A power supply method based on carbon image

Abstract

The application discloses a power supply method based on carbon portrait, and the steps of the method comprise the following: a power grid related system performs carbon emission accounting of an electricity load node; if the carbon emission accounting is passed, carbon characteristic related features of the electricity load node are obtained; according to the carbon characteristic related features, a dimension reduction data set is obtained through a classification method based on a feature space; dimension reduction feature classification is performed according to the dimension reduction data set, and a plurality of categories are obtained; carbon labels corresponding to the plurality of categories are obtained, so that a carbon portrait is obtained; and power supply is performed according to the carbon portrait of the electricity load node. Through the method, the carbon emission of the power grid in the process of supplying power to the electricity load node is reduced, and the accuracy of the power grid in selecting a power supply mode is ensured.

Description

Technical Field

[0001] This invention relates to a power supply method based on carbon profiling, and the technical field involved is the field of power grid supply. Background Technology

[0002] In current power systems, reducing carbon emissions during grid operation has become a key technological objective in order to effectively address environmental challenges and promote the sustainable transformation of the energy structure. Traditional grid dispatching and power supply models often focus on meeting users' electricity demand and ensuring the reliability and economy of power supply. They typically treat electricity loads as objects to be passively satisfied, primarily relying on load size and forecast curves to arrange generation plans and network flow. However, this approach ignores the deep-seated differences in the "carbon characteristics" exhibited by different load nodes when consuming electricity. Different types of loads have significantly different energy consumption periods, power fluctuation characteristics, sensitivity to power quality, and their own adjustability potential. These characteristics directly affect what types of generation resources the grid needs to mobilize and how much reserve capacity needs to be configured to cope with their uncertainties when supplying power. If we can gain a deeper understanding and utilize the carbon-related characteristics of these load nodes to create their "carbon profiles," the grid can make more refined and low-carbon choices in power supply decisions, especially at the levels of generation dispatching and power flow organization.

[0003] The key lies in the accurate and precise identification of this carbon profile. A carbon profile is not merely a simple label, but a comprehensive depiction of the multi-dimensional carbon-related characteristics of load nodes. The selection of power supply strategies, such as which power source combination to match, which demand response mechanism to apply, and what power quality assurance level to set, all heavily rely on the accurate interpretation of the load node carbon profile. If the carbon profile identification is flawed, then the power supply strategy formulated based on this incorrect profile will inevitably be ineffective or even harmful. Incorrect matching not only fails to achieve the expected carbon reduction effect but may also lead to the waste of renewable energy. Therefore, developing and applying methods that can accurately identify the carbon profiles of electricity load nodes, ensuring the authenticity and comprehensiveness of the profile information, is the foundation and key technical link for ensuring the correctness and effectiveness of subsequent power supply decisions based on carbon profiles. It directly relates to whether the entire low-carbon power supply strategy can be successfully implemented and whether its carbon reduction benefits can be truly realized.

[0004] In existing technologies, no technical solution for power supply based on carbon profiles of electricity load nodes has been found. Similar approaches include modeling and simulating the power supply system using a digital twin system to obtain simulated power supply data; then inputting this data into a power supply carbon emission verification model for verification; if the verification result is unqualified, activating a power supply carbon emission optimization algorithm to optimize the target power supply scheme, obtaining a carbon emission optimized power supply scheme, and controlling the power supply accordingly. For example, a power supply carbon emission regulation method and system based on optimization algorithms has been published on the China Patent Network, application number 202411181505.2. However, this patent lacks in-depth analysis of node-level carbon characteristics; it focuses on the carbon emission verification and optimization of the overall power supply scheme, with its analysis and control granularity at the "power supply scheme" level. It does not delve into the extraction and analysis of the intrinsic, multi-dimensional carbon characteristics of individual load nodes, and lacks the classification of load node carbon emission characteristics. Therefore, its regulation target is the overall power supply scheme, and it lacks the identification of node carbon profiles. This patented technology cannot directly support refined and differentiated power supply strategies based on the carbon characteristics of different nodes, and it is difficult to achieve precise green electricity matching or customized demand-side management guidance based on the specific carbon profile of a node. Thus, it cannot further reduce the carbon emissions of the power grid during the power supply process. Its optimization adjustment is systematic, rather than node-specific. Summary of the Invention

[0005] The purpose of this invention is to provide a power supply method based on carbon profiling. This method selects the power supply mode for the power load based on the carbon profile of the power load nodes in the power system, thereby reducing carbon emissions during the power supply process.

[0006] Another objective of this invention is to provide a power supply method based on carbon profiling, which improves the accuracy of the obtained carbon tags, ensures the accuracy of the obtained carbon profiles of power load nodes, and thus ensures the accuracy of selecting the appropriate power supply method for the power load nodes.

[0007] To achieve the above objectives, a carbon profile-based power supply method is provided. The method includes the following steps: calculating the carbon emissions of an electricity load node; if the carbon emission calculation is successful, obtaining the carbon characteristic-related features of the electricity load node; obtaining a dimensionality-reduced dataset based on the carbon characteristic-related features using a feature space-based classification method; performing dimensionality-reduced feature classification based on the dimensionality-reduced dataset to obtain several categories; obtaining carbon tags corresponding to the several categories to obtain a carbon profile; and supplying power based on the carbon profile of the electricity load node.

[0008] Preferably, the process of obtaining the dimensionality reduction dataset includes: obtaining the density of the neighborhood sample point set of each sample, and determining the scale parameter of the kernel function based on the density of the neighborhood sample point set; constructing the adjacency weight matrix and the relevance matrix of the feature through the kernel function; constructing the Laplacian matrix of the feature through the adjacency weight matrix and the relevance matrix, obtaining the vectors corresponding to the first few eigenvalues ​​of the Laplacian matrix, performing feature dimensionality reduction for each sample, and obtaining the dimensionality reduction dataset.

[0009] Preferably, the dimensionality reduction feature classification process includes: constructing a sparse network topology based on the dimensionality reduction dataset; dividing the sparse network topology into several sub-network topologies; merging the sub-network topologies in the several sub-network topologies using a similarity measure until the number of sub-network topologies equals the number of categories, and obtaining several categories corresponding to the several sub-network topologies.

[0010] Preferably, the process of dimensionality reduction feature classification includes: obtaining several categories from the dimensionality reduction dataset using the K-medoids algorithm.

[0011] Preferably, if the carbon emission calculation at the electricity load node fails, the process of the method is terminated.

[0012] Preferably, the carbon characteristic-related features include: coupling degree features; the coupling degree features are specifically: the proportion of nodes in the non-critical path to all nodes.

[0013] Preferably, the carbon-related characteristics include: load transfer potential; the load transfer potential is specifically the sum of the power of all transferable equipment multiplied by the start-stop energy consumption cost.

[0014] Preferably, the carbon-related characteristics include: carbon emissions; specifically, the carbon emissions generated by the electricity load node during electricity consumption.

[0015] Preferably, different carbon profiles correspond to different power supply methods; the power grid related system acquires the carbon profiles of the power load nodes in real time, and if the carbon profile of the power load node changes, the power grid related system selects a power supply method to supply power to the power load node.

[0016] Preferably, the number of carbon label types is at least 3, and the number of carbon image types is at least 3.

[0017] The beneficial effect of this invention is that the method selects different power supply methods for different types of carbon profiles of electricity load nodes, thereby reducing carbon emissions during the power supply process. The method improves the accuracy of the obtained carbon tags, thus improving the accuracy of obtaining carbon profiles of electricity load nodes and ensuring the accuracy of power supply method selection based on carbon profiles. Attached Figure Description

[0018] Figure 1 This invention relates to a flowchart of a method.

[0019] Figure 2 This is the neighborhood graph of the sample points in a certain two-dimensional feature space.

[0020] Figure 3 This invention relates to a flowchart of a method. Detailed Implementation

[0021] Example 1: This example discloses a specific implementation flow of the method involved in the present invention. (See attached document.) Figure 1 .

[0022] The specific implementation process of the method involved in this embodiment is as follows.

[0023] Obtain the carbon emissions of each energy-consuming link at the electricity load node, and calculate the carbon emissions of each energy-consuming link.

[0024] The carbon characteristics of the electrical load nodes are described below.

[0025] After carbon emission accounting, the carbon emissions of the electricity load node are specifically: the carbon emissions generated by the electricity load node during the electricity consumption process.

[0026] In this embodiment, the carbon-related features include: adjustable load ratio, which is the adjustable load divided by the total load; and response time, which is the time from receiving the instruction to the completion of load adjustment.

[0027] In this embodiment, the carbon-related characteristics include: production planning flexibility index, specifically the number of adjustable processes divided by the total number of processes and then multiplied by the time adjustment range; equipment flexibility, specifically a value of 1 divided by the minimum start-stop interval of the equipment multiplied by the start-stop energy cost (1 / (minimum start-stop interval of the equipment multiplied by the start-stop energy cost)); load transfer potential, specifically the sum of the power of all transferable equipment multiplied by the start-stop energy cost; and coupling characteristics, which involve using a graph theory model to analyze the dependencies between production processes, identifying non-critical paths, and obtaining the proportion of nodes on non-critical paths to all nodes.

[0028] In this embodiment, the carbon-related characteristics include: green electricity consumption improvement rate, specifically: (energy storage consumption of green electricity - the sum of curtailed solar and wind power when there is no energy storage) / total green electricity generation × 100%.

[0029] In this embodiment, carbon-related characteristics include: installed capacity potential assessment value, specifically usable area (m²). 2 ) × Installed capacity per unit area (kW / m 2 × Efficiency coefficient (shading, tilt correction); Predicted power generation, specifically installed capacity (kW) × local annual equivalent full-load hours × system efficiency (%); Carbon emission reduction, specifically self-generated power × grid emission factor - self-generated carbon emissions.

[0030] The waste heat and pressure recovery and utilization characteristics, in this embodiment, include: recoverable energy potential, specifically waste heat flow (kg / h) × specific heat capacity × temperature difference (°C) × operating time × recovery efficiency (%).

[0031] The characteristics of distributed microgrid construction, in this embodiment, include: microgrid capacity configuration, specifically photovoltaic / wind power installed capacity (kW) + energy storage capacity (kWh) + backup power supply (kW); energy self-sufficiency rate, specifically (microgrid self-generated power + energy storage discharge) / total plant power consumption × 100%; and operational reliability, specifically (1 - (microgrid outage time / total operating time)) × 100%.

[0032] The characteristics of the power grid interaction mode, in this embodiment, include: power trading participation rate, specifically the amount of electricity traded in the market / total electricity consumption × 100%; and demand response contribution rate, specifically the actual load reduction / committed load reduction × 100%.

[0033] It should be noted that in this embodiment, the power load node is mainly a power load node. A carbon profile is generated based on the carbon-related characteristics of the power load node, and power is supplied to the power load node according to the carbon profile.

[0034] After obtaining the carbon characteristics of each electricity load node, a spatial feature-based classification method is used to obtain several categories, thereby obtaining carbon labels corresponding to several categories.

[0035] It should be noted that the values ​​of the carbon characteristics related features of the electrical load nodes are all greater than 0.

[0036] In this embodiment, the classification method process is as follows.

[0037] Obtain the density of the neighborhood sample point set of each sample (each sample corresponds to an electrical load node), and determine the scale parameter of the kernel function based on the neighborhood density of each sample.

[0038] The process for determining the density of the neighborhood sample point set for each sample is as follows.

[0039] If there are n samples, and in this embodiment there are 17 carbon characteristic-related features, then a sample matrix of 17 rows and n columns is constructed.

[0040] After constructing the sample matrix, the entropy of each feature in the sample matrix is ​​obtained. The feature corresponding to the two largest entropies among the entropies of each feature is obtained. If the ratio of the sum of the entropies of the corresponding feature to the sum of the entropies of all features is not less than 0.8, then the feature is used as the feature of the sample. A 2xn matrix is ​​generated. The sample set of each sample in the feature space is obtained where the distance between each sample and each other is less than a first distance threshold. After obtaining the sample set of each sample, the normalized radian standard deviation of the sample set corresponding to each sample and the average Euclidean distance between each sample and the corresponding sample set are obtained. The density of the neighborhood sample point set of each sample is obtained by multiplying the normalized radian standard deviation by the reciprocal of the Euclidean distance standard deviation.

[0041] In each sample, for the i-th sample, the corresponding sample set is S. i The i-th sample and the sample set S i The expression for the normalized standard deviation of radians. Where, σ i The normalized standard deviation of radians in the sample set is expressed as follows: θ i，j θ is the radian value corresponding to the j-th sample point in the sample set corresponding to the i-th sample; i,avg Let be the average radian value in the sample set corresponding to the i-th sample.

[0042] In each sample, for the i-th sample, the corresponding sample set is S. i The i-th sample and the sample set S i The expression for the standard deviation of the Euclidean distance. Where, x i Let x be the feature vector of the i-th sample; i,j For the sample set S i The feature vector of the j-th sample; dist(x) i x i，j ) is x i With x i，j The Euclidean distance between them.

[0043] In each sample, for the i-th sample, the corresponding sample set is S. i The expression for the density of the neighborhood sample set of the i-th sample, Nd. i=1 / (σ i *AEd i The density of the neighborhood sample point set is related to the sample point and the density of sample points in the region where the sample point set is located.

[0044] If the ratio of the sum of the entropies of the corresponding features to the sum of the entropies of all features is less than 0.8, then PCA or KPCA dimensionality reduction is performed on the sample matrix to obtain the eigenvector with the largest two eigenvalues. This eigenvector is then used to reduce the dimensionality of the sample matrix to a 2xn matrix. Based on the reduced dimensionality sample matrix, the sample set corresponding to each sample is obtained. Then, the normalized radian standard deviation of the sample set corresponding to each sample and the average Euclidean distance between each sample and its corresponding sample set are obtained. The neighborhood sample point density of each sample is obtained by multiplying the normalized radian standard deviation by the reciprocal of the Euclidean distance standard deviation.

[0045] If we only take the reciprocal of the standard deviation of the Euclidean distance between each sample and its corresponding sample set as the density of the neighborhood sample point set of each sample, the high similarity among samples in the corresponding sample set may lead to a large standard deviation of the Euclidean distance between each sample and its corresponding sample set. Furthermore, there exists a situation where different samples have the same standard deviation of the Euclidean distance with their corresponding sample sets, resulting in the same density of neighborhood sample point sets for samples even when the distribution of neighborhood sample sets differs or the sample diversity within the sample sets varies. Therefore, the reciprocal of the standard deviation of the Euclidean distance cannot accurately measure the density of the neighborhood sample point set of a sample. An example will illustrate this below.

[0046] In one example of the neighborhood of a sample point in this embodiment, see [reference]. Figure 3 In a circular neighborhood of a sample's two-dimensional feature space, if all sample points in the sample set lie within an arc of 120° to 130° within this circular neighborhood, it indicates a high similarity between the sample points. Furthermore, a large variation in the Euclidean distance between these sample points and the sample itself signifies a large standard deviation of the Euclidean distance between the sample and its corresponding sample points. Changing the arc corresponding to a sample point does not alter the standard deviation of the Euclidean distance between that sample and the corresponding sample set.

[0047] Therefore, using the reciprocal of the standard deviation of the Euclidean distance between a sample and its corresponding sample cannot accurately measure the density of the neighborhood sample set of each sample point. However, the neighborhood sample set density involved in this embodiment not only considers the degree of change of the Euclidean distance between each sample in the sample set and its corresponding sample, but also considers the distribution of each sample in the sample set, making the neighborhood sample set density involved in this embodiment more accurate.

[0048] To further illustrate the neighborhood sample point set density involved in this embodiment, the neighborhood calculation method of this embodiment will be described in detail below with an example.

[0049] It should be noted that, in the set of sample points, the radian of a sample point is: the radian of the angle between the straight line connecting the sample point to the sample point in the set of sample points and the diameter of the circle with the diameter of the sample point in the two-dimensional feature space, which is the diameter of the first distance threshold circle; the sample point is not included in the set of sample points.

[0050] In one executable example of this embodiment, see [reference]. Figure 3 ,exist Figure 3 In the sample, there are sample points 1 to 8, where sample points 2 to 8 are located in the neighborhood of sample point 1, and the distance between sample points 2 to 8 and sample point 1 is less than the first distance threshold, which is 37 units.

[0051] Before changing the radians of the sample points, the Euclidean distance between sample point 1 and sample point 2 was 2 units, and the radian distance corresponding to sample point 2 was... The Euclidean distance between sample point 1 and sample point 3 is 4 units, and the radian distance corresponding to sample point 3 is... The Euclidean distance between sample point 1 and sample point 4 is 16 units, and the radian distance corresponding to sample point 4 is... The Euclidean distance between sample point 1 and sample point 5 is 7 units, and the radian distance corresponding to sample point 4 is... The Euclidean distance between sample point 1 and sample point 6 is 27 units, and the radian distance corresponding to sample point 6 is... The Euclidean distance between sample point 1 and sample point 7 is 30 units, and the radian distance corresponding to sample point 7 is... The Euclidean distance between sample point 1 and sample point 8 is 19 units, and the radian distance corresponding to sample point 8 is...

[0052] After changing the radian of the sample points (without changing the Euclidean distance between the sample points in the sample point set and the corresponding sample points in the sample point set), the radian corresponding to sample point 2 is 0.25*π; the radian corresponding to sample point 3 is... The radian value corresponding to sample point 4 is 0.5*π; the radian value corresponding to sample point 5 is 0.75*π; the radian value corresponding to sample point 6 is π; the radian value corresponding to sample point 7 is 1.25*π; and the radian value corresponding to sample point 8 is 1.5*π.

[0053] The standard deviation of the Euclidean distance is used as the density of the neighborhood sample point set of sample point 1. Before the change in the radius of the sample points in the sample set of sample point 1, the density of the neighborhood sample point set of sample point 1 is 0.97 (rounded to two decimal places). After the change in the radius of the sample points in the sample set of sample point 1, the density of the neighborhood sample point set of sample point 1 is 0.97 (rounded to two decimal places).

[0054] Using the neighborhood sample point set density involved in this embodiment, before the radian change of the sample point, the neighborhood sample point density of sample point 1 is 2.16 (rounded to two decimal places). After the radian change of the sample point in the sample set of sample point 1, the neighborhood sample point set density of sample point 1 is 0.44 (rounded to two decimal places).

[0055] from Figure 3 As can be seen, before the radian change, the density of sample point 1 and the region containing the corresponding set of sample points is high, and the density of the neighboring sample point set of node 1 is also high. After the radian change, the density of sample point 1 and the region containing the corresponding set of sample points is low, and the density of the neighboring sample point set of node 1 is also low.

[0056] The results above show that when using the Euclidean distance standard deviation as the neighborhood node set density, the neighborhood sample set density of sample point 1 remains unchanged even when the neighborhood sample set density of sample point 1 changes. When using the neighborhood sample set density change involved in this embodiment, before the radian change, the neighborhood sample set density of node 1 is relatively high, with a value of 2.16; after the radian change, the neighborhood sample set density of node 1 is relatively low, with a value of 0.44, which better matches the distribution of the sample point set of node 1.

[0057] After determining the density of the neighborhood sample point set of each sample point, the scale parameter of the kernel function is determined based on the density of the neighborhood sample point set of each sample point. The distance metric between each sample point is determined based on the kernel function. The fully connected function of the sample matrix is ​​constructed based on the distance metric.

[0058] In this embodiment, the kernel function is the Gaussian similarity distance kernel function, and its expression is as follows.

[0059] The expression for the Gaussian similarity distance kernel function is: Where, γ i With γj is the scaling parameter of the Gaussian similarity distance kernel function.

[0060] In this embodiment, γ i The scale parameter is the sample point x. i The density of the neighborhood sample point set, γ j The scale parameter is the sample point x. j The density of the neighborhood sample point set is calculated, and the kernel function value between sample points is used as the distance metric between sample points.

[0061] For the i-th sample point and the j-th sample point, in this embodiment, the kernel function between the sample points is: The distance between sample points is K(x) i x j ) = W i，j .

[0062] If the number of sample points is n, then construct an n x n adjacency weight matrix A. In the adjacency weight matrix A... n，n In the diagram, for element A of class j in row i... i，j The value of this element can be represented as: A i，j =W i，j .

[0063] Based on the adjacency matrix, an n x n degree matrix D is constructed. The degree matrix D is a diagonal matrix. Specifically, in the degree matrix D, if the sequence value corresponding to the row of an element is different from the sequence value corresponding to the column, then the value of that element is 0; if the sequence value corresponding to the row of an element is different from the sequence value corresponding to the column, then for the element in the i-th column and i-th row, the value of that element can be represented as...

[0064] A Laplace matrix is ​​constructed based on the adjacency weight matrix and the degree matrix. The Laplace matrix L can be expressed as L = DA. The Laplace matrix L is then normalized to obtain the normalized Laplace matrix L. * The normalized Laplace matrix L * It can be represented as L * =ID -0.5 AD -0.5 .

[0065] Find the normalized Laplace matrix L. * The eigenvalues ​​and corresponding eigenvectors are obtained. The eigenvectors corresponding to the first few smallest eigenvalues ​​are selected, and an eigenvector matrix is ​​constructed using these eigenvectors, thus obtaining the dimensionality-reduced dataset.

[0066] If there are n sample points, the eigenvectors corresponding to the first k smallest eigenvalues ​​are taken, and then a k-row, n-column eigenvector matrix M is constructed using these eigenvectors. In the eigenvector matrix M, each column represents the dimensionality reduction feature of the sample points.

[0067] After obtaining the feature vector matrix, several categories are obtained through dimensionality reduction feature classification, and then the carbon labels corresponding to the several categories are obtained.

[0068] In this embodiment, the process of dimensionality reduction feature classification is as follows.

[0069] A sparse network topology is constructed, and the sparse network topology is divided into several sub-network topologies. The several sub-network topologies are merged to obtain the carbon tag.

[0070] The specific process for constructing a sparse network topology is as follows.

[0071] Based on the feature vector matrix, traverse all sample point pairs, and if the following conditions are met between samples.

[0072] For the i-th sample point and the j-th sample point, if the condition: j∈f r (i)Λi∈f r (j), where f r (i) represents the set of the first r smallest samples whose distance to the i-th sample is Λ, and Λ represents the AND in the logical relation word. Then there is a connecting edge between the i-th sample point and the j-th sample point; otherwise, there is no connecting edge.

[0073] After traversing all point pairs, the sparse network topology is obtained.

[0074] The specific process of dividing the sparse network topology into several sub-network topologies is as follows.

[0075] Traverse the edges of the sparse network topology and determine whether the traversed edges satisfy the sparsity condition.

[0076] If the sparsity condition is met, the edges of the traversal are retained; if the sparsity condition is not met, the edges of the traversal are not retained.

[0077] Until the natural edge traversal is completed, several sub-network topologies of the sparse network topology are obtained.

[0078] The sparsity condition is: for any two sample points on a traversed edge, at least one of the sample points has a number of neighbors greater than a neighbor threshold.

[0079] Mutual neighbors are defined as follows: for the i-th sample point, if the j-th sample point is a mutual neighbor of the i-th sample point, then the following conditions are met: in; It is represented as an equivalence relation; Nteo(i) represents the mutual neighbors of the i-th sample point.

[0080] Based on this, the number of mutual neighbors is defined as the number of mutual neighbors of a sample point.

[0081] The process of merging the sub-network topologies is described below.

[0082] Traverse all combinations of the two sub-network topologies, obtain the maximum similarity of the combination, merge the two sub-network topologies in the combination corresponding to the maximum similarity, and update the set of sub-network topologies and the number of sub-network topologies in the set of sub-network topologies.

[0083] If the number of sub-network topologies is greater than the set number of categories, the above steps continue; if the number of sub-network topologies is equal to the set number of categories, the entire merging process ends.

[0084] Each sub-network topology in the set of sub-network topologies is regarded as a category, and the carbon label corresponding to each category is obtained.

[0085] The similarity between the sub-network topologies is measured by dividing the number of identical mutual neighbors between the sub-network topologies by the sum of the number of mutual neighbors of the sub-network topologies.

[0086] For subnetwork topology G i With subnetwork topology G j The expression for their similarity is: SUM is a function that returns the number of elements in a set.

[0087] In this embodiment, the number of categories is at least three, and the number of carbon labels is at least three. Different carbon labels correspond to different carbon profiles, and different carbon profiles correspond to different power supply methods. During the power supply process, the power grid-related system selects the power supply method based on the carbon profile of the power load node. The power grid-related system acquires the carbon profile of the power load node in real time. If the carbon profile of the power load node changes, the power grid-related system selects the appropriate power supply method to supply power to the power load node.

[0088] In this embodiment, the process of supplying power to the grid based on the carbon profile is as follows.

[0089] In the power grid power supply decision-making system based on carbon profiling, the core of power supply mode formulation lies in mapping the carbon profile characteristics of power load nodes to power grid operating parameters, and realizing refined power allocation by constructing a dynamic matching mechanism between carbon profiles and power supply modes.

[0090] The power grid dispatching system categorizes power supply levels into five types based on carbon-related characteristics analyzed in carbon profiles: zero-carbon priority, low-carbon optimization, medium-carbon regulation, high-carbon constraint, and process-sensitive. The system acquires the carbon-related characteristics of each load node in real time, and based on these characteristics, generates a real-time carbon profile for each load node. If the carbon profile of a load node changes, the system selects the appropriate power supply method to supply power to that load node based on its carbon profile.

[0091] For nodes with high carbon constraints as indicated by carbon profiles, the power grid will implement a mandatory renewable energy penetration rate enhancement strategy. During peak load periods, the power supply circuit will be switched to a direct connection to a nearby wind farm through dynamic topology reconfiguration, and the power smoothing function of the energy storage system will be activated to ensure that the node must absorb no less than 50% of the photovoltaic power generation during the daily period from 12:00 to 16:00. At the same time, carbon flow tracking devices will be deployed to monitor the implicit carbon emissions of the power supply link in real time. When the penetration rate of thermal power exceeds the threshold, the reactive power compensation equipment will be automatically triggered to adjust and reduce the equivalent carbon emissions by improving the power factor.

[0092] For zero-carbon priority data center nodes with prominent clean energy absorption capabilities in the carbon profile, the power grid dispatch system opens a two-way energy exchange interface, allowing their energy storage units to participate in frequency regulation ancillary services. By constructing a dynamic electricity price-carbon price coupling model, the power supply demand of such nodes is prioritized during the electricity market clearing process. When the system detects the potential carbon emission risk of their standby diesel generator sets, it immediately initiates the green electricity standby channel switching protocol, forcibly connecting them to the dispatchable resource pool of the regional virtual power plant. For low-carbon optimized machinery manufacturing plants with high carbon emission reduction potential scores, the power supply strategy adopts a flexible load control mode. During the low-carbon operation of the power grid, demand response signals guide the centralized operation of adjustable loads such as electroplating production lines and heat treatment workshops. At the same time, differentiated power supply is implemented based on equipment energy efficiency levels: energy efficiency level A equipment is connected to the main grid direct supply circuit, level B equipment is switched to the energy storage buffer power supply channel, and level C equipment needs to be matched with customized power supplies from carbon capture power plants.

[0093] Power supply strategies for process-sensitive nodes need to be deeply integrated with production process characteristics. For example, if the carbon profile of a steel company shows periodic carbon emission pulses during the electric arc furnace smelting stage, the power grid will deploy a phase-synchronous power supply control system. This system predicts the peak carbon emission period within the next 15 minutes by collecting process parameters such as molten steel temperature and electrode lifting speed. It then switches the power supply bus to thermal power units equipped with carbon capture devices in advance and adjusts the operating power of electrostatic precipitators to offset harmonic pollution. After the carbon emission pulse ends, it automatically reverts to a high-proportion renewable energy power supply mode. Simultaneously, it optimizes the power allocation of adjacent nodes through dynamic power flow calculations to avoid grid oscillations caused by sudden load changes. For injection molding load nodes with excessive carbon emission intensity fluctuations, the power supply system introduces a time-domain-frequency-domain joint analysis module to identify characteristic harmonic components during the mold heating stage. It designs customized power supply waveforms with carbon emission suppression functions and injects reverse harmonic current through active filters to reduce equivalent carbon emission intensity while ensuring power quality.

[0094] The power grid dispatch center uses a carbon profile feature extraction engine to analyze the spatiotemporal distribution of carbon emissions at each node in real time, constructing a three-dimensional decision matrix for power supply strategies: the horizontal dimension is the carbon emission intensity level, the vertical dimension is the clean energy absorption capacity, and the depth dimension is the carbon emission reduction response speed. For example, if the carbon profile of a semiconductor manufacturing node shows that it has rapid adjustment capabilities, even if it is currently at a medium-carbon regulation level, the power grid can still activate the "advanced carbon optimization power supply mode," pre-storing green electricity quotas for it during off-peak periods. When the production process enters the high-precision lithography stage, it switches to a zero-carbon power supply independent circuit, ensuring that the carbon footprint of the chip manufacturing process is traceable. For carbon profile collaboration scenarios in regional clusters, such as multiple high-carbon-constrained nodes forming a spatial cluster in an industrial park, the power grid will activate a carbon flow collaborative optimization algorithm. By reconstructing the 10kV distribution network topology, a "carbon-isolated power supply corridor" will be constructed. The power supply buses of such nodes will be physically isolated and connected to dedicated feeders of carbon capture power plants. At the same time, a distributed photovoltaic absorption priority controller will be deployed to ensure that renewable energy generation prioritizes coverage of loads within the corridor.

[0095] By implementing the specific process of the method involved in this embodiment, the adaptability and accuracy of the generated carbon tags are improved. This enables power grid-related systems to generate corresponding carbon tags based on the carbon characteristics of electricity load nodes using the carbon tags generated by the method involved in this invention. This allows for more accurate classification of each electricity load node, thereby accurately obtaining the carbon profile corresponding to the electricity load node and reducing the carbon emissions of the power grid during the power supply process.

[0096] Example 2: This example discloses a specific implementation flow of the method involved in the present invention. (See attached document.) Figure 3 .

[0097] In this embodiment, the carbon-related features are the same as those in Example 1, and will not be described again here.

[0098] In this embodiment, the specific process for calculating the carbon emissions of each energy-consuming link at each level is the same as that in Embodiment 1, and will not be repeated here.

[0099] In this embodiment, the process of obtaining the adjacency weight matrix and degree matrix is ​​the same as that in Embodiment 1, and will not be repeated here.

[0100] In this embodiment, the construction process of the Laplace matrix is ​​the same as that in Embodiment 1, and will not be repeated here.

[0101] In this embodiment, the dimensionality reduction dataset is the same as that in Embodiment 1, and will not be described again here.

[0102] In this embodiment, the process of dimensionality reduction feature classification is different from that in Embodiment 1. The K-medoids algorithm is used for classification to obtain the carbon tag corresponding to the category.

[0103] The process of dimensionality reduction feature classification in this embodiment will be described in detail below.

[0104] Based on the dimensionality reduction feature dataset and the set number of categories, sample points are randomly selected as initial category centers, and the number of initial category centers is the same as the set number of categories.

[0105] In one feasible process of this embodiment, Euclidean distance is used as the distance metric between sample points, and all sample points in the dimensionality reduction feature dataset are traversed. For a certain category center, if the distance between the traversed sample point and the category center is the smallest compared to the distances between other category centers, then the traversed sample point belongs to the category corresponding to the category center; if the distance between the traversed sample point and the category center is not the smallest compared to the distances between other category centers, then the traversed sample point does not belong to the category corresponding to the category center.

[0106] After traversing all sample points, update the class center of all class sample sets. The update method is as follows: obtain the sample centroid in the class center sample set, and obtain a sample point in the class sample set such that the Euclidean distance between the sample point and the sample centroid is minimized, and regard the sample point as the new class center.

[0107] It should be noted that, in this embodiment, for the i-th category, the specific method for obtaining the centroid of the category center sample set corresponding to that category is as follows: Among them, U i Let be the set of category center samples corresponding to the i-th category.

[0108] If the updated category center does not change, then obtain the category corresponding to the category sample set and obtain the carbon label corresponding to the category; if the updated category center changes, then continue to execute the above process.

[0109] In one feasible implementation process of this embodiment, based on the above-described specific implementation process, Manhattan distance is used as the distance metric between sample points.

[0110] In one feasible implementation process of this embodiment, based on the above-described specific implementation process, Chebyshev distance is used as the distance metric between sample points.

[0111] In one feasible implementation process of this embodiment, based on the above-described specific implementation process, cosine distance is used as the distance metric between sample points.

[0112] In one application scenario of this embodiment, the power grid system can obtain carbon tags corresponding to the electricity load nodes through the method involved in this invention, generate carbon profiles corresponding to the electricity load nodes based on the obtained carbon tags, and adopt different power supply measures for different carbon profile electricity load nodes in order to reduce carbon emissions during the power supply process.

[0113] The power grid dispatch center first uses carbon profiles of electricity load nodes to accurately classify a large number of industrial electricity load nodes connected to the grid. By analyzing key information contained in the carbon profiles, such as electricity load characteristics, production process flexibility, and energy management system capabilities, the dispatch center can identify which electricity load nodes have characteristics of "elasticity," "adjustability," or "no special operation required."

[0114] For example, some power load nodes may have production processes that allow for flexible scheduling, or they may have energy storage devices or standby generators. These features allow their power load to be adjusted and controlled to a certain extent, thus classifying them as "flexible loads" or "adjustable loads" power load nodes.

[0115] When the power grid faces peak electricity demand periods, such as the high temperatures of summer or the cold of winter, overall electricity demand rises sharply, increasing the pressure on grid operation. At this time, the grid dispatch center will activate the demand-side response mechanism, sending demand-side response signals to previously identified "elastic loads" or "adjustable loads." These signals are typically transmitted in a timely manner to the energy management system or dispatch personnel of the load nodes via various means, such as SMS and dedicated data channels, through the power communication network. The signal content clearly informs the load nodes of the current grid load status, such as "Grid load has reached peak, please participate in demand-side response," and may provide specific response targets, such as "Please reduce the load by X% within the next two hours" or "Please adjust the power consumption of key production lines to a certain level."

[0116] Upon receiving demand-side response signals, these electricity load nodes can flexibly adjust their production operation strategies based on their own production plans and carbon profile information. A common adjustment method is to implement peak-shifting, which involves shifting some non-urgent, non-continuous production tasks, such as the operation of auxiliary production lines and equipment maintenance, from peak grid load periods to off-peak periods. Another method is to reduce the operating power of some non-critical production lines; for example, while ensuring product quality and basic capacity, appropriately reducing the operating speed of production lines or the power output of equipment. For electricity load nodes with energy storage devices, the energy storage system can be activated to release pre-stored electricity, replacing grid draw, thereby directly reducing the electricity load on the load node from the grid.

[0117] By guiding "flexible load" or "adjustable load" nodes to proactively adjust their electricity consumption behavior during peak grid periods, the grid dispatch center can effectively reduce peak grid load without affecting the normal production and operation activities of these load nodes. This demand-side response measure acts like smoothing out the "peaks" of the grid load curve, reducing the operational pressure on the grid during peak periods and making grid operation more stable and secure. More importantly, by reducing electricity demand during peak periods, the grid's reliance on high-carbon emission power generation resources that typically start during peak periods can be reduced, such as coal-fired peak-shaving units or oil / gas-fired units, thereby reducing carbon emissions from the power system at the source and promoting a green and low-carbon transformation of the energy structure.

[0118] This invention has illustrated its purpose, technical solutions, and beneficial effects through specific embodiments. However, these embodiments are merely examples to demonstrate the application of the invention and do not constitute a limitation on the scope of protection of this invention. We explicitly state that any reasonable modifications, equivalent substitutions, or technical improvements guided by the spirit and principles of this invention should be included within the scope of protection of this invention. This means that as long as these changes do not deviate from the core idea and basic function of the invention, they should be protected by patent rights. The scope of protection of this invention should be broad, including all directly obvious variations as well as non-obvious innovations that a technical expert can reasonably deduce from the disclosure of this invention. This broad protection aims to promote further research and development based on this invention, while ensuring that its innovativeness and practicality are fully protected by law.

Claims

1. A power supply method based on carbon microscopy, characterized in that, The steps of the method include: The power grid-related systems calculate the carbon emissions of the electricity load nodes. If the carbon emission calculation is successful, the carbon characteristics of the electricity load nodes are obtained. Based on the carbon-related characteristics, a dimensionality-reduced dataset is obtained using a feature space-based classification method. Based on the dimensionality-reduced dataset, dimensionality-reduced feature classification is performed to obtain several categories. Obtain carbon tags corresponding to several categories to obtain carbon profiles, and supply power according to the carbon profiles of the power load nodes.

2. The power supply method based on carbon profiling according to claim 1, characterized in that, The process of obtaining the dimensionality reduction dataset includes: Obtain the density of the neighborhood sample point set for each sample, and determine the scale parameter of the kernel function based on the density of the neighborhood sample point set; The adjacency weight matrix and relevance matrix of the feature are constructed using the kernel function. The Laplacian matrix of the feature is constructed using the adjacency weight matrix and the degree matrix. The vectors corresponding to the first few eigenvalues ​​of the Laplacian matrix are obtained, and the dimensionality of each sample is reduced to obtain the dimensionality-reduced dataset.

3. A power supply method based on carbon microscopy according to claim 1 or 2, characterized in that, The process of dimensionality reduction feature classification includes: Construct a sparse network topology based on the dimensionality-reduced dataset; The sparse network topology is divided into several sub-network topologies; By using similarity measurement, the sub-network topologies are merged among several sub-network topologies until the number of sub-network topologies equals the number of categories, thereby obtaining several categories corresponding to several sub-network topologies.

4. A power supply method based on carbon profiling according to claim 1 or 2, characterized in that, The process of dimensionality reduction feature classification includes: The K-medoids algorithm is used to obtain several categories from the dimensionality-reduced dataset.

5. A power supply method based on carbon microscopy according to claim 1, characterized in that, If the carbon emission calculation at the electricity load node fails, the process of the method ends.

6. A power supply method based on carbon microscopy according to claim 1 or 5, characterized in that, The carbon-related characteristics include: coupling degree characteristics; The coupling degree feature is specifically the proportion of nodes on non-critical paths out of all nodes.

7. A power supply method based on carbon microscopy according to claim 1 or 5, characterized in that, The carbon-related characteristics include: load transfer potential; The load transfer potential is specifically defined as the sum of the power of all transferable equipment multiplied by the start-stop energy cost.

8. A power supply method based on carbon microscopy according to claim 1 or 5, characterized in that, The carbon-related characteristics include: carbon emissions; The carbon emissions specifically refer to the carbon emissions generated by the electricity load node during the electricity consumption process.

9. A power supply method based on carbon microscopy according to claim 1 or 5, characterized in that, Different carbon portraits correspond to different power supply methods; The power grid-related system acquires the carbon profile of the power load node in real time. If the carbon profile of the power load node changes, the power grid-related system selects a power supply method to supply power to the power load node.

10. A power supply method based on carbon microscopy according to claim 1 or 5, characterized in that, The number of carbon label types is at least 3, and the number of carbon image types is at least 3.

Images