential coefficients of the j-th type constraint, according to The original values ​​of the comprehensive weights of the corresponding constraints are scaled and adjusted respectively. After the adjustment is completed, all duplicates are normalized to ensure that the sum of the weights is 1. At the same time, the values ​​of the weights of a single type of constraint are limited to a preset reasonable range. The industry adjustment coefficient is adjusted by parameters to... Calculate the coefficient scaling factor to handle the deviation at the coefficient end. The original value of the adjustable load ratio is used to correct the original value of the industry adjustment coefficient in the same direction. The corrected value is limited to the preset reasonable range of the industry.

[0047] The corrected constraint weights and industry adjustment coefficients are substituted back into the progressive calculation model to verify the deviation between the load adjustable ratio recalculated by the model and the load adjustable ratio after direct calibration. If the deviation is less than the preset accuracy threshold, the corrected three types of parameters and the calculation results are updated to online effective values ​​simultaneously, completing one closed-loop online correction.

[0048] like Figure 2 As shown, S1 serves as the data input and classification initialization layer. By collecting static basic data and dynamic operational data, and after data cleaning, a multi-dimensional refined load classification system is constructed, outputting a standardized classification system including major load categories, subcategories, single-equipment levels, and time scales. S2 and S3 serve as the constraint construction and weighting layer, sequentially completing the determination of three types of core constraints, construction of constraint evaluation matrices, matrix standardization, combined weighting using entropy weighting and analytic hierarchy process (AHP), and calculation of single-equipment constraint correction coefficients, outputting a constraint correction coefficient dataset. S4 and S5 serve as the core measurement and uncertainty analysis layer. S4, through a hierarchical progressive measurement model, completes the calculation from single-equipment... The system aggregates the total load step by step, outputting the total adjustable load, the adjustable load ratio, and the industry adjustment coefficient K. Based on this, S5 identifies key random parameters and sets their probability distributions. It then uses adaptive Monte Carlo simulation to quantify uncertainty, outputting the risk boundary quantification results, which include the mean, standard deviation, and 95% confidence interval. S6 serves as the dynamic optimization and result output layer. Based on historical datasets, an XGBoost deviation prediction model is trained to perform online corrections on the adjustable load ratio, industry adjustment coefficient, and the comprehensive weights of various constraints. Finally, it outputs the closed-loop optimized adjustable load value, adjustable load ratio, industry adjustment coefficient K, and 95% confidence interval.

[0049] The method in this embodiment achieves accurate, stable, and scenario-based quantitative calculation of industrial adjustable loads through multi-dimensional refined classification, subjective and objective combined weighting, hierarchical recursive calculation, adaptive Monte Carlo simulation, and XGBoost dynamic correction, providing reliable technical support for demand response, virtual power plant scheduling, and load resource pool management.

[0050] Example 2: The difference between this example and Example 1 is that the convergence determination of the adaptive Monte Carlo simulation adopts a dual-threshold coupling mechanism, specifically: The first convergence threshold is set as the relative rate of change of the average adjustable load ratio between two consecutive iterations. The second convergence threshold is the relative rate of change of the standard deviation of the adjustable load ratio between two adjacent iterations. After each preset iteration step and sampling, the mean of the adjustable load proportion of all samples is calculated simultaneously. Standard deviation And the relative rate of change of the mean of the previous iteration results. and the relative rate of change of standard deviation ; Only when the relative rate of change of the mean is less than And the relative rate of change of the standard deviation is simultaneously less than When the simulation converges, the iteration stops; if any relative rate of change does not reach the threshold, the number of simulations continues to increase until the maximum number of simulations is reached. For the converged sample results, in addition to outputting the mean, standard deviation and 95% confidence interval of the load adjustable ratio, the skewness coefficient and kurtosis coefficient of the load adjustable ratio are calculated simultaneously. When the absolute value of the skewness coefficient is greater than the preset threshold, the probability density function is reconstructed by replacing the normal distribution assumption with kernel density estimation, and then the 95% confidence interval is recalculated, and the uncertainty measurement result after distribution shape correction is output.

[0051] The dual-threshold coupling mechanism in this embodiment overcomes the shortcomings of single-threshold convergence judgment, which only focuses on mean stability and ignores the evolution of distribution pattern, and is prone to premature convergence or false convergence. The introduction of skewness-kurtosis test and kernel density estimation makes the uncertainty quantification results more in line with the actual skewed distribution characteristics, and significantly improves the accuracy of risk boundary measurement in extreme scenarios.

[0052] Example 3: An industrial adjustable load measurement device based on multidimensional load classification and scenario adaptation, comprising: One or more processors; Memory, used to store one or more computer programs; When one or more programs are executed by one or more processors, the one or more processors perform the method in Embodiment 1 or Embodiment 2.

[0053] Example 4: A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method in Example 1 or Example 2.

[0054] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can make equivalent substitutions or modifications based on the technical solution and concept of the present invention within the scope of the technology disclosed in the present invention, and such modifications should also be considered to fall within the scope of protection of the present invention.

Claims

1. A method for calculating industrial adjustable load based on multidimensional load classification and scenario adaptation, characterized in that, Includes the following steps: S1. For industrial loads, acquire raw data including static basic data and dynamic operating data. After data cleaning, establish a multi-dimensional refined load classification system, dividing industrial loads into four dimensions: major load categories, sub-categories, single equipment levels, and time scales. S2. Construct a multi-dimensional constraint parameter system that includes production safety constraints, environmental compliance constraints, and equipment characteristic constraints; formulate standardized value rules for the correction coefficients of various constraint items; collect single-equipment constraint parameters for the target measurement object; and establish a constraint evaluation matrix for m types of constraints and n pieces of equipment. ,in For the i-th device, under the j-th type of constraint, there is a partial correction factor; S3. Normalize the constraint evaluation matrix, and perform combined weighting based on the entropy weight method and the analytic hierarchy process to obtain the comprehensive weight of various constraints. Then, calculate the constraint correction coefficient for each device. The process is as follows: S3.1 Perform Min-Max normalization on the constraint evaluation matrix to eliminate the dimensional differences between different constraint indicators and obtain a standardized matrix. , This represents the standardized component correction factor for the i-th device under the j-th type of constraint; S3.2 Calculate the information entropy of the j-th type of constraint. Based on the negative correlation between information entropy and weight, the objective weight of the j-th type of constraint is calculated. ; S3.

3. Based on the analytic hierarchy process, obtain the subjective weights of the j-th type of constraint. ; S3.

4. Based on the principle of minimum discriminative information, the comprehensive weights of various constraints are obtained: ； In the formula, This represents the overall weight of the j-th type of constraint; S3.5 Calculate the constraint correction factor for a single device: ； In the formula, represents the constraint correction coefficient for the i-th device; S4. Establish a progressive calculation model for single equipment, sub-categories, load categories, and total load. Using a single equipment as the smallest unit of load calculation, sequentially summarize the sub-categories, load categories, and total load, outputting the total adjustable load. Calculate the load adjustability ratio and the industry adjustment coefficient. The industry adjustment coefficient K is calculated as follows: ； In the formula, This indicates the proportion of the total adjustable load of load category c to the total industry load; This represents the power-weighted average constraint correction factor for the c-th load category; This indicates the number of devices in the c-th load category; Indicates the total number of equipment in an industry or enterprise; S5. Identify key random parameters that significantly affect the measurement results, set the probability distribution corresponding to the key random parameters, use the adaptive Monte Carlo simulation method with convergence judgment mechanism to generate random parameter samples based on the set probability distribution, substitute them into the progressive measurement model to complete the batch calculation of the load adjustable ratio, perform convergence judgment, perform statistical analysis on all sample results after convergence, and output the uncertainty measurement results. S6. Train the XGBoost regression model to predict the deviation correction value of the current measurement result, and make online corrections to the comprehensive weights of the load adjustable ratio, industry adjustment coefficient and various constraints, and output the final measurement result after closed-loop optimization.

2. The industrial adjustable load calculation method based on multidimensional load classification and scenario adaptation according to claim 1, characterized in that, In S1, the static basic data includes the rated parameters of industrial enterprise equipment, production process attributes, load category of equipment, and grid connection point information; the dynamic operation data includes the real-time operating power of equipment, production status data, grid dispatch instruction data, meteorological data, environmental compliance constraint data, and equipment operation and maintenance monitoring data. After obtaining the raw data, data cleaning is performed to remove invalid duplicate data and abnormal data that does not conform to physical boundaries.

3. The industrial adjustable load calculation method based on multidimensional load classification and scenario adaptation according to claim 1, characterized in that, In S1, the multidimensional refined load classification system divides industrial load into four dimensions, specifically: The first dimension is the load category, which includes three load categories: core production load, auxiliary production load, and non-production load. The second dimension is the subcategories, which break down the three major load categories into core and optional categories, including six subcategories: core process load, non-core process load, necessary auxiliary load, optional auxiliary load, necessary non-production load, and optional non-production load. The third dimension is the single device level, which establishes a standardized parameter input interface for each device, covering rated power, device type, process stage, load category, and historical operating data. The fourth dimension is the time scale, which is divided into three time scales according to the core demand response scenarios of the power grid: minute-level emergency response, hour-level peak response, and day-level day-ahead response. For each sub-category, the upper limit of the allowable adjustment ratio and the constraint boundary under the corresponding time scale are matched.

4. The industrial adjustable load calculation method based on multidimensional load classification and scenario adaptation according to claim 1, characterized in that, In S2, the standardized value rules for the correction coefficients of various constraints are as follows: For production safety constraints, the correction coefficient for core non-shutdown equipment is [0.1, 0.3], and the correction coefficient for non-core shut-down equipment is [0.8, 1.0]. For environmental compliance constraints, the correction coefficient for environmental protection equipment that needs to operate synchronously with the main equipment is [0.2, 0.4], and the correction coefficient for other equipment is [0.9, 1.0]. For equipment characteristic constraints, the correction coefficient for equipment whose response time and adjustment range meet the scheduling requirements is [0.9, 1.0], and the correction coefficient for equipment that does not meet the scheduling requirements is [0.3, 0.5].

5. The industrial adjustable load calculation method based on multidimensional load classification and scenario adaptation according to claim 1, characterized in that, In S4, the process of outputting the total adjustable load and calculating the load adjustable ratio and industry adjustment coefficient is as follows: S4.1 Calculate the actual adjustable load of a single device: ； In the formula, This represents the actual adjustable load of the i-th device; This represents the design rated power of the i-th device; This represents the allowable adjustment ratio of the i-th device in the target time scale scenario; S4.2 Calculate the total adjustable load of each subclass: ； In the formula, This represents the total adjustable load of the subclass "sub"; The number of devices in each sub-category; S4.3 Calculate the total adjustable load of the load category: ； In the formula, This represents the total adjustable load of load category c. ; This represents the number of subcategories under the c-th major load category; S4.4, Total adjustable output load: ； In the formula, This represents the total adjustable load of an industry or enterprise; S4.5, Calculate the adjustable load ratio : ； In the formula, Indicates the total load of an industry or enterprise; S4.6 Calculate the industry adjustment coefficient K.

6. The industrial adjustable load calculation method based on multidimensional load classification and scenario adaptation according to claim 4, characterized in that, In S5, the key random parameters include the single-equipment allowable adjustment ratio, the production safety constraint sub-item correction coefficient, the environmental compliance constraint sub-item correction coefficient, and the equipment characteristic constraint sub-item correction coefficient. The probability distribution corresponding to the key random parameters is set based on industry statistical data and production operation characteristics. In the adaptive Monte Carlo simulation method with convergence determination mechanism, the initial number of simulations is set. Maximum number of simulations Convergence threshold Iteration step size Random parameter samples are generated based on a set probability distribution and substituted into a progressive calculation model to complete batch calculations of the load adjustable ratio. The next sampling is performed to calculate the mean and coefficient of variation of the current adjustable load proportions for all samples. Perform a convergence check; if the absolute value of the difference between the coefficients of variation of two adjacent iterations is less than the convergence check threshold... Or the number of simulations reaches the maximum number of simulations. If the result is not found, the iteration stops; otherwise, the number of simulations is increased, and the sampling and calculation process is repeated. Perform statistical analysis on all converged sample results and output the uncertainty calculation results, including the mean, standard deviation and 95% confidence interval of the load adjustable ratio.

7. The industrial adjustable load calculation method based on multidimensional load classification and scenario adaptation according to claim 1, characterized in that, In S6, the actual adjustable load, actual adjustment ratio, equipment operating parameters, constraint data, and scenario parameter data from historical demand response events of the target industry or enterprise are collected to construct a historical dataset including feature sets and label sets. The historical dataset is divided into training and testing sets in a 7:3 ratio to train the XGBoost regression model. The input parameters calculated by the progressive calculation model are substituted into the trained XGBoost regression model to predict the deviation correction value of the current calculation result. Based on the deviation correction value, the comprehensive weights of the load adjustable ratio, industry adjustment coefficient and various constraints are corrected online. The calculation results of the progressive calculation model are updated based on the correction results, and the final calculation result after closed-loop optimization is output.

Images