Probabilistic carbon flow assessment method for integrated energy system based on random scenario analysis

Abstract

The application discloses a comprehensive energy system probability carbon flow evaluation method based on random scene analysis, and relates to the technical field of low-carbon operation evaluation of comprehensive energy systems. The specific steps of the method are as follows: first, washing the carbon data of grid-park interconnection, and constructing a random probability model; obtaining a weighted typical scene through multi-scale coupling sampling; generating a multi-layer carbon flow sequence by relying on energy storage carbon flow aggregation; building a carbon flow matrix to complete carbon tracking and establish carbon indexes; coupling multi-source tail carbon risks, quantifying extreme carbon emission risks, and outputting an evaluation report; the application builds a three-layer nested time sequence architecture to generate a weighted typical scene, eliminates carbon flow errors caused by single-scale modeling, relies on energy storage memory carbon flow aggregation to restore carbon transfer across time periods, constructs a two-way carbon flow matrix to form standardized carbon indexes, couples multi-source tail carbon risks to quantify extreme carbon emission risks, integrates carbon accounting and risk evaluation, and supports grid-park collaborative low-carbon management and control.

Description

Technical Field

[0001] This invention relates to the field of low-carbon operation assessment technology for integrated energy systems, specifically a probabilistic carbon flow assessment method for integrated energy systems based on stochastic scenario analysis. Background Technology

[0002] The construction of regional energy integration continues to advance, and the distribution network and industrial parks have formed a deeply interconnected integrated energy operation system. The system integrates multiple types of power generation equipment, diverse cooling, heating, and electrical energy units, energy storage equipment, and carbon capture devices. The distribution network and the park form a continuous two-way power interaction relationship through the interconnection line. During the energy transmission process, carbon media are transferred and conducted simultaneously in different subjects and at different time dimensions. Carbon flow quantitative assessment is the core technical means to evaluate the low-carbon operation status of the system and formulate energy operation optimization strategies. The assessment work needs to fully take into account the random fluctuations in operation caused by multiple variables such as new energy power generation output, user-side energy demand, energy trading prices, and grid power supply carbon intensity. At the same time, it needs to adapt to the analysis needs of three different time dimensions: long-term planning, daytime scheduling, and real-time operation. Currently, the industry generally adopts probabilistic modeling to describe the fluctuation characteristics of various variables in the energy system, simplifies massive operating conditions by relying on scenario generation methods, and uses carbon flow tracking to statistically analyze the carbon emission status of each node in the network. Carbon emission risk assessment is also gradually being incorporated into the integrated energy system analysis framework. Probabilistic carbon flow quantitative assessment technology under multi-scale and multi-subject coupling conditions has become a key research direction in the field of integrated energy.

[0003] Existing technologies for carbon flow assessment of integrated energy systems have several significant limitations. Most modeling systems only select a single time scale to build scenario models, failing to achieve nested long-term and short-term time series constraints. They also struggle to simultaneously reflect the correlation between annual energy resource fluctuations, daily supply and demand baseline changes, and short-term load shocks. Furthermore, they do not distinguish between the two differentiated interactive operating states of the park purchasing electricity from the grid and the park supplying electricity to the grid. The generated scenario samples cannot fully cover all actual operating modes of the system. Traditional models use a simplified static approach to the carbon transfer process of energy storage devices, ignoring carbon intensity retention during the charging phase and carbon content during the discharging phase. The time-series memory characteristics of the distribution network mean that the aggregation and calculation of carbon flow data at different time scales lacks dynamic carbon memory update logic, resulting in significant deviations in the final carbon flow accounting values. The carbon flow transfer calculation only considers the carbon inflow brought by the power grid to the park in one direction, lacking a two-way carbon exchange quantitative analysis framework. It is impossible to accurately count the total amount of two-way carbon flow between the distribution network and the park. Most carbon emission risk assessments calculate local risks separately and cannot simultaneously couple the three types of risk sources: internal energy fluctuations, external power purchase fluctuations, and interconnection line interaction fluctuations. It is difficult to quantify the tail risk of excess carbon emissions under extreme operating conditions and cannot output complete risk management support content at different scales and levels. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a probabilistic carbon flow assessment method for integrated energy systems based on stochastic scenario analysis. First, multi-dimensional operational data from the distribution network and industrial parks are collected. Abnormal samples are cleaned and probability distributions of various uncertainty parameters are fitted. A global probability benchmark model is built, and a three-layer nested time-series architecture is constructed. A spatiotemporal coupling fusion algorithm generates stochastic scenarios that differentiate between electricity purchase and sale conditions. After clustering and compression, a set of typical scenarios with probability weights is obtained. A three-level time-series carbon flow progressive deduction is completed using an energy storage memory carbon flow aggregation algorithm, restoring the cross-time carbon transfer characteristics of energy storage. A bidirectional carbon flow transfer matrix is ​​built to complete the tracking of carbon flow across the entire system and construct a complete carbon flow indicator system. Finally, a three-source tail carbon risk coupling algorithm is used to quantify the multi-scale tail risk of carbon emissions, decompose the risk sources, and output a control and assessment report.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a probabilistic carbon flow assessment method for integrated energy systems based on stochastic scenario analysis, the specific steps of which are as follows:

[0006] S1. Collect historical data of carbon capture from the source-grid-load-storage system of the distribution network and industrial park interconnection system, remove abnormal samples, extract uncertainty parameters and fit probability distribution, and construct a global random variable probability benchmark model.

[0007] S2, Based on the probability benchmark model, a year-day-real-time time series architecture is built. The runtime space-time coupled scenario fusion algorithm performs coupled sampling of random variables at the year, day, and real-time scales to generate random scenarios that distinguish between purchasing and selling electricity interactions. Clustering and compression are then used to obtain a set of typical scenarios with probability weights.

[0008] S3, taking the typical scenario set as input, run the energy storage memory carbon flow aggregation algorithm, and use the energy storage carbon flow memory update logic to retain the 15-minute real-time carbon flow distribution as a real-time scale carbon flow sequence, and aggregate it upwards level by level into a day-ahead scale carbon flow probability sequence and an annual scale carbon flow probability sequence.

[0009] S4, call the multi-scale carbon flow probability sequence, establish a two-way carbon flow transfer matrix between distribution network nodes and multi-energy links in the park, traverse the scenario to track carbon flow, and count the node carbon potential, branch carbon flow rate and the two-way carbon exchange volume between the network and the park. Combine the probability weight to generate a hierarchical time-series carbon flow index and a two-way carbon exchange index system.

[0010] S5. Import the hierarchical time-series carbon flow index and the two-way carbon exchange index system, run the three-source tail carbon risk coupling algorithm, perform coupled calculations on the tail carbon risk of the distribution network, park and tie line interaction, obtain the risk value under multiple time scale conditions, decompose the carbon risk sources and quantify the tail carbon emission risk in extreme scenarios, and generate a carbon risk management and assessment report.

[0011] Furthermore, the historical data collected in S1 is divided into two categories: distribution network side data and park side data. Distribution network side data includes centralized wind power daily output data, centralized photovoltaic daily output data, grid hourly total load data, regional time-of-use electricity purchase price data, regional time-of-use electricity sales price data, grid monthly marginal power supply carbon intensity data, grid main transformer maintenance plan data, and grid transmission line maintenance plan data. Park side data includes distributed photovoltaic 15-minute resolution output data, distributed wind power 15-minute resolution output data, park electricity load time-of-use energy consumption data, park heat load time-of-use energy consumption data, park cooling load time-of-use energy consumption data, park gas load time-of-use energy consumption data, energy storage time-period charging and discharging power data, carbon capture monthly carbon sequestration data, and tie line bidirectional power interaction time series data.

[0012] Furthermore, the abnormal samples removed in S1 include vacuum values ​​caused by equipment failure and shutdown, deviation values ​​from the operating range caused by line maintenance, and abnormal extreme values ​​caused by data transmission failures; the valid data after removing abnormal samples are uniformly organized into a standardized time series dataset; the uncertainty parameters extracted from the standardized time series dataset are divided into two groups: distribution network side uncertainty parameters and park side uncertainty parameters.

[0013] Furthermore, the uncertainty parameters on the distribution network side include the active power output of centralized wind power, the active power output of centralized photovoltaic power, the total active power load of the grid, the marginal carbon intensity of the grid, the time-of-use electricity purchase price, and the time-of-use electricity sales price; the uncertainty parameters on the industrial park side include the active power output of distributed wind power, the active power output of distributed photovoltaic power, the electrical load of the industrial park, the thermal load of the industrial park, the cooling load of the industrial park, the gas load of the industrial park, the charging and discharging power of energy storage, the carbon capture and sequestration power, and the net interactive power of the tie line; the parameters of new energy output are matched with a bounded probability distribution, which is a Beta distribution or a truncated normal distribution; the parameters of load and carbon intensity are matched with a normal distribution, and the parameters of electricity price are matched with a log-normal distribution; after all parameters have completed the distribution fitting, a fixed significance level is set to conduct a goodness-of-fit test, and the parameters that fail the test are re-selected with valid samples for iterative fitting, and after all parameters pass the test, a global random variable probability benchmark model is constructed.

[0014] Furthermore, the year-day-real-time time series architecture built in S2 includes three levels: annual long cycle, day-ahead mid-cycle, and 15-minute real-time short cycle. The annual long cycle uses the natural month as the statistical granularity, corresponding to the monthly fluctuation characteristics of wind and solar resources, load, carbon price, and maintenance plans throughout the year. The day-ahead mid-cycle uses 1 hour as the time node, inheriting the fluctuation constraints of the annual long cycle, corresponding to the daily energy supply and demand benchmark fluctuation dimension. The 15-minute real-time short cycle uses 15 minutes as the time node, inheriting the benchmark sample of the day-ahead mid-cycle, corresponding to the disturbance dimension of intraday irradiance fluctuation and short-term load impact. The three levels form a top-down constraint nesting relationship.

[0015] Furthermore, the random scenarios for electricity purchase and sale interaction distinguished in S2 include two categories: electricity purchase boundary scenario and electricity sale boundary scenario. The electricity purchase boundary scenario corresponds to the operating condition where the sum of the power generation of the source within the park and the discharge power of the energy storage is less than the total energy load of the park, and the park purchases electricity from the distribution network. The electricity sale boundary scenario corresponds to the operating condition where the sum of the output of the source of new energy within the park and the discharge power of the energy storage is greater than the total energy load of the park, and the park transmits electricity to the distribution network. The two types of scenarios are generated by batch sampling independently.

[0016] Furthermore, in step S2, the K-means clustering algorithm is used to perform clustering compression processing of random scenarios; the Euclidean distance between time-series data of scenarios is used as the basis for clustering, and random scenarios with similar fluctuation characteristics and interaction conditions are divided into the same cluster; the occurrence probabilities of all random scenarios within a single cluster are accumulated, and the accumulated result is used as the probability weight of the corresponding typical scenario; after the clustering process is completed, duplicate and redundant scenario samples are removed to generate a set of typical scenarios with probability weights.

[0017] Furthermore, the multi-scale carbon flow aggregation in S3 includes two progressive aggregation processes; the first aggregation process converts the 15-minute real-time carbon flow distribution into a day-ahead carbon flow distribution; the second aggregation process converts the day-ahead carbon flow distribution into an annual carbon flow distribution; the two aggregation processes synchronously perform energy storage carbon flow memory update operations throughout; the electricity carbon intensity value corresponding to the energy storage charging period is stored in the carbon memory state variable and is not included in the current carbon emission statistics; during the energy storage discharging period, the carbon intensity value stored in the carbon memory state variable is allocated to the corresponding time period carbon flow data according to the discharge power ratio; during each scale aggregation operation, the carbon memory state variable participates in the weighted calculation synchronously with the corresponding time period carbon flow data.

[0018] Furthermore, the bidirectional carbon flow transfer matrix established in S4 includes two types of carbon flow transfer calculation units; the first type of calculation unit corresponds to the carbon inflow transfer calculation logic for power transmission from the distribution network to the park; the second type of calculation unit corresponds to the carbon outflow transfer calculation logic for power transmission from the park to the distribution network; carbon flow tracing covers all levels of transmission line nodes and substation power supply nodes on the distribution network side, as well as distributed power supply nodes, multi-energy load nodes, energy storage power nodes, carbon capture power nodes, and multi-energy conversion nodes on the park side; after traversing all typical scenarios to complete carbon flow tracing, the carbon potential value of each node, the carbon flow rate value of each branch, and the bidirectional carbon exchange value between the network and the park are statistically analyzed.

[0019] Furthermore, S5 sets differentiated carbon emission risk thresholds for three time scales: annual, day-ahead, and real-time. The risk thresholds for each level are determined based on the 90%, 95%, and 99% statistical quantiles of the corresponding level's carbon flow probability distribution. The decomposed carbon risk sources include three categories: carbon risk from internal sources within the industrial park, carbon risk from purchased electricity in the distribution network, and carbon risk from bidirectional interactive fluctuations in the tie line. The generated carbon risk management and assessment report includes the risk warning level for each time series level, the contribution ratio of the three types of carbon risk sources, the quantitative value of tail carbon emission risk in extreme scenarios, and control and adjustment suggestions corresponding to different risk levels.

[0020] Compared with existing technologies, this probabilistic carbon flow assessment method for integrated energy systems based on stochastic scenario analysis has the following advantages:

[0021] I. This invention constructs a three-layer nested time-series architecture (year-day-real-time) combined with spatiotemporal coupling scenario fusion logic. It unifies and coordinates the probability distribution characteristics of multiple uncertain variables on both the distribution network and the industrial park sides. Relying on hierarchical scenario sampling and cluster compression to form a set of typical scenarios with probability weights, it solves the problem of difficulty in synchronously depicting the time-series coupling of multi-scale energy variables and the bidirectional power interaction between the grid and the industrial park. It fully covers the two types of operational boundary conditions of power purchase and sale in the industrial park, taking into account the random fluctuations of multiple dimensions such as renewable energy output, diversified loads, electricity prices, and carbon intensity. It can accurately restore the potential fluctuation patterns of energy flow and carbon transfer in the system throughout the entire time period. It unifies and standardizes the abnormal data cleaning and multi-type parameter adaptation probability distribution fitting process, realizes the quantitative verification of the distribution characteristics of all random variables, avoids the carbon flow calculation bias caused by single-scale and single-condition scenario modeling, and provides a statistically representative basic input for subsequent carbon flow tracking and risk assessment, significantly improving the objectivity and credibility of the carbon flow assessment results of the integrated energy system.

[0022] II. This invention achieves hierarchical extrapolation of three-level time-series carbon flow data through a progressive aggregation mechanism of energy storage memory carbon flow. It distinguishes between the carbon retention during charging and discharging and the carbon allocation during discharging based on dynamic update rules of energy storage charging carbon memory status, fully restoring the physical characteristics of carbon transfer across time periods of energy storage equipment. This eliminates the accounting distortion caused by neglecting the time-series migration of carbon in traditional static carbon flow models. A network-park bidirectional carbon flow transfer matrix is ​​constructed to track the carbon potential and carbon exchange volume of all nodes and branches. A standardized indicator system for hierarchical time-series carbon flow and bidirectional carbon exchange is established, clearly quantifying the scale of bidirectional carbon flow between the distribution network and the park. A three-source tail carbon risk coupling calculation logic is introduced, distinguishing three types of carbon risk sources and combining them with extreme tail carbon emission deviations to complete comprehensive risk quantification. Differentiated risk thresholds are set at different scales to output graded early warning and control schemes, achieving accurate carbon flow accounting and integrated assessment of carbon emission tail risk, supporting the collaborative low-carbon management and optimization of the park and distribution network.

[0023] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0025] Figure 1 Flowchart of the overall implementation of the probabilistic carbon flow assessment method for integrated energy systems based on stochastic scenario analysis;

[0026] Figure 2 Logical block diagram of the energy storage memory carbon flow stratified polymerization algorithm;

[0027] Figure 3 Flowchart for the construction of a two-way carbon flow tracking and hierarchical carbon index system. Detailed Implementation

[0028] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0029] Reference Figure 1 This invention proposes a probabilistic carbon flow assessment method for integrated energy systems based on stochastic scenario analysis. It adopts a data processing mechanism that integrates a three-level time-series nested architecture (year-day-real-time) and a data processing mechanism that combines global random variable probability fitting with energy storage memory carbon flow hierarchical aggregation. Under the premise of fully covering the bidirectional power interaction conditions between the distribution network and industrial parks, it can accurately solve the multi-scale carbon flow probability distribution and conduct coupled quantitative assessment of the tail risks of three types of carbon sources: parks, distribution networks, and interconnection lines. It has the advantages of multi-time scale adaptation, complete tracking of bidirectional carbon flow between the grid and parks, and quantitative management of extreme carbon emission risks.

[0030] The method described in this embodiment specifically includes:

[0031] Collect multi-dimensional historical operation data of the source-grid-load-storage-carbon capture system of the power distribution network and industrial park interconnection system, remove various abnormal operation samples, extract all uncertain parameters of the system after standardization, match the corresponding probability distribution function for different types of parameters and complete the goodness of fit verification, and build a global random variable probability benchmark model.

[0032] Based on the global random variable probability benchmark model, a real-time three-level time series nested architecture of annual-day-15-minute is built. The runtime space-time coupling scenario fusion algorithm completes the joint coupling sampling of random variables on both sides of the three-level time series scale and distribution network-park. It distinguishes between the two types of boundary operation conditions of park power purchase and power sale to generate a large number of original random scenarios in batches. The K-means clustering algorithm is used to compress and reduce the dimensionality of the original scenario set to generate a typical scenario set with independent probability weights.

[0033] Using a set of typical scenarios with probability weights as the input data source, the energy storage memory carbon flow aggregation algorithm is run. Based on the recursive update logic of the energy storage carbon memory state, the 15-minute real-time carbon flow probability distribution sequence is completely retained. The two-level progressive aggregation operation is completed step by step upward according to the time sequence level, and the day-ahead and year-ahead carbon flow probability sequences are output respectively.

[0034] The three-level time-series hierarchical carbon flow probability sequence is invoked to construct a carbon flow transfer matrix adapted to the bidirectional power interaction of the network and park. The carbon flow tracking and statistics of all nodes and branches in the whole system are completed by traversing all typical scenarios. The carbon potential of nodes, carbon flow rate of branches, and bidirectional carbon exchange volume of the network and park are integrated. The hierarchical time-series carbon flow index and bidirectional carbon exchange index system are built by combining the probability weights of each scenario.

[0035] Import a complete carbon flow indicator system, run a three-source tail carbon risk coupling algorithm, configure differentiated carbon emission risk thresholds for three time scales: annual, day-ahead, and real-time, separate three types of carbon risk sources: internal sources in the park, purchased electricity from the distribution network, and interconnection line interaction, couple and solve the comprehensive carbon risk quantification index and multi-scale conditional risk value, quantify the tail excess carbon emission deviation in extreme scenarios, and output a complete carbon risk management and assessment report.

[0036] Specifically, the technical process of this invention is based on a probabilistic carbon flow assessment method for integrated energy systems using random scenario analysis. This method first decouples all random disturbance factors in the distribution network-industrial park interconnected integrated energy system by parameter decoupling. Through categorized probability distribution fitting and significance testing, a global probabilistic benchmark model capable of fully reproducing the system's random fluctuation patterns is constructed. Relying on a three-level time-series nested architecture, top-down constraint correlations of long-cycle, medium-cycle, and short-cycle disturbances are achieved. A spatiotemporal coupled scenario fusion operator completes joint sampling of time-series scale and network-industrial park dual-side variables, distinguishing between power purchase and sales operation conditions to generate original random scenarios covering all operational possibilities. Then, redundant scenarios are eliminated through clustering compression, and a single scenario probability is assigned, reducing the computational cost of subsequent carbon flow calculations. The energy storage memory carbon flow aggregation algorithm is specifically designed with recursive update logic tailored to the carbon intensity transfer characteristics during energy storage charging and discharging. During the charging phase, carbon intensity is stored in memory and does not participate in current carbon emission statistics. During the discharging phase, the carbon memory storage is allocated to the corresponding time-period carbon flow according to the power ratio. Layered aggregation synchronously carries carbon memory state variables to participate in weighted calculations, ensuring physical consistency of carbon flow data at different time scales. The bidirectional carbon flow transmission matrix employs two independent calculation logics: one for power transmission from the distribution network to the industrial park, and the other for power transmission from the industrial park to the distribution network. This covers all equipment nodes in the distribution network, substations, distributed power sources within the industrial park, multi-energy loads, energy storage, and carbon capture, enabling carbon source tracing. Finally, a three-source tail-end carbon risk coupling algorithm, combined with a three-level time-series carbon flow probability distribution, dynamically allocates the contribution weights of the three types of carbon risks. This is then overlaid with a tail-end carbon emission correction term for extreme scenarios to obtain a comprehensive carbon risk value for the entire domain. Simultaneously, the contribution ratio of each risk source is broken down, and tiered early warning and control optimization suggestions are output.

[0037] The process of collecting historical carbon capture data from the interconnected power distribution network and industrial park systems, removing outliers, extracting uncertainty parameters, fitting probability distributions, and constructing a global random variable probability benchmark model includes: dividing the data collection boundary into two categories: the power distribution network side and the industrial park side, and collecting full-volume time-series operational data for each corresponding dimension; identifying and removing three types of outliers: equipment downtime vacuum values, deviation values ​​during line maintenance intervals, and abnormal extreme values ​​of transmission faults; standardizing the remaining valid data into a uniform time-series dataset with equal time steps; separating the two sets of uncertainty parameters from the standardized time-series dataset: the power distribution network side and the industrial park side; matching specific probability distribution functions for different categories of parameters, performing goodness-of-fit tests at a fixed significance level after completing the distribution fitting, re-selecting valid samples for parameters that fail the test, iteratively fitting them, and integrating all parameters after successful verification to construct a global random variable probability benchmark model.

[0038] Specifically, the data collection scope is first defined. Data collected from the distribution network includes centralized wind power daily output, centralized photovoltaic daily output, total grid load per hour, regional time-of-use electricity purchase price, regional time-of-use electricity sales price, monthly marginal carbon intensity of grid supply, grid main transformer maintenance plan, and transmission line maintenance plan. Data collected from the industrial park includes 15-minute resolution distributed photovoltaic output, distributed wind power output, time-of-use electricity / heat / cooling / gas load of the park, time-of-use charging and discharging power of energy storage, monthly carbon sequestration amount, and bidirectional power interaction time-series data from tie lines. The data collection process relies on the SCADA energy monitoring system, the park's integrated energy management platform, and the local recording unit of the carbon capture device for synchronous collection, with unified timestamp alignment to eliminate time-series offsets.

[0039] Secondly, anomaly sample cleaning was carried out, identifying three types of invalid data: numerical gaps caused by the failure and shutdown of wind turbines, photovoltaics, and energy storage equipment; power data that significantly deviates from the normal operating power range during the monthly maintenance window of the main power grid transformer and transmission lines; and abnormal maximum / minimum values ​​far exceeding the rated operating range of equipment caused by communication transmission interruptions and sensor failures. After identification, the corresponding entire time series sample was directly deleted without interpolation to avoid false data interfering with the accuracy of probability distribution fitting. After cleaning, all data were uniformly converted into a standardized dimensionless time series dataset with a unified time step unit: monthly data was normalized to a monthly granularity, hourly data to a 1-hour granularity, and short-cycle data within the park was uniformly normalized to a 15-minute granularity.

[0040] Then, two sets of uncertainty parameters were split from the standardized time-series dataset: distribution network side uncertainty parameters: centralized wind power active power output, centralized photovoltaic active power output, total grid active power load, grid marginal carbon intensity, time-of-use electricity purchase price, and time-of-use electricity sales price; and industrial park side uncertainty parameters: distributed wind power active power output, distributed photovoltaic active power output, industrial park electrical load, industrial park thermal load, industrial park cooling load, industrial park gas load, energy storage charging and discharging power, carbon capture and carbon sequestration power, and tie-line net interaction power.

[0041] For different parameters and physical characteristics, corresponding distribution models are matched: parameters for new energy output have upper and lower bound constraints, and are uniformly matched with a Beta distribution or a truncated normal distribution; various types of cooling, heating, and electrical loads, and the marginal carbon intensity of the power grid have no hard upper and lower boundaries, and are matched with a normal distribution; time-of-use electricity prices have right-skewed long-tail characteristics, and are matched with a log-normal distribution. For each set of parameters, maximum likelihood estimation is used to solve for the distribution characteristic parameters (mean, variance, shape parameters, scale parameters, etc.), and a chi-square goodness-of-fit test is conducted with a significance level of 0.05. If the test result of a certain parameter is lower than the significance threshold, it is determined that the current sample set cannot match the corresponding distribution, the data time range is narrowed again to select valid samples with stable operating conditions, and the fitting and testing process is repeated; after all parameters pass the goodness-of-fit test, all parameter distribution functions, characteristic parameters, and constraint thresholds are integrated and encapsulated to form a global random variable probability benchmark model.

[0042] For example, in constructing a global random variable probability benchmark model, complete time-series data of the 110kV distribution network of an industrial park throughout the year were selected for processing. The original dataset contained 105,120 samples with a 15-minute granularity. After an anomaly removal process, 2,136 samples of equipment downtime null values, 1,542 samples of line maintenance deviations, and 897 samples of sensor anomalies were identified, resulting in the removal of 4,575 invalid samples, leaving 100,545 valid standardized time-series data. From this, three typical uncertainty parameters—distributed photovoltaic power output, park electricity load, and time-of-use electricity price—were fitted with their respective distributions: Distributed photovoltaic power output is a bounded output variable, and a Beta distribution was used for fitting. The chi-square test showed a p-value of 0.072, which is greater than the significance threshold of 0.05, indicating a valid fit; the park electricity load was fitted with a normal distribution, with a p-value of 0.091, passing the validation; and the time-of-use electricity price was fitted with a log-normal distribution, with a p-value of 0.068, meeting the validation requirements. After all 21 uncertainty parameters have been verified for goodness of fit, all distribution functions and feature parameters are integrated to generate a global random variable probability benchmark model, which serves as the input basis for subsequent scene fusion algorithms.

[0043] The process of building a year-day-real-time time-series architecture based on the aforementioned probability benchmark model, and using a spatiotemporal coupled scenario fusion algorithm to couple and sample random variables at the year, day, and real-time scales to generate random scenarios that distinguish between power purchase and sales interactions, includes: building a three-layer top-down constrained nested time-series architecture; a long-term annual period with natural months as the statistical granularity, bearing the monthly fluctuation constraints of wind and solar power, load, carbon price, and maintenance throughout the year; a mid-term day-ahead period with 1 hour as the time node, bearing the monthly fluctuation boundary, and characterizing the daily energy supply and demand benchmark fluctuation; and a short-term 15-minute real-time period with 15 minutes as the node, bearing the day-ahead benchmark sample, and characterizing intraday irradiance and short-term load disturbances; calling the mathematical expression of the spatiotemporal coupled scenario fusion algorithm to complete the joint orthogonal fusion of the three-layer time-series variables and the distribution network-park dual-side variables, and batch generating the global joint random scenario probability domain; dividing the power purchase boundary and power sales boundary into two types of operating conditions based on the power supply and demand balance relationship of the park, independently completing batch sampling of the two types of operating conditions, and generating the original random scenario set.

[0044] The mathematical expression for the spatiotemporal coupled scene fusion algorithm is:

[0045]

[0046] Time-scale adaptive association weights Network interaction coupling strength coefficient The calculation expressions are as follows: , In the formula The standard deviation of the joint fluctuation of the three-level time series random variables. The standard deviation of the total fluctuation of all uncertain parameters in the system. This refers to the time-series correlation coefficient between the power on the distribution network side and the power exchanged between the distribution network side and the park side. This represents the maximum correlation coefficient of power between both sides when the system tie line is operating at full capacity. Represents the probability domain of a global joint random machine scenario; This represents a time-series nested transitive mapping operator; Represents time-scale adaptive association weights; This represents the set of marginal probability distributions of three levels of time-series random variables: annual, day-ahead, and real-time. The subscripts Y, D, and R correspond to the long-term annual period, the mid-term day-ahead period, and the short-term 15-minute real-time period, respectively. Represents a bidirectional interactive coupling mapping operator for the network. The coefficient representing the interaction coupling strength of the network; This represents the clustered distribution of random variables on both the distribution network side and the industrial park side, with the subscript G corresponding to the distribution network side and P corresponding to the industrial park side. Represents the probability orthogonal fusion operator; , Taken from a global random variable probability benchmark model. This is the output of the spatiotemporal coupled scene fusion algorithm.

[0047] Specifically, the three-layer time-series architecture implements hierarchical constraint propagation: the annual layer outputs monthly wind and solar power output ranges, monthly average loads, and monthly maintenance windows, serving as the upper and lower boundaries for day-ahead sampling; the day-ahead layer outputs hourly baseline loads and average renewable energy outputs, serving as the 15-minute disturbance sampling center value for the real-time layer; the real-time layer superimposes short-term random disturbances on the day-ahead baseline value to fully reproduce the system's intraday high-frequency fluctuations. A nested time-series propagation mapping operator is used. Adaptive association weights based on time-series scale The greater the constraint of the long cycle on the short cycle, the greater the monthly fluctuation in wind and light intensity. The higher the value, the greater the fluctuation range of short-period sampling. (Network-based bidirectional interactive coupling mapping operator) Based on the coupling strength coefficient The correlation between the distribution network and random variables on both sides of the industrial park is regulated. The larger the capacity of the industrial park's interconnection line, the more frequent the power interaction between the grid and the park. Higher values ​​result in a simultaneous increase in the correlation of the two-sided variable sampling. The probabilistic orthogonal fusion operator orthogonally superimposes two sets of independent probability distributions in the time-series dimension and the network-circle two-sided dimension, outputting a global joint random scene probability domain that includes all combinations of random perturbations. .

[0048] For example, a three-layer nested time-series architecture is constructed. The annual layer is divided into 12 calendar months, outputting the monthly average photovoltaic irradiance, wind power output range, and monthly maintenance plan. The day-ahead layer constructs a daily benchmark supply and demand curve using 24 1-hour nodes. The real-time layer is divided into 96 15-minute short-cycle nodes. Adaptive correlation weights are set for the time-series scale. Network interaction coupling strength coefficient Retrieve from the global probability baseline model , Substituting the distributed parameters into the expression of the spatiotemporal coupled field fusion algorithm, the probability domain of the global joint random scene is obtained. .based on Batch sampling was conducted, with 10,000 original scenarios independently sampled for electricity purchase and 8,000 original scenarios independently sampled for electricity sales, generating a total of 18,000 original random scenario sets covering all interaction scenarios.

[0049] The clustering compression process to obtain a set of typical scenarios with probability weights includes: extracting all original random scenario time-series data, calculating the time-series Euclidean distance between any two sets of scenarios as the basis for clustering; running the K-means clustering algorithm to divide the original scenarios with similar fluctuation characteristics and power purchase and sale conditions into the same cluster; accumulating the independent occurrence probabilities of all original scenarios within a single cluster, and using the accumulated result as the probability weight of the typical scenario corresponding to that cluster; after clustering, removing duplicate and highly similar redundant scenario samples, and outputting a set of typical scenarios with specific probability weights.

[0050] The core mathematical expressions of the K-means clustering algorithm are divided into two categories: sample distance calculation formula and cluster center iterative update formula. The Euclidean distance calculation expression for time-series scenarios is as follows: The cluster center iterative update expression is: In the formula For the m-th original scene With the nth original scene The temporal Euclidean distance between them, where T is the total dimension of a single scene temporal feature. , These are the t-th dimension temporal feature values ​​for the two scenarios, respectively. Let be the cluster center vector of the k-th cluster after the (i+1)-th iteration. This is the set of all original scenes partitioned into the k-th cluster during the i-th iteration. The total number of original scene samples in the set is represented by the iteration convergence criterion. In the formula, K is the total number of preset clusters. Represents the vector L2 norm. This is the preset cluster center convergence threshold.

[0051] Specifically, for each original random scenario, a multi-dimensional time-series feature vector is constructed, including annual and monthly indicators, day-ahead hourly power, 15-minute real-time output, tie-line interactive power, electricity price, and carbon intensity. The Euclidean distance between two scenario feature vectors quantifies their fluctuation similarity; the smaller the distance, the closer the scenario's operating state. A preset number of clusters is used, and the K-means algorithm iteratively updates the cluster centers. The iteration terminates when the cluster center offset is less than a preset convergence threshold. After convergence, each cluster contains a set of original scenario samples with similar operating conditions. The sampling probability values ​​of all original scenarios within the cluster are summed, serving as the probability weight for the unique typical scenario of that cluster, with the total weight remaining at 1. After clustering, only the cluster center vectors are retained as typical scenarios, and the remaining redundant original scenarios within the cluster are deleted, significantly reducing the computational scale of subsequent carbon flow tracing.

[0052] For example, for the aforementioned 18,000 original random scenarios, a 288-dimensional time-series feature vector was constructed. A clustering convergence threshold of 0.001 was set, and the preset number of clusters was 120. K-means clustering iterative computation was initiated, and after 36 iterations, all cluster centers converged. All original scenarios were divided into 120 independent clusters. The probability of samples within each cluster was accumulated to calculate the weight: for example, cluster 12 contains 146 original scenarios related to electricity purchase, with a total probability accumulation of 0.0086, corresponding to a typical scenario probability weight of 0.0086; cluster 68 contains 92 original scenarios related to electricity sales, with a total probability accumulation of 0.0053, corresponding to a typical scenario weight of 0.0053. All 120 typical scenarios and their corresponding probability weights together constitute a typical scenario set, reducing the number of scenarios from 18,000 to 120, and reducing the computational load by more than 99%.

[0053] The energy storage memory carbon flow aggregation algorithm utilizes the energy storage carbon flow memory update logic to retain the 15-minute real-time carbon flow distribution as a real-time scale carbon flow sequence, and aggregates it level by level into a day-ahead scale carbon flow probability sequence and an annual scale carbon flow probability sequence. The algorithm includes two sets of mathematical expressions: an energy storage memory carbon flow aggregation equation and a time-series carbon flow aggregation output equation. It completes the scale conversion according to a two-layer progressive aggregation process: the first layer converts the 15-minute real-time carbon flow distribution to a day-ahead scale carbon flow distribution, and the second layer converts the day-ahead scale carbon flow distribution to an annual scale carbon flow distribution. The two-layer aggregation operation is performed synchronously throughout the entire process, updating the energy storage carbon flow memory. During the energy storage charging period, the electricity carbon intensity is stored in the carbon memory state variable and is not included in the current carbon emission statistics. During the energy storage discharging period, the carbon memory storage is allocated to the corresponding time period carbon flow according to the discharge power ratio, and the carbon release attenuation coefficient is included in the aggregated carbon flow matrix. The carbon memory state variable is synchronously incorporated into the calculation at each scale weighting operation.

[0054] Recurrence relation for carbon memory state of energy storage:

[0055]

[0056] Time-sequential carbon stream polymerization output equation:

[0057]

[0058] Energy storage carbon release attenuation coefficient The calculation expression is as follows In the formula The total rated discharge capacity of the energy storage is [missing information]. The total rated charging capacity of the energy storage is [missing information]. This is a correction factor for carbon transfer loss in energy storage. Represents the three-level temporal stratification probability carbon flow distribution matrix; Represents an adaptive scaling aggregation function; The global joint random scene probability domain output by the spatiotemporal coupled scene fusion algorithm; Represents the cumulative carbon memory stock of energy storage after the current time-series update; This represents the accumulated carbon memory stock of energy storage retained at the previous time-series scale. This represents the amount of carbon memory newly stored during the current time-series energy storage charging period; This represents the amount of carbon transferred during the energy storage discharge period at the current timescale. Represents the carbon release decay coefficient of energy storage; during the energy storage charging phase, carbon memory only stores carbon and does not participate in carbon emission decay. Summation, the carbon memory allocation during the discharge phase is achieved through... The item is included in the stratified carbon flow matrix. This is the set of multi-scale carbon flow probability sequences that are ultimately output by this algorithm.

[0059] Specifically, the first layer of aggregation is based on a 15-minute real-time carbon flow sequence, and is performed using an adaptive scaling aggregation function. The carbon flows from 96 15-minute time periods on a single day are weighted and merged into 24 1-hour day-ahead carbon flow distributions. The second layer of aggregation merges the day-ahead carbon flow sequences for all 365 days of the year by monthly weighting, generating an annual monthly granular carbon flow probability distribution. Throughout the aggregation process, the energy storage carbon memory state variables are updated synchronously: when the scenario time series is during an energy storage charging period, the carbon intensity value of the electricity purchased during that period is superimposed and stored. It only updates the memory storage capacity and does not increase current carbon emissions; when the time sequence enters the discharge period, it is divided according to the current discharge power as a proportion of the total energy storage capacity. get Multiply by the carbon release attenuation coefficient Correct for carbon loss in energy storage and overlay it onto the corresponding time period's carbon flow data. Carbon memory stock obtained from aggregation at the next higher scale. As the initial value for this level of recursion, it ensures the continuous traceability of carbon memory data at the annual, current-day, and real-time levels.

[0060] Tracing back.

[0061] For example, real-time time-series data of a single typical scenario is selected to perform energy storage carbon memory update and scale aggregation calculation, and an energy storage carbon release decay coefficient is set. In the 12th real-time period, the energy storage was in a charging state, with a carbon intensity of 0.52t. MWh, charging power 2MN, and added carbon memory capacity. carbon memory stock at the previous moment Substituting into the recursive equation, we obtain the updated carbon memory inventory. The charging phase is not included in the current carbon flow statistics. In the 48th real-time period, the energy storage discharged with a power of 1.5MW, releasing carbon transfers of [amount missing]. Substituting into the aggregation output equation, after processing the basic carbon flow through an adaptive scaling aggregation function, the mixture is then superimposed. Carbon emission is calculated to obtain complete real-time carbon flow values ​​for that period. The 96 daily 15-minute carbon flow data points are weighted and aggregated to generate a 24-point day-ahead carbon flow probability distribution. Then, the daily day-ahead carbon flows for the entire year are aggregated monthly to output a 12-point annual carbon flow probability sequence, thus obtaining a complete three-level hierarchical carbon flow distribution matrix. ,like Figure 2 As shown.

[0062] The process of establishing a bidirectional carbon flow transfer matrix between distribution network nodes and multi-energy links in the industrial park, traversing scenarios to track carbon flows, and statistically analyzing node carbon potential, branch carbon flow rate, and bidirectional carbon exchange between the network and the park, combined with probability weights to generate a hierarchical time-series carbon flow index and a bidirectional carbon exchange index system, includes: constructing two types of carbon flow transfer calculation units to form a bidirectional carbon flow transfer matrix; the first type of unit is adapted to the carbon inflow calculation logic for power transmission from the distribution network to the industrial park, and the second type of unit is adapted to the carbon outflow calculation logic for power transmission from the industrial park to the distribution network; carbon flow tracking covers distribution network transmission lines and substation power supply nodes. The system covers all equipment nodes in the park, including distributed power sources, cooling and heating gas loads, energy storage, carbon capture, and multi-energy conversion. It traverses all weighted typical scenarios, performing carbon flow source calculations node by node and branch by branch. It statistically analyzes the carbon potential of each node, the carbon flow rate of each branch, and the total amount of bidirectional carbon exchange between the park and the distribution network in each scenario. The statistical results of each scenario are multiplied by the probability weight of the corresponding typical scenario, and weighted and accumulated to generate three-level time-series carbon flow indicators for annual, day-ahead, and real-time periods, as well as quantitative indicators for bidirectional carbon exchange between the park's electricity purchase carbon inflow and electricity sales carbon outflow, integrating them into a complete indicator system.

[0063] Specifically, the bidirectional carbon flow transfer matrix distinguishes between two independent carbon intensity transfer rules based on power flow direction: when power is supplied from the distribution network to the park, carbon flow flows unidirectionally from the grid-side substations along the tie lines into the park, and carbon potential is transferred step by step from the marginal carbon intensity of the grid to each load node in the park; when surplus power from renewable energy and energy storage generation in the park is supplied to the distribution network, carbon flow flows outward from the distributed power nodes in the park along the tie lines, and the carbon intensity from the internal sources in the park is transferred in the opposite direction to the lower-level branches of the distribution network. The carbon sequestration device directly deducts the current carbon emissions of the corresponding node, and the carbon flow value is deducted synchronously during the carbon flow tracking process. After completing the carbon flow calculation for the entire system by traversing all typical scenarios, the scenario probability weight is used as the weighting coefficient to weight the carbon flow statistics of all scenarios under the same time series, eliminating the interference of random fluctuations in a single scenario, and obtaining a hierarchical carbon flow index with probabilistic statistical significance; at the same time, the monthly, daily, and 15-minute total of bidirectional carbon exchange along the tie lines are separately calculated to form a grid-park bidirectional carbon exchange index. The two types of indicators are combined to support subsequent carbon risk solutions.

[0064] For example, carbon flow tracking was performed sequentially on 120 typical scenarios with probability weights. Taking the typical scenario of electricity purchase conditions with a weight of 0.0086 as an example: In this scenario, the distribution network transmits a total of 12.6 MWh of electricity to the park via the tie line, the average carbon intensity of the grid is 0.48 tCO2 / MWh, the total carbon inflow of the tie line is 6.048 tCO2, the carbon potential range of each electricity and heat load node in the park is 0.39-0.48 tCO2 / MWh, and the carbon sequestration of the park's carbon capture device in the current month is 12.5 tCO2, which offsets the total carbon emissions of the system. The total carbon flow statistics of this scenario are multiplied by the weight of 0.0086 to obtain the weighted carbon flow contribution value; the weighted calculation is repeated for the remaining 119 scenarios, and after all are accumulated, the real-time average node carbon potential, branch carbon flow rate, and monthly grid-park carbon exchange total are obtained. These are then organized into a three-layered time-series carbon flow index set (annual, day-ahead, and real-time) and a two-way carbon exchange index set, as shown below. Figure 3 As shown.

[0065] The algorithm for coupling carbon risk at the tail of the three sources is used to perform coupled calculations on the tail risks of carbon sources interacting with the distribution network, industrial park, and interconnection lines. It calculates risk values ​​under multiple time scales, breaks down carbon risk sources, quantifies tail carbon emission risks in extreme scenarios, and generates a carbon risk management and assessment report. This report includes: setting differentiated carbon emission risk thresholds based on the statistical quantiles of the probability distribution of carbon flows at each of the annual, day-ahead, and real-time time-series scales; retrieving stratified time-series carbon flow indicators; and calculating the benchmark value of fluctuating carbon risk within the industrial park. Benchmark value for carbon risk of power fluctuations from power distribution network purchases Bidirectional interactive fluctuations coupled with carbon risk benchmark value of the connecting line Run the three-source tail carbon risk coupling algorithm to dynamically solve the normalized contribution weights of the three types of risks. , , By superimposing adaptive correction coefficients for extreme scenarios and tail-end excess carbon emission deviations, a comprehensive carbon risk quantification index for the entire domain is obtained. Based on the comprehensive carbon risk value and risk thresholds at all levels, the risk warning levels are divided into multiple time series levels; the contribution ratio of three types of carbon risk sources is broken down, and the tail excess carbon emission value in extreme scenarios is quantified; all calculation results are integrated to generate a carbon risk management and control assessment report, with system regulation and adjustment suggestions corresponding to different risk levels.

[0066] The mathematical expression for the three-source tail carbon risk coupling algorithm is:

[0067]

[0068] Dynamic contribution weights of three types of risks , , The calculation expressions are as follows: , , Adaptive correction coefficient for extreme scenarios The calculation expression is Excess carbon emissions at the tail end In the formula , , These represent the variances of three carbon risk benchmarks: internal sources within the park, external purchases from the distribution network, and interaction between the grid and the park. The conditional value of risk at the 99th percentile of the carbon flow distribution. Value at risk corresponding to the 99th percentile of the carbon flow distribution. This represents the average carbon emissions in extreme scenarios where carbon emissions exceed the 99th percentile threshold. , , These represent the dynamic contribution weights of three types of risks: risks originating within the park, risks purchased from the distribution network, and risks interacting with the park, respectively, satisfying normalization constraints. ; The representative indicator of comprehensive carbon risk quantification for the entire distribution network-industrial park interconnection system; The benchmark value representing the carbon risk corresponding to the fluctuations in carbon sources within the park; The carbon risk benchmark value represents the fluctuation of purchased electricity in the distribution network; The baseline value for coupled carbon risk corresponding to the bidirectional power interaction fluctuations in the network park; Represents the adaptive correction coefficient for extreme scenarios; This represents the excess carbon emission deviation at the adaptive tail.

[0069] Specifically, the three-tiered time-series risk thresholds are set according to the 90th, 95th, and 99th percentiles of the carbon flow distribution, with the 99th percentile corresponding to the extreme tail risk trigger threshold. Based on the tiered carbon flow probability distribution, various risk benchmark values ​​are statistically analyzed: Intra-park risk arises from distributed wind and solar power output and random fluctuations in park heating and cooling loads; external purchase risk in the distribution network is caused by fluctuations in marginal carbon intensity of the grid and time-of-use electricity prices; and interconnection risk is generated by the coupling of random fluctuations in bidirectional power between the grid and the park. Normalized weights are dynamically assigned based on the volatility variance of each risk benchmark value. , , The larger the variance of a certain type of risk fluctuation, the higher its corresponding weight. Tail-end excess carbon emission deviation. The difference between the mean and threshold carbon emissions from extreme scenarios exceeding the 99th percentile of the carbon flow distribution is used, along with an adaptive correction coefficient. To correct the underestimation of risk under extreme operating conditions, a comprehensive carbon risk index for the entire region was finally obtained. .contrast The system classifies risks into three levels: Level 1, Level 2, and Level 3, based on risk thresholds at each level; statistics are compiled. , , The numerical proportions clearly identify the main sources of carbon emission risk in the system; the quantitative figures for excess carbon emissions at the tail end of extreme scenarios are summarized, and corresponding control measures are formulated for different warning levels: for low-risk levels, the charging and discharging sequence of energy storage in the park is optimized to smooth internal source fluctuations; for medium-risk levels, the output forecast of distributed renewable energy in the park is adjusted and the power purchase and sale period is optimized; for high extreme risk levels, the carbon capture operation power is increased and additional green electricity purchase quotas are allocated. All the above data are uniformly compiled to form a standardized carbon risk management and assessment report.

[0070] For example, three types of risk benchmark values ​​are solved based on the three-level time-series carbon flow probability distribution matrix: the intra-park carbon risk benchmark value. Carbon risk benchmark for purchased electricity in power distribution network Inter-coupling carbon risk benchmark value ; Dynamically assign normalized weights based on the variance of three types of risk , , The condition 0.28 + 0.52 + 0.20 = 1 is satisfied; extreme scenarios outside the 99th quantile of the carbon flow distribution are extracted, and the tail excess carbon emission deviation is calculated. Adaptive correction coefficient for extreme scenarios Substituting the values ​​into the coupled algorithm expression, the overall carbon risk index is calculated: ,Will Compared with the 99th percentile risk threshold of 20 at the real-time scale, a level-two risk warning was triggered. The weighted values ​​indicate that fluctuations in purchased electricity from the distribution network are the primary source of carbon risk in the system, contributing 52%. In extreme scenarios, the average tail-end excess carbon emissions are 7.2 tons. The carbon risk management assessment report marks a real-time time-series level-two early warning, identifies the carbon fluctuation of purchased electricity as the main risk source, and provides corresponding control recommendations: increase the proportion of daytime energy storage discharge, purchase low-carbon electricity in batches during off-peak hours, and increase the average daily operating time of carbon capture equipment, thus completing the entire assessment output.

[0071] The method also includes: extracting risk warning levels at each time-series level, the contribution ratio of the three types of carbon risk sources, and quantitative values ​​of extreme tail carbon emissions from the carbon risk management and assessment report; in order to reduce overall carbon risk across the entire region. To optimize objectives, the park's energy storage charging and discharging strategy, distributed new energy consumption scheme, carbon capture operation power, and grid-connected electricity purchase and sale trading plan will be adjusted. The adjusted operating parameters will be re-substituted into the global random variable probability benchmark model, and the entire process of scenario generation, carbon flow aggregation, and risk coupling will be repeated to update the overall carbon risk index. If the updated overall carbon risk value decreases significantly, the corresponding control scheme will be included as a recommended optimization strategy in the evaluation report.

[0072] Specifically, the risk data from the assessment report is read to identify the carbon risk sources contributing the highest percentage, and multiple control options are generated accordingly: if the risk from external carbon purchases from the distribution network is the highest, priority is given to optimizing time-of-use electricity purchase periods, increasing the amount of electricity purchased during periods of low grid intensity, and relying on energy storage discharge for self-sufficiency during periods of high carbon emissions; if the risk from internal sources within the industrial park is the highest, the intraday charging and discharging sequence of energy storage is optimized to smooth out fluctuations in wind and solar power output; if the risk from interconnection lines is significant, constraints on power fluctuations in interconnection lines are added to limit short-term high-power bidirectional power transmission. The load, energy storage, and electricity purchase parameters corresponding to each control option are updated to the global probability benchmark model, and typical field quantity sets are regenerated and the three-level carbon flow sequence and the overall carbon risk across the entire region are recalculated. By comparing the reduction in risk indicators before and after regulation, the regulation scheme with the largest reduction in risk is selected as the optimal management and control recommendation and added to the carbon risk management and control assessment report to provide a quantitative decision-making basis for park and power grid operation and maintenance scheduling.

[0073] For example, in the aforementioned case, the carbon risk contribution of purchased electricity from the distribution network accounts for 52%, making it the primary source of system risk. Two control alternatives were generated: Option 1 adjusts the energy storage charging period to the early morning low-carbon period of the grid, allowing the energy storage to discharge at full capacity during daytime peak hours; Option 2 increases the daily operating time of the carbon capture equipment by 4 hours. The parameters of both options are updated in the global probability baseline model and recalculated. After control of Option 1, the overall carbon risk across the entire region is... It dropped to 16.21, a decrease of 30.97%; after adjustment under Plan 2. The risk level dropped to 19.75, a decrease of 15.90%. Comparing the risk reduction of the two schemes, Scheme 1 was selected as the optimal control strategy and incorporated into the carbon risk management assessment report. The optimization details of time-sharing charging and discharging of energy storage were clarified, and the entire assessment and verification process was completed.

[0074] Optionally, before collecting historical carbon capture data from the power grid and industrial park interconnection system, the system further includes: building a hardware architecture for synchronous multi-source data acquisition, deploying a power grid-side SCADA acquisition terminal, a park integrated energy management platform, a distributed wind and solar inverter data acquisition module, an energy storage BMS monitoring unit, and a local recording terminal for the carbon capture device; uniformly configuring the system master clock to synchronously trigger the data acquisition and analog-to-digital conversion processes of all terminals; setting multiple sampling granularities, with monthly and daily statistical data using hourly sampling, and real-time operating data of the power grid, load, and storage system within the park uniformly using 15-minute high-frequency sampling.

[0075] Specifically, the hardware acquisition architecture covers the power distribution network and all energy and carbon capture equipment in the park. Data transmission between terminals is achieved via industrial Ethernet, with the master clock signal synchronously sent to all acquisition units every second, eliminating timing offset errors between multiple devices. Sampling granularity is differentiated for different levels of timing requirements: annual and monthly statistical indicators use 1-hour sampling and storage of average values; distributed power sources, heating and cooling loads, energy storage power, and interconnection power within the park use 15-minute high-frequency sampling to fully capture short-term power fluctuations; carbon capture devices summarize the total carbon sequestration monthly and synchronously record monthly operating time. The synchronous acquisition mechanism ensures that when constructing standardized time-series datasets, the timestamps of all device data are strictly aligned, avoiding timing misalignment that could cause distortion in probability distribution fitting and deviations in carbon flow tracking calculations.

[0076] For example, a complete data acquisition architecture is established, with distribution network-side SCADA terminals, park energy management platforms, wind and solar inverters, energy storage BMS, and carbon capture device terminals connected to a unified industrial ring network, and data acquisition is synchronously triggered by a millisecond-level master clock. Monthly marginal carbon intensity and maintenance plans for the power grid are sampled and summarized hourly; the park's distributed photovoltaic, energy storage charging and discharging power, and bidirectional power of the tie line are sampled at 15-minute intervals, with time-series data stored continuously throughout the year; the carbon capture device summarizes carbon sequestration monthly, and monthly operation records are automatically exported on the 5th of each month. The time-series data collected throughout the year is aligned without offset, providing a complete and time-consistent raw data source for subsequent anomaly sample cleaning and probability distribution fitting.

[0077] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A probabilistic carbon flow assessment method for integrated energy systems based on stochastic scenario analysis, characterized in that, The specific steps of this method are as follows: S1. Collect historical data of carbon capture from the source-grid-load-storage system of the distribution network and industrial park interconnection system, remove abnormal samples, extract uncertainty parameters and fit probability distribution, and construct a global random variable probability benchmark model. S2, Based on the probability benchmark model, a year-day-real-time time series architecture is built. The runtime space-time coupled scenario fusion algorithm performs coupled sampling of random variables at the year, day, and real-time scales to generate random scenarios that distinguish between purchasing and selling electricity interactions. Clustering and compression are then used to obtain a set of typical scenarios with probability weights. S3, taking the typical scenario set as input, run the energy storage memory carbon flow aggregation algorithm, and use the energy storage carbon flow memory update logic to retain the 15-minute real-time carbon flow distribution as a real-time scale carbon flow sequence, and aggregate it upwards level by level into a day-ahead scale carbon flow probability sequence and an annual scale carbon flow probability sequence. S4, call the multi-scale carbon flow probability sequence, establish a two-way carbon flow transfer matrix between distribution network nodes and multi-energy links in the park, traverse the scenario to track carbon flow, and count the node carbon potential, branch carbon flow rate and the two-way carbon exchange volume between the network and the park. Combine the probability weight to generate a hierarchical time-series carbon flow index and a two-way carbon exchange index system. S5. Import the hierarchical time-series carbon flow index and the two-way carbon exchange index system, run the three-source tail carbon risk coupling algorithm, perform coupled calculations on the tail carbon risk of the distribution network, park and tie line interaction, obtain the risk value under multiple time scale conditions, decompose the carbon risk sources and quantify the tail carbon emission risk in extreme scenarios, and generate a carbon risk management and assessment report.

2. The method for probabilistic carbon flow assessment of integrated energy systems based on stochastic scenario analysis according to claim 1, characterized in that, The historical data collected in S1 is divided into two categories: distribution network side data and park side data. Distribution network side data includes centralized wind power daily output data, centralized photovoltaic daily output data, grid hourly total load data, regional time-of-use electricity purchase price data, regional time-of-use electricity sales price data, grid monthly marginal power supply carbon intensity data, grid main transformer maintenance plan data, and grid transmission line maintenance plan data. Park side data includes distributed photovoltaic 15-minute resolution output data, distributed wind power 15-minute resolution output data, park electricity load time-of-use energy consumption data, park heat load time-of-use energy consumption data, park cooling load time-of-use energy consumption data, park gas load time-of-use energy consumption data, energy storage time-period charging and discharging power data, carbon capture monthly carbon sequestration data, and tie line bidirectional power interaction time series data.

3. The method for probabilistic carbon flow assessment of integrated energy systems based on stochastic scenario analysis according to claim 1, characterized in that, The abnormal samples removed in S1 include vacuum values ​​caused by equipment failure and shutdown, deviations from the operating range caused by line maintenance, and abnormal extreme values ​​caused by data transmission failures. After the abnormal samples were removed, the valid data were uniformly organized into a standardized time series dataset. The uncertainty parameters extracted from the standardized time series dataset were divided into two groups: distribution network uncertainty parameters and park uncertainty parameters.

4. The method for probabilistic carbon flow assessment of integrated energy systems based on stochastic scenario analysis according to claim 3, characterized in that, The uncertainty parameters on the distribution network side include the active power output of centralized wind power, the active power output of centralized photovoltaic power, the total active power load of the grid, the marginal carbon intensity of the grid, the time-of-use electricity purchase price, and the time-of-use electricity sales price. The uncertainty parameters on the industrial park side include the active power output of distributed wind power, the active power output of distributed photovoltaic power, the park's electrical load, the park's thermal load, the park's cooling load, the park's gas load, the energy storage charging and discharging power, the carbon capture and sequestration power, and the net interactive power of the tie line. The parameters of new energy output are matched with a bounded probability distribution, which is either a Beta distribution or a truncated normal distribution. The parameters of load and carbon intensity are matched with a normal distribution, and the parameters of electricity price are matched with a log-normal distribution. After all parameters have completed the distribution fitting, a fixed significance level is set to conduct a goodness-of-fit test. Parameters that fail the test are re-selected with valid samples for iterative fitting. After all parameters pass the test, a global random variable probability benchmark model is constructed.

5. The method for probabilistic carbon flow assessment of integrated energy systems based on stochastic scenario analysis according to claim 1, characterized in that, The year-day-real-time time series architecture built in S2 includes three levels: annual long cycle, day-ahead medium cycle, and 15-minute real-time short cycle. The annual long cycle uses the natural month as the statistical granularity, corresponding to the monthly fluctuation characteristics of wind and solar resources, load, carbon price, and maintenance plan throughout the year. The recent medium-term cycle uses 1 hour as a time node to bear the fluctuation constraints of the long-term annual cycle, corresponding to the daily energy supply and demand benchmark fluctuation dimension; The 15-minute real-time short cycle uses 15 minutes as a time node to serve as the benchmark sample for the daytime mid-cycle, corresponding to the disturbance dimensions of intraday irradiance fluctuations and short-term load impacts. The three levels form a top-down nested constraint relationship.

6. The method for probabilistic carbon flow assessment of integrated energy systems based on stochastic scenario analysis according to claim 1, characterized in that, The random scenarios for electricity purchase and sale interaction distinguished in S2 include two categories: electricity purchase boundary scenario and electricity sale boundary scenario. The electricity purchase boundary scenario corresponds to the operating condition where the sum of the power generation of the source within the park and the discharge power of the energy storage is less than the total energy load of the park, and the park purchases electricity from the distribution network. The electricity sale boundary scenario corresponds to the operating condition where the sum of the output of the source of new energy within the park and the discharge power of the energy storage is greater than the total energy load of the park, and the park transmits electricity to the distribution network. Batch sampling generation is completed independently for both types of scenarios.

7. The method for probabilistic carbon flow assessment of integrated energy systems based on stochastic scenario analysis according to claim 1, characterized in that, In S2, the K-means clustering algorithm is used to complete the clustering compression processing of random scenarios; Using the Euclidean distance between time-series data of different scenarios as the basis for clustering, random scenarios with similar fluctuation characteristics and interaction conditions are grouped into the same cluster. The occurrence probabilities of all random scenarios within a single cluster are accumulated, and the accumulated result is used as the probability weight of the corresponding typical scenario. After the clustering process is completed, duplicate and redundant scenario samples are removed to generate a set of typical scenarios with probability weights.

8. The method for probabilistic carbon flow assessment of integrated energy systems based on stochastic scenario analysis according to claim 1, characterized in that, The multi-scale carbon flow polymerization in S3 comprises a two-layer progressive polymerization process. The first-level aggregation process converts the 15-minute real-time carbon flow distribution into a day-ahead carbon flow distribution. The second-level aggregation process converts the day-ahead carbon flow distribution into the annual carbon flow distribution. The two-layer polymerization process synchronously performs energy storage carbon flow memory update operations throughout the entire process; The carbon intensity value of electricity corresponding to the energy storage charging period is stored in the carbon memory state variable and is not included in the current carbon emission statistics. During the energy storage discharge period, the carbon intensity value stored in the carbon memory state variable is distributed to the corresponding carbon flow data according to the discharge power ratio; during each scale aggregation operation, the carbon memory state variable participates in the weighted calculation synchronously along with the corresponding carbon flow data.

9. The method for probabilistic carbon flow assessment of integrated energy systems based on stochastic scenario analysis according to claim 1, characterized in that, The bidirectional carbon flow transfer matrix established in S4 includes two types of carbon flow transfer calculation units; the first type of calculation unit corresponds to the carbon inflow transfer calculation logic for power transmission from the distribution network to the park; the second type of calculation unit corresponds to the carbon outflow transfer calculation logic for power transmission from the park to the distribution network; carbon flow tracing covers all levels of transmission line nodes and substation power supply nodes on the distribution network side, as well as distributed power supply nodes, multi-energy load nodes, energy storage power nodes, carbon capture power nodes, and multi-energy conversion nodes on the park side; after traversing all typical scenarios to complete carbon flow tracing, the carbon potential value of each node, the carbon flow rate value of each branch, and the bidirectional carbon exchange value between the network and the park are statistically analyzed.

10. The method for probabilistic carbon flow assessment of integrated energy systems based on stochastic scenario analysis according to claim 1, characterized in that, In S5, differentiated carbon emission risk thresholds are set for three time scales: annual, day-ahead, and real-time. The risk thresholds at each level are determined based on the 90%, 95%, and 99% statistical quantiles of the carbon flow probability distribution at the corresponding level. The carbon risk sources are divided into three categories: carbon risk from internal sources in the park, carbon risk from purchased electricity in the distribution network, and carbon risk from bidirectional interactive fluctuations in the tie line. The generated carbon risk management and assessment report includes risk warning levels at each time series level, contribution ratios of the three types of carbon risk sources, quantitative values ​​of tail carbon emission risks in extreme scenarios, and control and adjustment recommendations corresponding to different risk levels.

Images