Method, apparatus, computer equipment, readable storage medium, and program products for probabilistic prediction of the adjustability of building clusters based on electro-carbon coupling.

By constructing eigenvector and conditional quantile models, and combining the electric-carbon coupling strength and carbon compliance pressure to correct the load prediction range, the problem of overestimation of regulation potential caused by the lack of consideration of multiple factors in the existing technology is solved, and a more accurate assessment of the regulation capacity of building clusters is achieved, reducing the risks of electricity and carbon compliance.

CN123073723APending Publication Date: 2026-11-10SHENZHEN POWER SUPPLY BUREAU
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611311165.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-27
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing methods for assessing the regulation potential of building clusters fail to comprehensively consider various factors, leading to an overestimation of the actual achievable regulation potential and triggering risks of power response failures and carbon compliance violations.

Method used

By acquiring actual load, meteorological data, and building status data of building clusters, feature vectors are constructed. A conditional quantile model is used to predict the basic load, and the carbon-electric coupling strength and carbon compliance pressure are combined for correction to obtain the carbon-corrected prediction range. Considering carbon market constraints, the load uncertainty and risk are quantified.

Benefits of technology

It provides more accurate predictions of the adjustment capacity of building clusters, reduces the risk of default in market declarations, and enhances the credibility and practical reference value of the assessment results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123073723A_ABST
    Figure CN123073723A_ABST
Patent Text Reader

Abstract

This application relates to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for probabilistic prediction of the adjustability capacity of building clusters based on electro-carbon coupling. The method includes: acquiring the actual load, weather, time period, and building status of each sub-building within the building cluster during the prediction period; aggregating the sub-building data to obtain cluster-dimensional data and constructing a feature vector; inputting the feature vector into a trained conditional quantile model to obtain the basic load prediction interval for the building cluster; correcting the basic load prediction interval based on the electro-carbon coupling strength related to carbon price and marginal emission factors, as well as carbon compliance pressure; and combining this with physical feasibility truncation to obtain the achievable adjustment potential interval. This invention combines probabilistic interval prediction, electro-carbon risk correction, and physical constraint truncation, taking into account both carbon compliance risk and hard constraints on building operation, reducing the risk of market declaration default by aggregators, and improving the engineering practicality and reliability of demand-side resource assessment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of intelligent building energy and carbon management technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium and computer program product for probabilistic assessment of the adjustment potential of building clusters. Background Technology

[0002] Against the backdrop of the low-carbon transformation of the power system and the market-based participation of demand-side resources, a system for coupled and coordinated regulation and market-based trading of electricity and carbon is gradually taking shape. Building clusters are demand-side controllable resources, and assessing their adjustable potential can provide a reference for market application and resource dispatch. This has led to the development of various technologies for assessing the adjustable potential of building clusters, such as equipment rated parameter calculation, thermal mechanism simulation, and quantile regression probability prediction.

[0003] However, the current mainstream method for assessing the regulatory potential of building clusters is still mainly based on traditional thermal mechanism simulation and equipment rated parameter calculation. It estimates the capacity reduction based on temperature, load history, building thermal inertia, or equipment capacity, and usually outputs a single value.

[0004] However, the current mainstream technical solutions based on thermal mechanism simulation and equipment rated parameter calculation do not take into account the influence of various factors, which can easily overestimate the actual adjustment potential of buildings, resulting in the inability to fulfill market-declared capacity, and thus triggering risks of power response failure and carbon compliance exceeding standards. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for predicting and evaluating the adjustability of building clusters based on electro-carbon coupling, in order to address the above-mentioned technical problems.

[0006] Firstly, this application provides a probabilistic prediction method for the adjustability of building clusters based on electro-carbon coupling, including:

[0007] Obtain the actual load, meteorological data, time period data, and building status data of each sub-building in the building cluster within the predicted time period;

[0008] Feature vectors are obtained based on actual load, meteorological data, time period data, and building status data;

[0009] The feature vectors are input into the trained conditional quantile model to obtain the basic load prediction range for the building cluster.

[0010] The base load forecast range is corrected based on the electro-carbon coupling strength and carbon compliance pressure to obtain the carbon-corrected forecast range; where the electro-carbon coupling strength is related to carbon price and marginal emission factor.

[0011] In one embodiment, the base load forecast interval is corrected based on the electro-carbon coupling strength and carbon compliance pressure to obtain a carbon-corrected forecast interval, including:

[0012] The electro-carbon coupling strength and carbon compliance pressure were normalized respectively.

[0013] The carbon impact variables are determined based on the normalized electro-carbon coupling strength and carbon compliance pressure, as well as the preset electro-carbon coupling weights and carbon compliance pressure weights.

[0014] The carbon-corrected forecast interval is obtained by correcting the midpoint of the base load forecast interval based on the carbon impact variable and by correcting the width of the base load forecast interval based on the carbon price volatility within a preset time period.

[0015] In one embodiment, the method further includes:

[0016] Obtain the candidate carbon compliance pressure weight set and the preset comprehensive calibration target;

[0017] The carbon compliance pressure weights are determined based on the obtained set of candidate carbon compliance pressure weights and the preset comprehensive calibration target.

[0018] In one embodiment, the method further includes:

[0019] Under the condition that the preset boundary conditions are met, obtain the reference adjustment capacity threshold of each sub-building in the building cluster within the preset time period;

[0020] In one embodiment, the carbon correction prediction interval is truncated based on a reference adjustment capacity threshold, including:

[0021] The lower limit of correction is determined based on the midpoint and width of the carbon correction prediction interval;

[0022] The upper limit of correction is determined based on the reference regulation capacity threshold and the midpoint and width of the carbon correction prediction interval; wherein the upper limit of correction is less than or equal to the reference regulation capacity threshold.

[0023] The potential prediction range is obtained based on the lower correction limit.

[0024] In one embodiment, the training method for the conditional quantile model includes:

[0025] Obtain the historical aggregated baseline load and historical aggregated actual load of the building cluster;

[0026] Historical target load is determined based on the difference between historical aggregated baseline load and historical aggregated actual load.

[0027] Acquire historical load data, historical meteorological data, historical time period data, and historical building status data;

[0028] Construct sample feature vectors based on historical load, historical meteorological data, historical time period data, and historical building status data;

[0029] The historical target load is set as the ground truth label for training the conditional quantile model. Based on the historical target load and the sample feature vector, the training of the conditional quantile model is completed.

[0030] Secondly, this application also provides a probabilistic prediction device for the adjustable capacity of building clusters based on electro-carbon coupling, comprising:

[0031] The data acquisition module is used to acquire the actual load, meteorological data, time period data, and building status data of each sub-building in the building cluster during the predicted time period.

[0032] The feature vector construction module is used to obtain feature vectors based on the actual load, the meteorological data, the time period data, and the building status data;

[0033] The basic interval construction module is used to input the feature vector into the trained conditional quantile model to obtain the basic load prediction interval of the building cluster.

[0034] The potential prediction interval construction module is used to correct the base load prediction interval based on the electro-carbon coupling strength and carbon compliance pressure to obtain a carbon-corrected prediction interval; wherein, the electro-carbon coupling strength is related to carbon price and marginal emission factor.

[0035] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0036] Obtain the actual load, meteorological data, time period data, and building status data of each sub-building in the building cluster within the predicted time period;

[0037] Feature vectors are obtained based on actual load, meteorological data, time period data, and building status data;

[0038] The feature vectors are input into the trained conditional quantile model to obtain the basic load prediction range for the building cluster.

[0039] The base load forecast range is corrected based on the electro-carbon coupling strength and carbon compliance pressure to obtain the carbon-corrected forecast range; where the electro-carbon coupling strength is related to carbon price and marginal emission factor.

[0040] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0041] Obtain the actual load, meteorological data, time period data, and building status data of each sub-building in the building cluster within the predicted time period;

[0042] Feature vectors are obtained based on historical load, meteorological data, time period data, and building status data;

[0043] The feature vectors are input into the trained conditional quantile model to obtain the basic load prediction range for the building cluster.

[0044] The base load forecast range is corrected based on the electro-carbon coupling strength and carbon compliance pressure to obtain the carbon-corrected forecast range; where the electro-carbon coupling strength is related to carbon price and marginal emission factor.

[0045] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0046] Obtain the actual load, meteorological data, time period data, and building status data of each sub-building in the building cluster within the predicted time period;

[0047] Feature vectors are obtained based on actual load, meteorological data, time period data, and building status data;

[0048] The feature vectors are input into the trained conditional quantile model to obtain the basic load prediction range for the building cluster.

[0049] The base load forecast range is corrected based on the electro-carbon coupling strength and carbon compliance pressure to obtain the carbon-corrected forecast range; where the electro-carbon coupling strength is related to carbon price and marginal emission factor.

[0050] The aforementioned method, apparatus, computer equipment, computer-readable storage medium, and computer program product for probabilistic prediction of the adjustability capacity of building clusters based on electro-carbon coupling collect raw data on actual load, weather, time period, and building status of each sub-building within the building cluster. This raw data is aggregated to obtain aggregated data corresponding to the building cluster. A feature vector is constructed from the aggregated data, and this cluster feature vector is input into a conditional quantile model to obtain the basic load prediction range for the building cluster. The range is then corrected by combining electro-carbon coupling strength and carbon compliance pressure. On one hand, by aggregating the raw information of each sub-building to construct the cluster feature vector, the operating status of multiple buildings within the cluster can be comprehensively considered, fully preserving the operational statistics at the cluster level. The basic load prediction range output by the conditional quantile model can quantify the uncertainty of the building cluster load, overcoming the deficiency of deterministic single-point output in representing load fluctuations. On the other hand, by introducing the electro-carbon coupling strength related to carbon price and marginal emission factors, as well as carbon compliance pressure, to correct the basic load range, carbon market constraints are incorporated into the potential assessment process. This ensures that the obtained carbon-corrected prediction range can take into account both power operation characteristics and carbon compliance risks, providing a more realistic reference result for aggregators to conduct joint electro-carbon market declarations. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is an application environment diagram of a probabilistic prediction method for the adjustability of building clusters based on electro-carbon coupling in one embodiment.

[0053] Figure 2 This is a flowchart illustrating a probabilistic prediction method for the adjustability of building clusters based on electro-carbon coupling in one embodiment.

[0054] Figure 3 This is a flowchart illustrating the carbon correction steps for the base load forecast range in one embodiment.

[0055] Figure 4 This is a flowchart illustrating the steps for physically truncating the carbon-corrected prediction range in one embodiment.

[0056] Figure 5 This is a flowchart illustrating the training steps of a conditional quantile model in one embodiment.

[0057] Figure 6This is a structural block diagram of a probabilistic prediction device for the adjustable capability of a building cluster based on electro-carbon coupling in one embodiment.

[0058] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0061] The probabilistic prediction method for the adjustability of building clusters based on electro-carbon coupling provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server.

[0062] Specifically, terminal 102 can transmit the actual load, meteorological data, time period data, and building status data of each sub-building in the building cluster within the predicted time period to server 104. Server 104 obtains feature vectors based on the actual load, meteorological data, time period data, and building status data. Then, the feature vectors are input into the conditional quantile model trained in server 104 to obtain the basic load prediction interval of the building cluster. Finally, the basic load prediction interval is corrected according to the electro-carbon coupling strength and carbon compliance pressure to obtain the carbon-corrected prediction interval.

[0063] Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection equipment. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0064] In one exemplary embodiment, such as Figure 2 As shown, a probabilistic prediction method for the adjustability of building clusters based on electro-carbon coupling is provided, and this method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 202 to 206. Wherein:

[0065] Step 202: Obtain the actual load, meteorological data, time period data, and building status data of each sub-building in the building cluster within the predicted time period.

[0066] A building cluster refers to a collection of two or more geographically concentrated individual buildings that can participate in demand-side response regulation; the cluster includes one or more types of buildings such as office buildings, commercial buildings, and park buildings. A sub-building refers to an individual building within the building cluster. Actual load refers to the sum of the measured electricity loads of all sub-buildings within the building cluster during the corresponding metering period. Meteorological data includes indoor and outdoor temperature and humidity, as well as weather type. Outdoor temperature and humidity refer to the measured temperature and relative humidity of the building's external environment, which are core meteorological drivers affecting air conditioning heating and cooling loads; indoor temperature and humidity refer to the actual indoor temperature and humidity of each room in the building, constraining the upper and lower limits of air conditioning regulation, reflecting the user's thermal comfort boundary, and are key characteristics of building physical operation. Weather type refers to the classification of the day's meteorological conditions, such as sunny, cloudy, rainy, windy, high temperature, cold wave, etc.; weather type is used to distinguish the building load baseline under different meteorological scenarios. Time-period data refers to calendar and weekday identifiers, a type of time-series static feature that includes year, month, day, hour, holiday markers, and binary identifiers for weekdays and rest days. It is used to distinguish the differentiated energy consumption patterns of residential and commercial buildings on weekdays, weekends, and statutory holidays, and to eliminate load disturbances caused by time cycles. Building status data refers to the status data of equipment such as air conditioning, energy storage, and charging piles inside the building. This overall equipment status data characterizes the hardware constraints of the building's current adjustable flexible resources and is the core physical characteristic that determines the building's maximum adjustable potential.

[0067] In practice, the server acquires the actual load, meteorological data, time period data, and building status data of the building cluster within the predicted time period. First, the server collects the historical load time series data of each sub-building through the BMS building automation system and EMS energy management system of each sub-building; it acquires meteorological data such as outdoor temperature, outdoor humidity, and weather type through on-site meteorological acquisition terminals; it generates time period data with calendar identifiers and workday identifiers according to discrete time intervals consistent with the metering granularity of electricity market settlement; then, it collects building status data such as air conditioning operation status, energy storage SOC, and charging pile operating conditions.

[0068] Finally, the server performs time-series alignment and outlier removal on the actual load, meteorological data, time period data, and building status data of the building cluster.

[0069] Step 204: Obtain feature vectors based on actual load, meteorological data, time period data, and building status data.

[0070] The feature vector includes the cluster's actual load, meteorological characteristics, calendar information, and building summary features. The building status summary features can be composed of the concatenation of the statuses of each sub-building, statistics, or a fixed-dimensional vector formed by a feature extractor.

[0071] For example, the server first performs missing value imputation and outlier filtering on the collected actual load data and meteorological data, encodes non-numerical information such as time period and building status, and extracts features corresponding to meteorological conditions, time period, and building status. These features are then concatenated to form a fixed-dimensional feature vector, thus converting the original collected data into a feature vector. The specific formula for representing the feature vector is as follows:

[0072] (1)

[0073] in This represents the actual load characteristics. As a meteorological feature, For calendar or time period characteristics, For cluster building status characteristics, This is the final feature vector formed by the server.

[0074] Step 206: Input the feature vector into the trained conditional quantile model to obtain the basic load prediction range of the building cluster.

[0075] The conditional quantile model represents the q-th quantile estimate of the target response under given physical operating characteristics. The basic load forecast interval for the building cluster refers to the basic load forecast interval directly output by the conditional quantile model based on the eigenvectors, used to characterize the load adjustment range or uncertainty caused by the building's own operating conditions. Next, the total span of the basic load forecast interval, obtained by subtracting the lower bound from the upper bound, is the basic width, which characterizes the uncertainty of the adjustment potential; the larger the basic width, the higher the uncertainty caused by load fluctuations and random equipment disturbances. The basic midpoint is obtained by the arithmetic mean of the upper and lower bounds of the basic load forecast interval, representing the median estimate of the building cluster's adjustment potential output by the conditional quantile model without carbon risk correction. The lower and upper bounds of the interval correspond to the minimum and maximum theoretically adjustable loads of the building cluster at different confidence levels, respectively, quantifying the load adjustment uncertainty solely based on the building's own patterns, equipment physical operating constraints, and environmental time-series characteristics.

[0076] In practice, the server uses the feature vector generated in the previous step as input to the conditional quantile model. The conditional quantile model, based on linear quantile regression, obtains the base load forecast interval. The base width and base midpoint are then derived from this interval. The specific calculation formula is as follows:

[0077] (2)

[0078] (3)

[0079] In formula (2) Characterizes the base width, Characterize the midpoint of the base. Characterizes the upper limit of the basic load forecast interval. Characterizes the lower bound of the basic load forecast interval.

[0080] In optional embodiments, the conditional quantile model includes, but is not limited to, gradient boosting quantile regression, random forest quantile regression, or neural network quantile models. Based on the above feature vectors and conditional quantile models, the required upper and lower conditional quantiles can be obtained. These upper and lower conditional quantiles are used as the upper and lower limits of the basic load forecast interval, thereby obtaining the basic load forecast interval.

[0081] Step 208: Correct the base load prediction range based on the electro-carbon coupling strength and carbon compliance pressure to obtain the carbon correction prediction range.

[0082] The electricity-carbon coupling strength is related to carbon price and marginal emission factor. Carbon price is the trading price of carbon emission allowances within the carbon trading market, representing the economic cost of carbon emissions. Marginal emission factor, on the other hand, represents the incremental carbon emissions corresponding to each additional unit of electricity consumption in the power system, used to quantify the carbon emission changes resulting from load regulation. Carbon price and marginal emission factor jointly participate in the calculation of electricity-carbon coupling strength, reflecting the carbon compliance risk corresponding to the electricity regulation behavior of building clusters. Electricity-carbon coupling strength is used to characterize the carbon cost or carbon burden corresponding to a unit of regulation.

[0083] Carbon compliance pressure is related to carbon price volatility, cumulative emissions, and free allowances or compliance benchmarks. Carbon price volatility characterizes the degree of fluctuation in carbon trading prices over a recent period, reflecting the uncertainty risk of carbon market settlement. A higher volatility value indicates significant recent price fluctuations, high future price uncertainty, and high market settlement risk; a lower volatility value indicates stable recent carbon prices, small price changes, and low market settlement risk. Carbon compliance pressure is used to characterize the relative extent to which cumulative emissions exceed free allowances or compliance benchmarks. For example, after obtaining the base load forecast range, the server obtains carbon price, carbon emissions, and electricity consumption through an external interface. Based on carbon emissions and electricity consumption, a marginal emission factor is calculated. Based on the carbon price and marginal emission factor, the electro-carbon coupling strength is obtained. The specific calculation formula is as follows:

[0084] (4)

[0085] In formula (4), E t For carbon emissions, D t For electricity consumption, For carbon price, χ t For the electrocarbon coupling strength, MEF t This is the marginal emission factor.

[0086] The server calculates the average carbon price within the rolling window based on carbon price volatility, using the following formula:

[0087] (5)

[0088] In formula (5) This represents the arithmetic mean of carbon prices within the rolling window at time t. Indicates historical carbon prices, The current carbon price; K represents the backtracking step, K=0 for the current time, K=1 for the previous time. The length of the rolling window used for carbon price calculation.

[0089] Based on the mean carbon price within the rolling window, the server calculates the sample standard deviation of the rolling window, using the following formula:

[0090] (6)

[0091] In formula (6), σc(t) represents the sample standard deviation of the rolling window. To eliminate differences in dimensions and market scale, the server standardizes σc(t) using the mean and standard deviation, and then takes the larger value after rounding down from zero to obtain the non-negative volatility correction input, i.e., carbon compliance pressure. The specific formula is as follows:

[0092] (7)

[0093] In formula (7), This refers to cumulative emissions, This refers to free quotas or performance standards. This is to prevent positive numbers with a denominator of 0 from being used in formula (7). This refers to carbon compliance pressure.

[0094] The carbon-corrected prediction range refers to the probability range of building cluster regulation load obtained by dynamically correcting the upper and lower boundaries of the basic load range based on the basic load prediction range output by the conditional quantile model, combined with carbon price, the electric carbon coupling strength related to marginal emission factors, and carbon compliance pressure.

[0095] In the above-mentioned probabilistic prediction method for the adjustable capacity of building clusters based on electricity-carbon coupling, the server constructs feature vectors and trains a conditional quantile model based on the feature vectors to obtain the basic load prediction range. Carbon market fluctuations and carbon compliance pressures are incorporated into the basic load prediction range to obtain a more accurate potential prediction range, reduce the default risk of market declarations by aggregators, fully explore the utilization value of flexible resources in building clusters, and provide a reliable assessment basis for electricity-carbon collaborative trading.

[0096] In one exemplary embodiment, such as Figure 3 As shown, the process of correcting the base load forecast range based on the electro-carbon coupling strength and carbon compliance pressure to obtain the carbon-corrected forecast range includes steps 302 to 306. Wherein:

[0097] Step 302: Normalize the electro-carbon coupling strength and carbon compliance pressure respectively.

[0098] Normalization processes include, but are not limited to, linear scaling normalization and zero-mean standardization. Linear scaling normalization compresses features of different dimensions and numerical ranges into the 0-1 interval, preserving the relative magnitudes of the original data. Zero-mean standardization eliminates the influence of dimensions, transforming features into a value representing a deviation from the mean by a certain number of standard deviations, characterizing relative fluctuations, and preserving variance information.

[0099] In practice, the server statistically obtains and saves the mean and standard deviation of the electro-carbon coupling strength and the performance pressure based on the historical training and calibration dataset; the server obtains the original electro-carbon coupling strength and performance pressure at the prediction time, reuses the mean and standard deviation obtained from the training set to perform zero-mean standardization on the two indicators respectively, and obtains the normalized electro-carbon coupling strength and the normalized performance pressure.

[0100] Step 304: Determine the carbon impact variables based on the normalized electro-carbon coupling strength and carbon compliance pressure, as well as the preset electro-carbon coupling weights and carbon compliance pressure weights.

[0101] The preset weights for electricity-carbon coupling and carbon compliance pressure are hyperparameters pre-defined in the conditional quantile model. The electricity-carbon coupling weight controls the magnitude of the influence of the electricity-carbon coupling strength on the overall carbon impact variable. A larger weight indicates a stronger effect of fluctuations in carbon prices and the marginal emission factor of the power grid on the downward adjustment of the midpoint of the interval; a smaller weight indicates a weaker correction effect from this factor. The carbon compliance pressure weight characterizes the magnitude of the influence of compliance pressure on the overall carbon impact variable. A larger weight indicates a stronger effect of carbon quota gaps and compliance pressure on the downward adjustment of the midpoint of the interval; a smaller weight indicates a weaker correction effect from compliance pressure.

[0102] In practice, the server determines the carbon impact variables based on the normalized electro-carbon coupling strength and the normalized carbon compliance pressure, combined with preset electro-carbon coupling weights and preset carbon compliance pressure weights. The specific formula is as follows:

[0103] (8)

[0104] In formula (8), For carbon-related variables, The preset electro-carbon coupling weights, The normalized electro-carbon coupling strength, The preset carbon compliance pressure weight, This represents the normalized carbon compliance pressure.

[0105] Step 306: Correct the midpoint of the base load forecast interval based on the carbon impact variable, and correct the width of the base load forecast interval based on the carbon price volatility within a preset time period to obtain the carbon-corrected forecast interval.

[0106] The midpoint correction indicates that when carbon burden or carbon compliance pressure increases, the server will lower the center of the committable capacity range to prevent aggregators from making overly aggressive declarations when carbon risk is high. The range width correction indicates that when carbon price volatility increases, the system will widen the range width as a buffer against market uncertainty.

[0107] In practice, the server obtains the midpoint of the basic load forecast interval and the width of the basic load forecast interval calculated by the conditional quantile model.

[0108] The midpoint of the interval correction is calculated using the following formula:

[0109] (9)

[0110] In formula (9), The midpoint of the corrected interval is obtained after considering the influence of carbon variables based on the midpoint of the base interval. The midpoint of the basic interval, For midpoint correction hyperparameters, The carbon-related variables are used as the basis for adjusting the midpoint of the predicted interval based on the midpoint correction hyperparameter and the carbon-related variables. This is done to shift the midpoint of the interval downwards when the carbon risk is high, thus preventing aggregators from submitting overly aggressive reports.

[0111] The server performs width correction calculations simultaneously with the midpoint calculation of the correction interval. The server calculates the corrected interval width based on the base interval width and the sample standard deviation of the scrolling window, using the following formula:

[0112] (10)

[0113] In formula (10), Characterizes the corrected interval width. Characterizes the width of the basic interval. The sample standard deviation characterizing the rolling window. This is the width correction hyperparameter. The server obtains the corrected interval width based on the base interval width and the sample standard deviation of the rolling window. This corrected interval width is used to accommodate price uncertainty in the carbon market; as carbon price volatility increases, the corrected interval width becomes larger.

[0114] The server uses the corrected midpoint of the interval. and corrected interval width Calculate the lower limit of correction respectively. Correct the upper limit value It is composed of the lower correction limit and the upper correction limit. , The interval is used as the carbon correction prediction interval.

[0115] In this embodiment, the midpoint of the interval is lowered by adjusting the coupling strength of electricity and carbon and the compliance pressure, and the width of the interval is expanded according to the carbon price fluctuation, which enhances the risk perception capability in the electricity-carbon synergy scenario and enables the assessment results to be dynamically adjusted with the carbon market status.

[0116] In one exemplary embodiment, the method further includes:

[0117] Obtain the candidate carbon compliance pressure weight set and the preset comprehensive calibration target, and then determine the compliance pressure weight based on the obtained candidate carbon compliance pressure weight set and the preset comprehensive calibration target.

[0118] The candidate carbon compliance pressure weight set refers to a pre-configured set of candidate parameters, containing multiple candidate carbon compliance pressure weights of different sizes. This set is used for parameter optimization and traversal to select the optimal carbon compliance pressure weight. The preset comprehensive calibration target refers to a pre-defined multi-indicator joint evaluation criterion used to evaluate the effectiveness of the weights. This includes pinball loss, predicted interval coverage probability (PICP), and average interval width (MPIW). It measures the overall performance of the corrected potential interval by comprehensively evaluating multiple indicators.

[0119] In practice, the server obtains a pre-configured set of candidate carbon compliance pressure weights and a preset comprehensive calibration target. The set of candidate carbon compliance pressure weights contains multiple sets of different alternative values ​​for carbon compliance pressure weights, and the preset comprehensive calibration target comprehensively includes pinball loss, predicted interval coverage probability (PICP), and average interval width (MPIW). The server iterates through each candidate value in the set of candidate carbon compliance pressure weights, substitutes them into the calibration calculation process, performs optimization evaluation based on the preset comprehensive calibration target, selects the candidate value with the best overall performance, determines the carbon compliance pressure weight, and saves it.

[0120] In one exemplary embodiment, the method further includes:

[0121] Under the condition that the preset boundary conditions are met, obtain the reference adjustment capacity threshold of each sub-building in the building cluster within the preset time period.

[0122] Among these, satisfying the preset boundary conditions means that in the process of solving for the theoretical maximum adjustable capacity, the upper and lower limits of the state, control, and SOC are used as inequality constraints. This requires that the building's thermal state at each prediction time not exceed the upper and lower limits of the state, the equipment control inputs not exceed the upper and lower limits of the control, and the energy storage's state of charge not exceed the upper and lower limits of the SOC. Only solutions that simultaneously satisfy all of these constraints are considered valid and feasible. The upper and lower limits of the state mean that, when recursively calculating the building's thermal state at each prediction time, thermal state variables such as indoor temperature must be between the upper and lower limits of the state, and cannot exceed the allowable range for human thermal comfort. The upper and lower limits of the control mean that all control inputs and values, such as air conditioning setpoints, energy storage charging and discharging power, and charging pile power, must fall within the lower and upper limits of the control allowed by the hardware equipment, and no control quantities that the equipment cannot execute should be output. The upper and lower limits of the SOC mean that the SOC of the energy storage device throughout the entire prediction period must always be between the upper and lower limits of the SOC, prohibiting overcharging and over-discharging of the battery.

[0123] The reference adjustment capacity threshold refers to the upper limit of physical flexibility, also known as the theoretical maximum adjustable capacity. It is the theoretically maximum load adjustment capacity that can be achieved by solving and optimizing the state space model of a single building under the condition of strictly meeting all hard constraints of the building.

[0124] In practice, the server establishes a discrete short-term state model for each sub-building based on the state vector, control vector, and disturbance vector, as shown in the following formula:

[0125] (11)

[0126] Equation (11) is the discrete-time state-space thermodynamic recursive equation for the i-th sub-building, and it is the core mechanism equation constraint for solving the upper bound of the physical flexibility of the building cluster. In Equation (9) The state vector representing the i-th sub-building at time t+k+1, the state vector Let be the state vector of the i-th sub-building at time t+k, representing the building's thermal state and energy storage state at the current time; it serves as the initial state for state recursion. The input control vector for the i-th sub-building at time t+k represents the control quantity, which includes, but is not limited to, control commands for adjustable devices such as air conditioning set temperature, energy storage charging and discharging power, and transferable load power. Let be the external disturbance vector of the i-th building at time t+k, representing uncontrollable external disturbances, including but not limited to external environmental and human disturbances such as outdoor temperature, solar radiation, human thermal disturbance, and humidity.

[0127] A(i), B(i), and G(i) are state-space coefficient matrices corresponding to the i-th sub-building. A(i) is the state transition matrix, used to describe the influence of the previous time period's state on the next time period's state; B(i) is the control action matrix, used to describe the influence of control inputs on state evolution; and G(i) is the disturbance action matrix, used to describe the influence of external disturbances such as weather and occupancy on state evolution. Each matrix can be obtained by discretizing the building's thermal parameters or by identifying them using historical operating data, and is recalibrated during the model update phase.

[0128] The server uses formula (9) to deduce the thermal state of each sub-building in the prediction time domain, and further obtains the load of each sub-building from the thermal state of each sub-building, and finally substitutes it into the equation. The optimized upper bound of the physical flexibility of the building cluster at time t+h is obtained. The specific formula is as follows:

[0129] (12)

[0130] In formula (12), It represents the aggregated baseline load of the building cluster, the total baseline power consumption of the cluster at time t+h, and the reference power consumption level when no control measures are implemented. It is the actual electrical load of the i-th sub-building at time t+h, which is determined by the control input, the building thermal state equation, and external disturbances. It is the upper bound of the physical flexibility of the building cluster at time t+h. This refers to finding a complete set of control sequences for air conditioning settings, energy storage charging and discharging, etc., within the allowable control range, so that the downward adjustment capacity can reach its maximum value. However, this formula of reaching the maximum value does not mean finding an unlimited maximum, but rather finding the maximum value under constraints. The specific constraint formula is as follows:

[0131] (13)

[0132] In formula (13), state constraints At each future step from K=1 to h, the control constraints are applied. Apply at each control step from K=0 to h-1; when configuring energy storage, also apply 0≤SOC(i,t+k)≤1 or the device given SOC range in the future. The server determines the set of feasible state-control trajectories based on Equations (9) and (11), and then selects the trajectory with the largest cluster load reduction in the target time period by Equation (10), and finally obtains the upper bound of the physical flexibility of the building cluster at the future time t+h.

[0133] In this embodiment, the thermal state of each sub-building within the predicted time domain is obtained through deduction. Combined with preset constraints such as state boundaries and control boundaries, optimization solutions are performed to calculate the upper limit of the physical flexibility of the building cluster. This solution process comprehensively considers the building's thermal dynamics, the controllable range of equipment, indoor comfort constraints, and the operational limitations of energy storage SOC, ensuring that the obtained reference regulation capacity threshold fully meets the actual physical operational feasibility of the building equipment. This avoids the shortcomings of relying solely on statistical predictions where the potential results exceed the actual regulation capacity of the equipment. This step provides a physical cutoff basis for subsequent potential prediction intervals, imposing physical feasibility constraints on the upper limit of the probabilistic prediction output interval. This prevents overestimation of potential during the market application stage, improves the engineering credibility of the building cluster's adjustable potential assessment results, and provides a true and effective physical upper limit reference for aggregators to conduct electricity market applications.

[0134] In one exemplary embodiment, such as Figure 4 As shown, the carbon correction prediction interval is truncated according to the reference adjustment capacity threshold, including steps 402 to 406, wherein:

[0135] Step 402: Determine the lower limit of correction based on the midpoint and width of the carbon correction prediction interval.

[0136] The lower limit of correction refers to the non-negative constraint value obtained by physically truncating the lower limit value in the carbon correction interval.

[0137] In practice, the server obtains the carbon correction midpoint and carbon correction width within the carbon correction prediction interval. First, it subtracts half the carbon correction width from the carbon correction midpoint to calculate the original lower limit of correction. The server then applies a non-negative constraint to this original lower limit, taking the larger of the original lower limit and the value 0 as the non-negative constraint value. This avoids outputting negative capacity results that lack engineering meaning. Subsequently, the server compares this non-negative constraint value with the physical flexibility boundary of the building cluster, taking the smaller of the two to obtain the lower limit of correction. The specific formula is as follows:

[0138] (14)

[0139] In formula (14), This refers to the lower limit of correction. It refers to a physical flexible boundary. This refers to the carbon correction midpoint. It is the carbon correction width. The original lower limit of the potential prediction interval can be obtained by calculating using this formula. This formula characterizes the process of applying non-negative constraints to the original lower limit of the potential prediction interval.

[0140] Step 404: Determine the upper limit of correction based on the reference regulation capacity threshold and the midpoint and width of the carbon correction prediction interval; wherein the upper limit of correction is less than or equal to the reference regulation capacity threshold.

[0141] Among them, the upper limit of correction refers to the non-negative constraint value obtained by physically truncating the upper limit of the carbon correction interval.

[0142] In practice, the server obtains the carbon correction midpoint and carbon correction width of the carbon correction prediction interval. First, it adds half the carbon correction width to the carbon correction midpoint to calculate the original upper limit of the correction. The server then applies a non-negative constraint to this original upper limit, taking the larger of the original upper limit and the value 0 as the non-negative constraint value. This avoids outputting negative capacity results that lack engineering meaning. Subsequently, the server compares this non-negative constraint value with the physical flexibility boundary of the building cluster, taking the smaller of the two to obtain the upper limit of the correction. The specific formula is as follows:

[0143] (15)

[0144] In formula (15), This refers to the upper limit of the correction. It refers to a physical flexible boundary. This refers to the carbon correction midpoint. It is the carbon correction width. The corrected original upper limit can be calculated from this formula. This formula characterizes the process of applying non-negative constraints to the original upper limit of correction.

[0145] Step 406: Obtain the potential prediction range based on the lower correction limit and the upper correction limit.

[0146] The potential prediction range is a range consisting of the lower bound and the upper bound. The lower bound of this range can be used as a conservative reporting reference, the midpoint can be used as an internal planning reference, and the upper bound can be used as the capability boundary within the physical boundary.

[0147] In practice, the server uses the value calculated by equation (14) as the lower bound of the potential prediction interval and the value calculated by equation (15) as the upper bound of the potential prediction interval. Then, the server combines the upper and lower bounds to form the potential prediction interval. The specific formula is as follows:

[0148] (16)

[0149] in For potential prediction range, This is the lower limit of the interval correction. This is the upper limit of the interval correction.

[0150] In this embodiment, the physical feasibility is truncated by the above-mentioned upper and lower limits to obtain the final potential prediction range, thus obtaining an adjustable potential range that takes into account both carbon market risks and building physical operation constraints.

[0151] In one exemplary implementation, such as Figure 5 As shown, the training method for the conditional quantile model includes steps 502 to 510, wherein:

[0152] Step 502: Obtain the historical aggregated baseline load and historical aggregated actual load of the building cluster.

[0153] Among them, the historical aggregated baseline load is the aggregated baseline load of the building cluster in each discrete historical period; the baseline load is the inherent electricity consumption benchmark load of the building cluster; the historical aggregated baseline load is the benchmark electricity consumption level obtained by retrospection at a historical time, which is consistent with the market settlement metering granularity and is obtained by aggregating and summing the baseline loads of each sub-building; the historical aggregated actual load is the total electricity consumption of the building cluster in the corresponding historical discrete metering period, which is obtained by aggregating the actual electricity consumption of each sub-building. This load includes the load changes caused by the control actions of air conditioning, energy storage, and charging piles, and can reflect the real electricity consumption results of the building cluster after control in the historical period.

[0154] In practice, the server reads the historical baseline load and historical actual load of each sub-building from the historical database for each historical period, and then sums the historical baseline load and historical actual load for each sub-building to obtain the corresponding historical aggregated baseline load and historical aggregated actual load. The specific formula is as follows:

[0155] (17)

[0156] In formula (17), This refers to the historical baseline load. This refers to the actual historical load.

[0157] Step 504: Determine the historical target load based on the difference between the historical aggregated baseline load and the historical aggregated actual load.

[0158] The historical target load is obtained by subtracting the historical aggregated baseline load from the historical aggregated actual load. The historical target load can be used as a training label in the conditional quantile model.

[0159] In practice, the server subtracts the historical aggregated baseline load from the historical aggregated actual load to obtain the historical target load. The server then labels and stores this historical target load in its internal memory for use in subsequent training of the conditional quantile model. The specific formula is as follows:

[0160] (18)

[0161] In formula (18), y(t+h) is the historical target load.

[0162] Step 506: Obtain historical load, historical meteorological data, historical time period data, and historical building status data.

[0163] Among them, historical load is the measured time-series data of electricity load of building clusters collected during historical metering periods; historical meteorological data is environmental observation data such as indoor and outdoor humidity and weather type that have been observed during historical periods; historical period data is time attribute information such as calendar, workday identifier, and time corresponding to historical samples. Historical building status data is the allowable status data of building thermal status, air conditioning, energy storage, charging piles, and other equipment recorded during historical periods.

[0164] In practice, the server iterates through each historical metering period, sequentially acquiring historical load, historical meteorological data, historical time period data, and historical building status data corresponding to each historical period. The server retrieves all data according to the time index, performs time alignment processing on multiple types of data based on discrete metering time, and associates and matches historical load, historical meteorological data, historical time period data, and historical building status data belonging to the same historical period, thus completing the collection and organization of multiple types of historical data.

[0165] Step 508: Construct sample feature vectors based on historical load, historical meteorological data, historical time period data, and historical building status data.

[0166] The sample feature vector refers to a fixed-dimensional vector formed by combining historical load, historical meteorological data, historical time period data, and historical building status data collected within the same historical measurement period according to preset feature construction rules. This vector serves as the input data for model training. It is composed of multiple types of information from the historical time period and is used to characterize the permissible and environmental conditions of the building cluster at that historical moment.

[0167] In practice, after completing the time alignment and correlation matching of historical load, historical meteorological data, historical time period data, and historical building status data for each historical period, the server standardizes and integrates the collected historical data according to preset unified feature construction rules. The server systematically splices and regularizes the historical load, historical meteorological data, historical time period data, and historical building status data corresponding to the same historical period, unifies the data units and dimensions, removes abnormal data, and finally generates vector data with fixed dimensions and unified format, which is the sample feature vector.

[0168] Step 510: Set the historical target load as the ground truth label for training the conditional quantile model, and complete the training of the conditional quantile model based on the historical target load and the sample feature vector.

[0169] The conditional quantile model refers to the output of the q-th quantile estimate of the target load y(t+h) under given physical and operational characteristics. This model can use linear quantile regression, whose output is a linear combination of the parameter vector and the feature vector, or it can use gradient boosting quantile regression, random forest quantile regression, or neural network quantile models, as long as the required upper and lower conditional quantiles are output respectively. In this embodiment, linear quantile regression is used.

[0170] In practice, the server first processes the historical data samples, constructing a historical feature vector for each sample n. ; Constructing tags The total sample set is N samples, which are then divided into a training sample set for parameter optimization. Next, the server determines two quantile targets based on the speech loss coverage rate α. The corresponding lower quantile corresponds to the lower bound of the interval. Corresponding to the upper quantile and the upper bound of the interval, two sets of independent parameters are trained using these two parameters respectively. Then, the model parameters are solved by minimizing the pinball loss according to equation (20). Taking quantile q as an example, the server adopts the linear model form of equation (19). Using pinball loss ρq(e) = qe (e ≥ 0) and ρq(e) = (q-1)e (e < 0) as loss functions, summation is performed over all N training samples; The optimal parameters for this quantile were obtained through optimization. The server respectively handles q= q= After performing the above optimization, two sets of parameters were obtained. The two sets of model parameters are stored on the server. Finally, when the server is running online, it constructs a feature vector based on the input. Substitute the eigenvectors into q= The trained linear model yields the lower bound of the basic interval. Substitute the same feature into q= The trained linear model yields the upper bound of the basic interval. And thus output the basic load forecast range [ , ].

[0171] (19)

[0172] (20)

[0173] (twenty one)

[0174] In this embodiment, the conditional quantile model is trained by using the difference between the historical aggregated baseline load and the historical aggregated actual load of the building cluster during the same time period as the historical target load. The actual adjustment response of the cluster is used as the model's learning object, rather than directly modeling the original load. Therefore, the conditional quantile model can directly learn the probability distribution of the building cluster's flexible adjustment capability. The model output directly corresponds to the adjustment potential, eliminating the need for additional result conversion and reducing computational errors introduced by intermediate conversion steps. Model training is completed by matching corresponding feature vectors, ensuring that the feature input and learning target are mutually adapted. This improves the probability prediction accuracy of the conditional quantile model for the building cluster's adjustment potential, resulting in a more reliable baseline load prediction range and reducing the risk of misjudging the cluster's adjustability.

[0175] This embodiment discloses a building cluster regulation potential assessment method based on adaptive correction of electricity carbon risk, which is applied to the scenario of collaborative electricity carbon market declaration for multiple building clusters in a park. The assessment cycle is adapted to the settlement rules of the electricity market and carbon market. It adopts a 15-minute discrete measurement period and the server is the sole execution entity of the whole method. It sequentially completes data collection, conditional quantile model inference, carbon risk midpoint correction, carbon volatility width correction, and physical feasibility truncation. Finally, it outputs a building cluster potential prediction range that can be directly used for market declaration. The specific implementation steps are as follows.

[0176] Step 1: The server collects basic operational data matching the market granularity.

[0177] The server collects real-time operational and external environmental data of the target building cluster in 15-minute market settlement metering discrete time periods. The building cluster comprises multiple commercial and office buildings geographically concentrated and equipped with flexible and adjustable devices such as air conditioning, energy storage, and charging piles, meeting unified control and market application requirements. Specific data collected by the server includes: historical electricity load of each individual building in the cluster, real-time aggregated baseline load, real-time aggregated actual load, outdoor temperature and humidity, sunshine and meteorological data, calendar time period characteristics, building equipment operating status, and energy storage SOC status data; simultaneously, it collects real-time carbon market trading prices, historical carbon price time-series data, and cluster carbon quota remaining data.

[0178] The server performs time-series alignment, outlier removal, and normalization preprocessing on all collected data, uniformly adapts to the 15-minute measurement time granularity, and constructs the model input feature vector to provide standardized data support for subsequent conditional quantile model inference and carbon risk index calculation.

[0179] Step 2: The server infers the basic load forecast interval using a linear conditional quantile model. In this embodiment, linear quantile regression is used to construct the conditional quantile model, and the entire model training and inference process is completed by the server. The server pre-builds a model dataset, using the physical operation characteristics of the building cluster as the input feature vector and the actual load adjustment response of the cluster as the label, and divides it into a training set and a calibration set for model training.

[0180] The model adopts a linear structure: ,in For model intercept, The model uses standardized physical feature vectors. The server optimizes the model by minimizing the sum of the overall losses on the training set using the pinball loss function. The optimization formula is: The server presets the interval coverage rate α=0.05, determines the upper and lower target quantile levels, namely the lower quantile α / 2=0.025 and the upper quantile 1-α / 2=0.975, and trains two independent sets of model parameters for each. After training, the server uses a calibration set combined with three indicators—Pinball loss, interval coverage rate PICP, and average interval width MPIW—to jointly select the optimal model parameters and store them.

[0181] During the online inference phase, the server inputs the preprocessed feature vectors into the trained bisector model in real time, and outputs the upper and lower bounds of the basic load prediction interval according to the formula: , .

[0182] For example, in this embodiment, the model inference outputs the lower bound of the basic load prediction interval. =400KW, the upper limit of the basic load forecast range is =600KW. The server calculates the foundation midpoint and foundation width based on the upper and lower bounds, using the following formula: = 500KW, = 200KW, where the midpoint of the base is the central reference regulating capacity of the base load forecast interval, and the base width is the uncertainty span of the base load forecast interval. Both are purely data-driven results and have not incorporated carbon market risk correction.

[0183] Step 3: The server calculates the carbon risk correction index:

[0184] Based on the preprocessed carbon market data, the server simultaneously calculates two types of core carbon risk indicators for subsequent interval adaptive correction:

[0185] First, calculate the carbon impact variables. The server combines the real-time carbon price with the corresponding electro-carbon coupling strength and the compliance pressure corresponding to the remaining carbon quota of the building cluster, and obtains the carbon impact variables after weighted normalization. In this embodiment, the calculated... =0.96, indicating that the overall carbon risk and carbon compliance pressure of the cluster are at a high level at the current moment. Next, the carbon price volatility is calculated. The server uses a fixed scrolling window to calculate the sample standard deviation of historical carbon price time-series data, quantifying the degree of price volatility and uncertainty in the carbon market. In this embodiment, the calculated standard deviation is... =0.8 indicates that carbon prices have fluctuated sharply recently and market settlement is highly uncertain.

[0186] Step 4: The server performs midpoint carbon risk correction.

[0187] The server retrieves the midpoint correction hyperparameters, which were previously obtained through joint calibration using the training and calibration sets. =80, this hyperparameter is determined using Pinball loss, PICP, and MPIW as joint optimization objectives, balancing model fitting accuracy and interval reliability. The server performs risk offsetting on the baseline midpoint based on the midpoint correction formula: =500-80×0.96=423.2KW. By shifting the carbon risk negatively, the reference capacity at the center of the interval tends to be conservative, effectively avoiding aggressive declarations by aggregators when the carbon risk is high, and reducing the dual risks of exceeding carbon compliance limits and defaulting on electricity market responses.

[0188] Step 5: The server performs width volatility correction.

[0189] The server retrieves the width correction hyperparameters obtained from the joint calibration. =0.3, and the baseline width is adaptively broadened and corrected by incorporating carbon price volatility, as shown in the following formula: =200×(1+0.3×0.8)=248KW. The higher the carbon price volatility, the wider the range, which can fully accommodate the settlement uncertainty brought about by carbon market price fluctuations and avoid the prediction failure problem caused by an excessively narrow range.

[0190] Step 6: The server generates the carbon correction prediction range.

[0191] Based on the corrected midpoint and corrected width, the server calculates the lower and upper limits of the carbon correction prediction interval. This interval has completed all carbon risk corrections and has not undergone physical constraint truncation; it represents the core intermediate calculation result. = =423.2-124=299.2KW = =423.2+124=547.2KW. The server generates carbon correction range [299.2KW, 547.2KW].

[0192] Step 7: The server solves for the upper bound of the physical flexibility of the building cluster.

[0193] Based on the thermodynamic mechanism model of each individual building in the building cluster, and under the premise of satisfying indoor thermal comfort constraints, power constraints of air conditioning and energy storage equipment, and energy storage SOC charging and discharging constraints, the server constructs an optimization objective that maximizes the cluster load reduction by using the future time period control input sequence of each sub-building as the optimization variable: The server obtains the upper bound of the physical flexibility of the building cluster for the current time period through constrained optimization. This value represents the ceiling of the cluster's objective physical adjustment capability and is only used as a constraint boundary, not directly for application. This embodiment obtains the upper bound of physical flexibility. =500KW.

[0194] Step 8: The server performs a physical feasibility truncation to generate the final potential prediction range.

[0195] The server performs physical constraint truncation on the upper and lower limits of the carbon correction prediction range to completely eliminate physically unrealizable values ​​and ensure the engineering feasibility of the output results.

[0196] Lower bound truncation (non-negativity constraint): The server takes the maximum value between 0 and the corrected lower bound value to obtain the physical truncation lower bound value. =299.2KW, eliminating the unreasonable result of negative regulation capacity. Upper limit truncation (physical boundary constraint): The server takes the minimum value between the physical flexibility upper limit and the correction upper limit value to obtain the physical truncation upper limit value. =500KW, avoiding excessively high adjustment capacity that exceeds the limits of equipment and thermal comfort. The final potential prediction range for the server output to adapt to the carbon synergy market is [299.2KW, 500KW].

[0197] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0198] Based on the same inventive concept, this application also provides an electro-carbon coupling-based probabilistic prediction device for building cluster adjustability probability prediction, which is used to implement the above-described electro-carbon coupling-based building cluster adjustability probability prediction method. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the electro-carbon coupling-based building cluster adjustability probability prediction device provided below can be found in the limitations of the electro-carbon coupling-based building cluster adjustability probability prediction method described above, and will not be repeated here.

[0199] In one exemplary embodiment, such as Figure 6 As shown, a probabilistic prediction device for the adjustable capacity of building clusters based on electro-carbon coupling is provided, comprising: a data acquisition module 601, a feature vector construction module 602, a basic interval construction module 603, and a potential prediction interval construction module 604, wherein:

[0200] Data acquisition module 601 is used to acquire historical load, meteorological data, time period data, and building status data of each sub-building in the building cluster within the predicted time period;

[0201] The feature vector construction module 602 is used to obtain feature vectors based on the historical load, the meteorological data, the time period data, and the building status data;

[0202] The basic interval construction module 603 is used to input the feature vector into the trained conditional quantile model to obtain the basic load prediction interval of the building cluster.

[0203] The potential prediction interval construction module 604 is used to correct the base load prediction interval based on the electro-carbon coupling strength and carbon compliance pressure to obtain the potential prediction interval; wherein, the electro-carbon coupling strength is related to carbon price and marginal emission factor.

[0204] In an exemplary embodiment, the data acquisition module 601 is used to acquire the aggregated baseline load and aggregated actual load corresponding to the building cluster during the same time period; and to determine the target load based on the difference between the aggregated baseline load and the aggregated actual load.

[0205] In an exemplary embodiment, the feature vector construction module 602 is used to train the conditional quantile model based on the feature vector, and to obtain the base midpoint and base width based on the trained conditional quantile model.

[0206] In an exemplary embodiment, the basic interval construction module 603 is used to obtain the reference adjustment capacity threshold of each sub-building in the building cluster within a preset time period, provided that preset boundary conditions are met.

[0207] In an exemplary embodiment, the potential prediction interval construction module 604 is used to normalize the electro-carbon coupling strength and the carbon compliance pressure respectively; determine carbon impact variables based on the normalized electro-carbon coupling strength and carbon compliance pressure, as well as preset electro-carbon coupling weights and carbon compliance pressure weights; correct the midpoint of the base load prediction interval based on the carbon impact variables, and correct the width of the base load prediction interval based on the carbon price volatility within the preset time period, to obtain the carbon-corrected prediction interval. Under preset boundary conditions, a reference adjustment capacity threshold for each sub-building in the building cluster within the preset time period is obtained; a lower correction limit is determined based on the midpoint and width of the carbon-corrected prediction interval; an upper correction limit is determined based on the reference adjustment capacity threshold and the midpoint and width of the carbon-corrected prediction interval; wherein the upper correction limit is less than or equal to the reference adjustment capacity threshold; and the potential prediction interval is obtained based on the lower correction limit.

[0208] Each module in the aforementioned probabilistic prediction device for the adjustable capacity of building clusters based on electro-carbon coupling can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0209] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores probabilistic prediction data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a probabilistic prediction method for the adjustable capabilities of building clusters based on electro-carbon coupling.

[0210] Those skilled in the art will understand that Figure 7The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0211] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0212] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the methods of the above embodiments.

[0213] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the methods of the above embodiments.

[0214] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0215] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0216] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A probabilistic prediction method for the adjustability of building clusters based on electro-carbon coupling, characterized in that, The method includes: Obtain the actual load, meteorological data, time period data, and building status data of each sub-building in the building cluster within the predicted time period; Feature vectors are obtained based on the actual load, the meteorological data, the time period data, and the building status data; The feature vector is input into the trained conditional quantile model to obtain the basic load prediction range of the building cluster; The base load forecast interval is corrected based on the electro-carbon coupling strength and carbon compliance pressure to obtain a carbon-corrected forecast interval; wherein the electro-carbon coupling strength is related to carbon price and marginal emission factor.

2. The method according to claim 1, characterized in that, The step of correcting the base load forecast range based on the electro-carbon coupling strength and the carbon compliance pressure to obtain the carbon-corrected forecast range includes: The electro-carbon coupling strength and the carbon compliance pressure are normalized respectively. The carbon impact variables are determined based on the normalized electro-carbon coupling strength and carbon compliance pressure, as well as the preset electro-carbon coupling weights and carbon compliance pressure weights. The carbon-corrected forecast interval is obtained by correcting the midpoint of the base load forecast interval based on the carbon impact variable and by correcting the width of the base load forecast interval based on the carbon price volatility within a preset time period.

3. The method according to claim 1, characterized in that, The method further includes: Obtain the candidate carbon compliance pressure weight set and the preset comprehensive calibration target; The carbon compliance pressure weights are determined based on the acquired set of candidate carbon compliance pressure weights and the preset comprehensive calibration target.

4. The method according to claim 1, characterized in that, The method further includes: Under the condition that the preset boundary conditions are met, the reference adjustment capacity threshold of each sub-building in the building cluster is obtained within a preset time period.

5. The method according to claim 1, characterized in that, The carbon correction prediction interval is truncated based on a reference adjustment capacity threshold, including: The lower limit of the correction is determined based on the midpoint and width of the carbon correction prediction interval; The upper limit of correction is determined based on the reference adjustment capacity threshold and the midpoint and width of the carbon correction prediction interval; wherein the upper limit of correction is less than or equal to the reference adjustment capacity threshold; Based on the lower correction limit, a potential prediction range is obtained.

6. The method according to claim 1, characterized in that, The training methods for the conditional quantile model include: Obtain the historical aggregated baseline load and historical aggregated actual load of the building cluster; The historical target load is determined based on the difference between the historical aggregated baseline load and the historical aggregated actual load. Acquire historical load data, historical meteorological data, historical time period data, and historical building status data; Construct sample feature vectors based on the historical load, historical meteorological data, historical time period data, and historical building status data; The historical target load is set as the ground truth label for training the conditional quantile model. Based on the historical target load and the sample feature vector, the training of the conditional quantile model is completed.

7. A probabilistic prediction device for the adjustability of building clusters based on electro-carbon coupling, characterized in that, The device includes: The data acquisition module is used to acquire the actual load, meteorological data, time period data, and building status data of each sub-building in the building cluster during the predicted time period. The feature vector construction module is used to obtain feature vectors based on the actual load, the meteorological data, the time period data, and the building status data; The basic interval construction module is used to input the feature vector into the trained conditional quantile model to obtain the basic load prediction interval of the building cluster. The potential prediction interval construction module is used to correct the base load prediction interval based on the electro-carbon coupling strength and carbon compliance pressure to obtain a carbon-corrected prediction interval; wherein the electro-carbon coupling strength is related to carbon price and marginal emission factor.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.