Method and related device for online coordinated optimization and active response of multiple complementary loads

Abstract

The present application relates to the technical field of power system optimization, and discloses a method for online collaborative optimization and active response of multi-energy complementary load and related equipment, wherein the present application collects terminal energy node, energy flow, carbon flow and load operation data, constructs an electricity-carbon-energy knowledge graph, initializes load carbon fingerprint characteristics, realizes energy-carbon correlation mapping and device-level carbon characteristic perception; a dynamic electricity-carbon coupling model is established by fusing deep learning and thermodynamic mechanism to solve the instantaneous carbon emission intensity of each energy supply path. By analyzing the carbon difference gradient of multi-energy supply, the correlation between load, carbon potential and device energy efficiency is explored, and a carbon emission reduction gain-oriented alternative energy supply path set is constructed; a multi-objective convex optimization model is built and converted into control instructions, and the control is executed relying on the cloud-edge collaborative architecture. Benefit evaluation and abnormal diagnosis are carried out, and the power marketing strategy is iteratively optimized to form a closed-loop management and control. The present application reduces the response link and operation time, and meets the low-carbon real-time regulation and control requirements of multi-energy system under high fluctuation scenarios.

Description

Technical Field

[0001] This invention relates to the field of power system optimization technology, specifically to a method and related equipment for online collaborative optimization and active response of multi-energy complementary loads. Background Technology

[0002] In recent years, the global energy system has been accelerating its transformation towards low-carbon and clean energy. As a crucial carrier of energy consumption and carbon emissions, the power system is gradually evolving from a single electricity supply system to a comprehensive energy system integrating electricity, gas, and heat. The widespread integration of various energy facilities, including distributed photovoltaics, wind power, power-to-heat conversion, gas-fired internal combustion engines, and energy storage, is driving the power grid's operation from a "source-following-load" model to a "source-grid-load-storage coordinated interaction." Against this backdrop, electricity-carbon integration has become an important direction for comprehensive energy optimization. This involves introducing carbon emission factors and grid carbon potential parameters while optimizing energy flow, achieving dynamic coupling analysis of electricity and carbon emission intensity. Multi-energy complementary load regulation is gradually expanding from simple economic optimization to multi-objective coordinated control that considers carbon intensity, energy structure, and system security, providing technical support for building new power systems and green, low-carbon energy systems.

[0003] However, due to the complexity of modeling the electro-carbon coupling mechanism, the high real-time requirements of multi-source energy collaborative control, and the difficulty of multi-objective optimization coordination, multi-energy complementary load collaborative optimization and active response face many challenges: (1) The modeling of the electro-carbon coupling mechanism is highly complex. The carbon potential of the power grid is time-varying and regionally differentiated, and is significantly affected by the output of new energy sources, cross-regional power trading, and load structure. There are different carbon emission marginal characteristics between electricity-to-heat, gas-fired power generation, and grid power in multi-energy systems. The electro-carbon coupling relationship exhibits nonlinear and dynamic coupling characteristics, making unified modeling and real-time calculation difficult. (2) The real-time requirements of multi-source energy collaborative control are high. Multi-energy complementary systems involve multiple physical networks such as electricity, gas, and heat, with different operating time scales and complex control logic. In the active load response scenario, carbon intensity assessment and control strategy switching need to be completed at the second or minute level, which places high demands on data acquisition, edge computing, and scheduling response capabilities. (3) The difficulty of multi-objective optimization coordination is high. Multi-energy collaborative regulation not only involves the optimization of economic costs, but also needs to take into account multiple objectives such as minimizing carbon emissions, equipment operation constraints, and user comfort. There are trade-offs between different objectives, which can easily lead to conflicts between carbon emission reduction and operating costs, and the stability and interpretability of optimization strategies face challenges. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the present invention aims to provide a method and related equipment for online collaborative optimization and active response of multi-energy complementary loads, so as to solve the technical problems of complex modeling of electric carbon coupling mechanism, strong real-time control of multi-source energy collaborative control, and great difficulty in multi-objective optimization coordination in the existing technology.

[0005] This invention is achieved through the following technical solution: In a first aspect, the present invention provides a method for online collaborative optimization and active response of multi-energy complementary loads, comprising: Acquire terminal-side energy node, energy flow, carbon flow and load operation data, construct an electricity-carbon-energy knowledge graph and initialize load carbon fingerprint features to obtain energy-carbon correlation mapping relationship and equipment-level carbon characteristic perception results; Based on the energy-carbon correlation mapping relationship and the device-level carbon characteristic perception results, a dynamic electric-carbon coupling model is constructed by integrating deep learning and thermodynamic mechanisms to obtain the instantaneous carbon emission intensity of each energy supply path. Based on the instantaneous carbon emission intensity of each energy supply path and combined with real-time load demand, the carbon difference gradient between multi-energy supply schemes is analyzed, the correlation between load, carbon potential and equipment energy efficiency is explored, and a set of alternative energy supply paths guided by carbon emission reduction gains is constructed. Based on the set of alternative power supply paths, a multi-objective convex optimization model is constructed, which is then transformed into executable control commands and executed through a cloud-edge collaborative architecture to obtain the control execution results. Based on the results of regulation implementation, the effectiveness of regulation is evaluated and anomalies are diagnosed. Combined with user load characteristics, the power marketing strategy is iterated to form a closed-loop optimization.

[0006] Preferably, the specific process of constructing an electricity-carbon-energy knowledge graph and initializing load carbon fingerprint features to obtain energy-carbon correlation mapping relationships and device-level carbon characteristic perception results includes: By introducing knowledge graph technology and non-intrusive load monitoring algorithms, an explicit characterization of the implicit carbon footprint at the terminal side is obtained; When constructing a knowledge graph, define the graph. ,in Representative energy nodes include transformer energy nodes and electric boiler energy nodes; Represents a connection of energy flow or carbon flow. Represents the mapping rules between nodes; the Laplacian feature mapping formula is introduced as follows: ,in, The adjacency matrix is ​​an energy-carbon relational matrix. The degree matrix is ​​used to identify the bottleneck node with the highest carbon emission intensity in the energy network by calculating the eigenvalues ​​of the matrix. During load carbon fingerprint initialization, the NILM formula based on instantaneous power decomposition is used, assuming the bus current signal is... Harmonic feature vectors of each device are extracted using discrete Fourier transform. Using the residual minimization decomposition formula:

[0007] in, For the first The on / off status of this type of equipment; n is the equipment serial number; n is the total number of all electrical devices in the system; For harmonic characteristic mapping terms; The sum of squared residuals is used to measure the magnitude of the current fitting error. To perform current superposition for all load devices; The bus-side mixed carbon flow is separated into individual device carbon traces to complete the full perception of micro-load carbon characteristics. Based on the full perception of micro-load carbon characteristics, the energy-carbon correlation mapping relationship and device-level carbon characteristic perception results are obtained.

[0008] Preferably, the specific process of constructing a dynamic electrocarbon coupling model includes: A dynamic electro-carbon coupling model for energy conversion is constructed, and a bidirectional long short-term memory network is used to process time-series energy data to capture the impact of historical load inertia and future carbon potential prediction on the current state. For electro-thermal conversion equipment, a thermodynamic exergy efficiency-corrected carbon intensity model is introduced, with the following formulas:

[0009] in, for The equivalent carbon emission intensity of the electro-thermal conversion equipment at any given time; for The marginal emission factor of the upstream power grid reflects the proportion of clean energy on the grid side in real time; This is the efficiency coefficient for electrothermal conversion; Carnot factor, Absolute temperature for environmental reference; This is the absolute temperature at the heat output terminal; for Random noise disturbance term of carbon emissions from the power grid at any given time; Introducing a weighted formula for attention mechanisms:

[0010] in, The weight representing the contribution of carbon intensity fluctuations at different times to the prediction target. and They are respectively The hidden layer state vector and global context vector at each time step represent the feature abstraction extracted by the deep learning model; It is a natural exponential function, which maps attention scores to non-negative values; This is an attention scoring function used to calculate the degree of association and matching between a single temporal hidden state and the global context; For any number The hidden layer state vector at time step n is obtained by iterating through all time steps; ∑ is the summation operation, which accumulates the exponential scores of all time steps. A dynamic electrocarbon coupling model is constructed by deeply coupling a data-driven deep learning model with a thermodynamic physical mechanism model.

[0011] Preferably, the correlation between load, carbon potential, and equipment energy efficiency is explored to construct a set of alternative energy supply paths guided by carbon emission reduction gains. The Apriori association rule algorithm is then used to explore the correlation between load, carbon potential, and equipment energy efficiency, and the carbon emission gradient of different energy path combinations is calculated in real time. The specific process includes: Distributed energy carbon differential analysis introduces a carbon emission comparison discriminant formula:

[0012] in, This represents the difference in instantaneous carbon emission reduction gradient. This represents the current real-time total load demand power. For the first Real-time output power of the internal distributed energy source; For the first Carbon emission coefficient per unit of electricity generated from a particular energy form; For the first The transmission loss correction factor for various energy supply paths; Multi-energy complementary path construction, introducing information gain decision criteria:

[0013] In the formula, It is the entropy value of the system's original carbon intensity state, representing a measure of the system's uncertainty before regulation; In order to perform specific regulatory actions Under the influence of this effect, the proportion and weight of subset samples to the total sample; The number of samples in the v-th state subset after being divided by the control action A; The total number of samples in the total sample set S; Using carbon difference as the target variable, the contribution of each control variable to carbon emission reduction is evaluated, and a multi-level decision search tree is automatically constructed, with carbon emission reduction gain as the primary search driver.

[0014] Preferably, based on the set of alternative power supply paths, a multi-objective convex optimization model is constructed, which is then transformed into executable control commands and executed through a cloud-edge collaborative architecture to obtain the control execution results. The specific process includes: The second-order cone programming (SOCP) algorithm is introduced to transform the nonlinear power flow constraint into a convex constraint, ensuring the online real-time performance and convergence of the control action; Construct a multi-objective optimization function, as shown in the following formula:

[0015] in, These are the weighting coefficients for each objective. For time-of-use electricity pricing, For the predicted real-time carbon emission intensity, This represents the system's adjustment increment for grid frequency offset; The summation is performed over time, where t is the scheduling time and T is the total scheduling duration. The economic cost weighting coefficient represents the priority of the electricity cost control target in the overall optimization. Let t be the electrical power exchanged between the system and the upper-level power grid at time t, with positive values ​​representing purchased electricity and negative values ​​representing sold electricity; The weighting coefficient for low-carbon emission reduction is used to adjust the priority of carbon emission targets in the overall optimization. This is the power grid frequency regulation weighting coefficient, which characterizes the priority of the system in participating in the power grid frequency stability regulation objective; The frequency adjustment increment squared term is used to weaken the influence of positive and negative deviations by adopting a squared form, thereby achieving smooth suppression of frequency fluctuations. We introduce the Lagrange multiplier method to construct the augmented objective function, as shown below:

[0016] in, To augment the Lagrangian function, the constrained optimization problem is transformed into an unconstrained optimization problem; For Lagrange multipliers, representing the th The marginal carbon cost and power deviation cost of each additional unit load at each node; Indicates the first The network loss power of each node; transforming the constrained optimization problem into an unconstrained optimization problem; The objective function for system comprehensive optimization includes economic cost, carbon emission cost, and frequency regulation cost; To sum over all M power grid / energy network nodes in the system sequentially; The real-time power output of the distributed power source at node j; The equivalent power of charging and discharging for the energy storage device at node j is given by the positive value for discharging and the negative value for charging. The real-time power output of the distributed power source at node j; Adopting a cloud-edge collaborative architecture, the cloud formulates global strategies at the minute level, and the edge controller performs frequency tracking at the second level; for flexible loads, pulse width modulation technology is used to fine-tune the input power, so that the load carbon footprint curve is fitted with the photovoltaic or wind power output curve in the region in real time, and active carbon footprint offsetting is carried out.

[0017] Preferably, the specific process of evaluating the effectiveness of regulation and diagnosing anomalies based on the results of regulation execution, and iterating the electricity marketing strategy in combination with user load characteristics to form a closed-loop optimization includes: The actual carbon reduction C_(act) after each load adjustment, the response speed T_(resp), the economic cost increment ΔM, and the grid carbon potential prediction deviation are calculated. A high-dimensional feature vector X is formed, and the abnormal score of each regulated sample is calculated using the abnormal score s(x,n). When s(x,n) approaches 1, it means that the regulated action has produced an abnormal deviation. The cause of the abnormality is located by the correlation analysis backtracking algorithm. At the marketing closed-loop level, a collaborative filtering recommendation model is introduced. Based on users' historical carbon reduction performance, low-carbon benchmark enterprises with similar load characteristics are identified through cosine similarity. Optimized electricity marketing packages are then pushed to users with poor performance, forming a closed loop of model correction and marketing strategy optimization.

[0018] Furthermore, anomaly scoring The formula is as follows:

[0019] in, The average path height of the sample across all isolation trees. The average constant of the search path; The cosine similarity formula is as follows:

[0020] in, This represents the rating index for user u in the i-th control dimension; Let v be the quantitative scoring index for user v under the i-th control dimension.

[0021] Secondly, the present invention also provides a system for online collaborative optimization and active response of multi-energy complementary loads, comprising: The graph fingerprint module is used to acquire terminal-side energy node, energy flow, carbon flow and load operation data, construct an electricity-carbon-energy knowledge graph and initialize load carbon fingerprint features, and obtain energy-carbon correlation mapping relationship and equipment-level carbon characteristic perception results; The coupling modeling module is used to construct a dynamic electric carbon coupling model based on the energy-carbon correlation mapping relationship and the device-level carbon characteristic perception results, and integrate deep learning and thermodynamic mechanisms to obtain the instantaneous carbon emission intensity of each energy supply path. The path construction module is used to analyze the carbon difference gradient between multiple energy supply schemes based on the instantaneous carbon emission intensity of each energy supply path and combined with real-time load demand, explore the correlation between load, carbon potential and equipment energy efficiency, and construct a set of alternative energy supply paths guided by carbon emission reduction gains. The regulation and execution module is used to construct a multi-objective convex optimization model based on the set of alternative power supply paths, transform it into executable control commands, and execute them through a cloud-edge collaborative architecture to obtain the regulation and execution results. The evaluation and iteration module is used to evaluate the effectiveness of regulation and diagnose anomalies based on the results of regulation execution, and to iterate the power marketing strategy in combination with user load characteristics to form a closed-loop optimization.

[0022] Thirdly, the present invention also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for online collaborative optimization and active response of multi-energy complementary load as described above.

[0023] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for online collaborative optimization and active response of multi-energy complementary loads as described above.

[0024] Compared with the prior art, the present invention has the following beneficial technical effects: This invention provides a method for online collaborative optimization and proactive response of multi-energy complementary loads. By combining an electricity-carbon-energy knowledge graph with a dynamic deep learning model, along with exergy efficiency correction and time-varying marginal emission factors, it achieves a fusion model of physical mechanisms and data-driven approaches. This effectively characterizes instantaneous carbon intensity changes under regional differences and fluctuating renewable energy scenarios, improving model prediction accuracy and operational stability under complex operating conditions. It also enables dynamically computable and traceable representation of the electricity-carbon coupling relationship. Furthermore, relying on a cloud-edge collaborative architecture combined with a convex optimization scheduling model, it achieves second-level load tracking at the edge and minute-level global carbon potential assessment and strategy updates in the cloud, thus effectively managing carbon... The difference analysis is integrated into the path search logic, which can quickly switch the energy supply structure under the high carbon pulse condition of the power grid, shorten the response link and calculation time, and meet the low-carbon real-time control requirements under the high fluctuation scenario of multi-energy system. This invention integrates economic cost, carbon cost and power grid support contribution into the convex optimization framework, and uses Lagrange multipliers to complete constraint relaxation and dynamic adjustment of marginal carbon price. It is equipped with anomaly diagnosis and power marketing iteration mechanism to build a complete control optimization closed loop. While ensuring the safety of equipment operation and user experience, it takes into account carbon emission reduction benefits and operation economy, and improves the operation stability and practical interpretability of the overall optimization strategy. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the method for online collaborative optimization and active response of multi-energy complementary loads in this embodiment of the invention. Figure 2 This is a framework diagram of the method for online collaborative optimization and active response of multi-energy complementary loads in an embodiment of the present invention; Figure 3 This is a system schematic diagram of online collaborative optimization and active response of multi-energy complementary loads in an embodiment of the present invention; In the diagram: 1. Graph fingerprint module; 2. Coupled modeling module; 3. Path construction module; 4. Regulation and execution module; 5. Evaluation and iteration module. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0027] The purpose of this invention is to provide a method and related equipment for online collaborative optimization and active response of multi-energy complementary loads, so as to solve the technical problems of complex modeling of electric carbon coupling mechanism, strong real-time control of multi-source energy collaborative control, and great difficulty in multi-objective optimization coordination in the prior art.

[0028] The present invention will now be described in further detail with reference to the accompanying drawings: Example 1 See Figure 1 and Figure 2 In one embodiment of the present invention, a method for online collaborative optimization and active response of multi-energy complementary loads is provided, comprising: Step 1: Obtain terminal-side energy node, energy flow, carbon flow and load operation data, construct an electricity-carbon-energy knowledge graph and initialize load carbon fingerprint features to obtain energy-carbon correlation mapping relationship and equipment-level carbon characteristic perception results; Specifically, a comprehensive "electricity-carbon-energy" sensing system covering both the physical and digital sides is being constructed. The system utilizes a dense network of intelligent sensing terminals within the industrial park to achieve high-frequency collection of full-volume data from mains power input, distributed generating units, electro-thermal conversion equipment, and energy storage systems. This step incorporates knowledge graph technology and non-intrusive load monitoring algorithms to achieve explicit characterization of the implicit carbon footprint at the terminal side.

[0029] At the knowledge graph construction level, the graph is defined. ,in Represents energy nodes (such as transformers and electric boilers). Represents a connection of energy flow or carbon flow. This represents the mapping rules between nodes. To quantify the carbon correlation density between nodes, the Laplace eigenmap formula is introduced:

[0030] in, The adjacency matrix is ​​an energy-carbon relational matrix. Let be the degree matrix. By calculating the eigenvalues ​​of this matrix, the system can identify the bottleneck node with the highest carbon emission intensity in the energy network. At the load fingerprint initialization level, to solve the problem of carbon accounting for terminal equipment at low cost, the NILM formula based on instantaneous power decomposition is adopted. Let the bus current signal be... Harmonic feature vectors of each device are extracted using discrete Fourier transform. The residual is minimized using the following formula:

[0031] in, For the first The on / off status of this type of equipment. n is the equipment serial number; n is the total number of all electrical devices in the system; For harmonic characteristic mapping terms; The sum of squared residuals is used to measure the magnitude of the current fitting error. The formula superimposes current to cover all load devices; its physical meaning lies in separating the "mixed carbon flow" on the bus side into "independent carbon traces" for individual devices. Based on this step, there is no need to install carbon monitoring devices on each terminal; the algorithm alone can achieve full perception of the carbon characteristics of micro-loads, greatly reducing the deployment cost of the perception layer. The system establishes a high-dimensional mapping relationship from the macro power grid to micro devices, providing a structured data foundation for subsequent precise control.

[0032] Step 2: Based on the energy-carbon correlation mapping relationship and the device-level carbon characteristic perception results, a dynamic electric-carbon coupling model is constructed by integrating deep learning and thermodynamic mechanisms to obtain the instantaneous carbon emission intensity of each energy supply path. Specifically, after acquiring multidimensional high-frequency data, the core of this step lies in constructing a dynamic electro-carbon coupling model for the energy conversion process, employing a bidirectional long short-term memory network (Bi-LSTM) to process energy data with time-series characteristics. The advantage of this method lies in its ability to simultaneously capture the impact of historical load inertia and future carbon potential predictions on the current state. For electro-thermal conversion equipment, not only is its thermal efficiency considered, but thermodynamic exergy efficiency is also introduced to correct the carbon intensity model.

[0033] in, for The equivalent carbon emission intensity of the electro-thermal conversion equipment at any given time; for The marginal emission factor of the upstream power grid reflects the proportion of clean energy on the grid side in real time; This is the efficiency coefficient for electrothermal conversion; Carnot factor Absolute temperature for environmental reference; This is the absolute temperature at the heat output terminal; for The random noise disturbance term for carbon emissions from the power grid at any given time measures the degree of energy devaluation after electrical energy is converted into heat. The physical significance of the formula lies in correcting the erroneous understanding in traditional models that "heat is equivalent to work," making the allocation of carbon emissions more consistent with physical reality.

[0034] Furthermore, to improve the model's robustness under new energy fluctuations, an attention mechanism weighted formula is introduced:

[0035] in, The weight representing the contribution of carbon intensity fluctuations at different times to the prediction target. and They are respectively The hidden layer state vector and global context vector at each time step represent the feature abstraction extracted by the deep learning model. It is a natural exponential function, which maps attention scores to non-negative values; This is an attention scoring function used to calculate the degree of association and matching between a single temporal hidden state and the global context; For any number The hidden layer state vector at time step n is obtained by iterating through all time steps; ∑ is the summation operation, which accumulates the exponential scores of all time steps.

[0036] This model deeply couples a data-driven deep learning model with a thermodynamic physical mechanism model, solving the problem of prediction failure of pure data models under extreme conditions. This modeling approach transforms the electricity-carbon coupling relationship from a static average emission factor to a dynamic instantaneous emission intensity, providing high-precision computing power support for subsequent searches for the path to maximize carbon difference, enabling the system to perceive the real environmental cost behind each kilowatt-hour in real time.

[0037] Step 3: Based on the instantaneous carbon emission intensity of each energy supply path and combined with real-time load demand, analyze the carbon difference gradient between multi-energy supply schemes, explore the correlation between load, carbon potential and equipment energy efficiency, and construct a set of alternative energy supply paths guided by carbon emission reduction gains. Specifically, this step serves as the decision-making hub for "proactive response," aiming to find the optimal energy alternative pathway. The system utilizes the Apriori association rule algorithm to uncover strong correlations among "production load, external carbon potential, and equipment energy efficiency." The core logic lies in the real-time calculation of carbon emission gradients for different energy pathway combinations.

[0038] Step 3.1: Distributed Energy Carbon Difference Analysis. Introduce the carbon emission comparison formula:

[0039] in, This represents the difference in instantaneous carbon emission reduction gradient. This represents the current real-time total load demand power. For the first Real-time output power of the internal distributed energy source; For the first Carbon emission coefficient per unit of electricity generated from a particular energy form; For the first The transport loss correction coefficient for various energy supply paths. The physical meaning of this formula is to quantify "how much more carbon emissions would be generated if no regulation were implemented, compared to the current energy supply through multi-energy complementary systems".

[0040] Step 3.2: Multi-energy complementary path construction. To quickly locate a path from a massive number of path combinations, an information gain decision criterion is introduced:

[0041] In the formula, It is the entropy value of the system's original carbon intensity state, representing a measure of the system's uncertainty before regulation; In order to perform specific regulatory actions Under the influence of this effect, the proportion and weight of subset samples to the total sample; The number of samples in the v-th state subset after being divided by the control action A; The total number of samples in the total sample set S. The system uses the carbon difference as the target variable and automatically constructs a multi-level decision search tree by evaluating the contribution of each control variable (such as the depth of energy storage discharge and the proportion of gas turbine output) to the reduction of carbon emissions.

[0042] This step changes the limitation of traditional power dispatching, which only focuses on "power gaps," and instead uses "carbon emission reduction gains" as the primary search driver. When a high carbon potential pulse is detected in the grid, the decision engine quickly identifies the alternative path with the largest carbon difference. This path construction method based on association mining can process hundreds of complex energy flow combinations within seconds, greatly improving the system's flexibility and real-time response in "carbon finding" within complex energy networks.

[0043] Step 4: Based on the set of alternative power supply paths, construct a multi-objective convex optimization model, transform it into executable control commands, and execute them through a cloud-edge collaborative architecture to obtain the control execution results; Specifically, this step aims to transform the candidate paths selected in step 3 into high-precision physical control commands. Due to the strong nonlinear coupling characteristics within multi-energy complementary systems, traditional linear programming often struggles to balance solution speed and global optimality. Therefore, this step innovatively introduces the second-order cone programming (SOCP) algorithm, transforming the nonlinear power flow constraint into a convex constraint, thereby ensuring the online real-time performance and convergence of the control actions.

[0044] The system was built with operating costs Environmental carbon costs and the contribution of power grid frequency regulation This is a multi-objective optimization function with variables. Its core objective function formula is defined as follows:

[0045] in, These are the weighting coefficients for each objective. Let t be the time-of-use electricity price. For the predicted real-time carbon emission intensity, This represents the system's adjustment increment for grid frequency offset; The summation is performed over time, where t is the scheduling time and T is the total scheduling duration. The economic cost weighting coefficient represents the priority of the electricity cost control target in the overall optimization. Let t be the electrical power exchanged between the system and the upper-level power grid at time t, with positive values ​​representing purchased electricity and negative values ​​representing sold electricity; The weighting coefficient for low-carbon emission reduction is used to adjust the priority of carbon emission targets in the overall optimization. This is the power grid frequency regulation weighting coefficient, which characterizes the priority of the system in participating in the power grid frequency stability regulation objective; The frequency adjustment increment squared term is used to weaken the influence of positive and negative deviations by adopting a squared form, thereby achieving smooth suppression of frequency fluctuations.

[0046] To address the spatiotemporal coordination problem among "electricity-heat conversion, energy storage, and flexible load" when dealing with complex physical boundary conditions, the Lagrange multiplier method is introduced to construct an augmented objective function, and dual variables are introduced. Real-time relaxation of power balance constraints at each node:

[0047] in, To augment the Lagrangian function, the constrained optimization problem is transformed into an unconstrained optimization problem; For Lagrange multipliers, representing the th The marginal carbon cost and power deviation cost of each additional unit load at each node; Indicates the first Network loss power of each node, The objective function for system comprehensive optimization includes economic cost, carbon emission cost, and frequency regulation cost; To sum over all M power grid / energy network nodes in the system sequentially; The real-time power output of the distributed power source at node j; The equivalent power of charging and discharging for the energy storage device at node j is given by the positive value for discharging and the negative value for charging. The real-time power output of the distributed power source at node j.

[0048] The physical significance of this formula lies in the fact that it not only finds the point with the lowest total carbon emissions but also monitors the "carbon pressure" and "voltage" status of each energy node in real time. If a node experiences a surge in carbon potential due to a decrease in the proportion of green electricity, the Lagrange multiplier... The "carbon price" of the node will be automatically increased, thereby forcing the control logic to automatically cut off the high-energy-consuming load of the node and activate distributed gas turbine units or energy storage for "physical carbon substitution".

[0049] During response execution, the system adopts a "cloud-edge collaboration" architecture. The cloud is responsible for formulating minute-level global strategies based on the aforementioned convex optimization model, while the edge controllers deployed on the production line side perform second-level frequency tracking based on model predictive control. For flexible loads such as air conditioners and variable frequency water pumps, the system uses pulse width modulation technology to fine-tune their input power, enabling their carbon footprint curves to be fitted in real time with the output curves of photovoltaic or wind power in the region.

[0050] This approach addresses the shortcomings of previous load response methods that focused solely on electricity prices without considering carbon intensity and total emissions without considering spatiotemporal distribution. By directly embedding carbon emission intensity into the physical power flow optimization equations, it ensures that during periods of strong grid carbon fluctuations, the multi-energy complementary system can proactively offset the surge in its carbon footprint through internal energy conversion, much like a biological instinct. This not only meets the physical requirements of power peak shaving but also achieves truly proactive low-carbon response, providing technical support for the stable operation of industrial parks under the backdrop of tight carbon quotas.

[0051] Step 5: Based on the results of regulation execution, evaluate the regulation effectiveness and diagnose anomalies, and combine user load characteristics to iterate the power marketing strategy to form a closed-loop optimization.

[0052] Specifically, this step constructs a closed-loop mechanism for evaluating the regulatory benefits and diagnosing anomalies based on the Isolation Forest algorithm. Isolation Forest is a non-parametric anomaly detection method suitable for high-dimensional data. Its core idea is that anomalous samples are more easily isolated, meaning they have shorter path lengths in the decision tree branches.

[0053] In practice, the actual carbon reduction after each load adjustment will be used as the basis for implementation. Response speed Incremental economic costs and the bias in the prediction of the carbon potential of the power grid Composition of high-dimensional feature vectors The anomaly score for each regulated sample is calculated using the following formula:

[0054] in, The average path height of the sample across all isolation trees. This is the average constant of the search path. Physically, when... When the value approaches 1, it indicates that the control action has deviated abnormally. The system then uses correlation analysis and backtracking algorithms to determine whether the error is due to inaccurate predictions from the front-end model or from the back-end execution mechanism.

[0055] At the marketing closed-loop level, this step introduces a collaborative filtering recommendation model, which calculates the cosine similarity between users based on their historical carbon reduction performance:

[0056] in, This represents the rating index for user u in the i-th control dimension; This is a quantitative scoring index for user v under the i-th control dimension. Using this formula, the system can automatically identify "low-carbon benchmark enterprises" with similar load characteristics and automatically push optimized electricity marketing packages to users with poor performance.

[0057] This step elevates the originally purely physical level control feedback to data-driven business decision-making recommendations, which not only ensures the long-term stability of the system's operation, but also provides the power grid company with precise data tools for carrying out power marketing services.

[0058] This embodiment addresses the challenge of complex and unifying the modeling of the electro-carbon coupling mechanism. It combines an "electricity-carbon-energy" knowledge graph with a dynamic deep learning model, introducing exergy efficiency correction and time-varying marginal emission factors to achieve a fusion modeling approach that integrates physical mechanisms and data-driven methods. Compared to existing technologies, this method can characterize instantaneous carbon intensity changes under regional differences and fluctuations in new energy sources, significantly improving the model's prediction accuracy and stability under complex operating conditions, and enabling a dynamically computable and traceable representation of the electro-carbon coupling relationship.

[0059] This embodiment addresses the issue of insufficient real-time performance in multi-energy collaborative control by constructing a cloud-edge collaborative architecture and a convex optimization scheduling model. It achieves second-level load tracking at the edge and minute-level global carbon potential assessment and strategy updates in the cloud. Compared to traditional centralized scheduling, embedding carbon difference analysis into the path search logic enables rapid switching of energy structures when high-carbon pulses occur, significantly shortening response time and computation time, and meeting the real-time low-carbon control requirements of multi-energy systems in highly volatile scenarios.

[0060] This embodiment addresses the challenge of coordinating multi-objective optimization by integrating economic costs, carbon costs, and grid support contributions into a unified convex optimization framework. Lagrange multipliers are used to achieve constraint relaxation and dynamic adjustment of marginal carbon prices. Furthermore, anomaly assessment and marketing iteration mechanisms are introduced to form a closed-loop system of regulation-assessment-optimization. This solution achieves a synergistic balance between carbon emission reduction and economic benefits while ensuring equipment safety and user experience, thereby improving the stability and interpretability of the optimization strategy.

[0061] In summary, the method for online collaborative optimization and proactive response of multi-energy complementary loads provided in this embodiment combines an electricity-carbon-energy knowledge graph with a dynamic deep learning model, along with exergy efficiency correction and time-varying marginal emission factors, to achieve a fusion model of physical mechanisms and data-driven approaches. This effectively characterizes instantaneous carbon intensity changes under regional differences and renewable energy fluctuations, improves model prediction accuracy and operational stability under complex operating conditions, and enables dynamically computable and traceable representation of the electricity-carbon coupling relationship. Furthermore, relying on a cloud-edge collaborative architecture combined with a convex optimization scheduling model, the edge side completes second-level load tracking, while the cloud side achieves minute-level global carbon potential assessment and strategy updates. The new approach integrates carbon difference analysis into the path search logic, enabling rapid switching of energy supply structures under high-carbon pulse conditions in the power grid. This shortens the response chain and computation time, meeting the real-time low-carbon control requirements of multi-energy systems in high-fluctuation scenarios. Furthermore, it unifies economic costs, carbon costs, and grid support contributions into a convex optimization framework. By leveraging Lagrange multipliers, it achieves constraint relaxation and dynamic adjustment of marginal carbon prices, along with supporting anomaly diagnosis and power marketing iteration mechanisms. This constructs a complete control optimization closed loop, ensuring equipment operation safety and user experience while balancing carbon emission reduction benefits and operational economics, thus improving the overall operational stability and practical interpretability of the optimization strategy.

[0062] Example 2 according to Figure 3 As shown, this embodiment also provides a system for online collaborative optimization and proactive response of multi-energy complementary loads, including: The graph fingerprint module 1 is used to acquire terminal-side energy node, energy flow, carbon flow and load operation data, construct an electricity-carbon-energy knowledge graph and initialize load carbon fingerprint features, and obtain energy-carbon correlation mapping relationship and equipment-level carbon characteristic perception results; Coupled modeling module 2 is used to construct a dynamic electric carbon coupling model based on the energy-carbon correlation mapping relationship and the device-level carbon characteristic perception results, and integrate deep learning and thermodynamic mechanisms to obtain the instantaneous carbon emission intensity of each energy supply path. The path construction module 3 is used to analyze the carbon difference gradient between multiple energy supply schemes based on the instantaneous carbon emission intensity of each energy supply path and combined with real-time load demand, explore the correlation between load, carbon potential and equipment energy efficiency, and construct a set of alternative energy supply paths guided by carbon emission reduction gains. The regulation and execution module 4 is used to construct a multi-objective convex optimization model based on the set of alternative power supply paths, convert it into executable control commands and execute them through the cloud-edge collaborative architecture to obtain the regulation and execution results; Evaluation Iteration Module 5 is used to evaluate the effectiveness of regulation and diagnose anomalies based on the results of regulation execution, and iterate the power marketing strategy in combination with user load characteristics to form a closed-loop optimization.

[0063] Among them, the graph fingerprint module 1 realizes device-level carbon fingerprint analysis and energy-carbon mapping relationship initialization, providing a data foundation for subsequent refined carbon accounting and coordinated regulation.

[0064] The coupled modeling module 2 dynamically depicts the coupling relationship between the grid carbon potential and the multi-energy conversion efficiency, and outputs the instantaneous carbon emission intensity of each energy supply path in real time.

[0065] The path construction module 3 calculates the carbon difference gradient between multiple energy supply schemes based on real-time carbon intensity and load demand, and constructs an alternative path decision set guided by carbon emission reduction gains.

[0066] The control and execution module 4 transforms the low-carbon substitution path into a multi-objective convex optimization problem, generates executable control commands, and realizes online coordinated control of electric-to-heat conversion, gas turbine units, and energy storage.

[0067] The evaluation iteration module 5 assesses the carbon reduction of the response and diagnoses anomalies, forming a closed loop of model correction and marketing strategy optimization to support the continuous iterative upgrade of low-carbon load response.

[0068] Example 3 This embodiment also provides a mobile terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor, such as a program for online collaborative optimization and proactive response of multi-energy complementary loads.

[0069] When the processor executes the computer program, it implements the steps of the above-described method for online collaborative optimization and active response of multi-energy complementary loads, for example: Acquire terminal-side energy node, energy flow, carbon flow and load operation data, construct an electricity-carbon-energy knowledge graph and initialize load carbon fingerprint features to obtain energy-carbon correlation mapping relationship and equipment-level carbon characteristic perception results; Based on the energy-carbon correlation mapping relationship and the device-level carbon characteristic perception results, a dynamic electric-carbon coupling model is constructed by integrating deep learning and thermodynamic mechanisms to obtain the instantaneous carbon emission intensity of each energy supply path. Based on the instantaneous carbon emission intensity of each energy supply path and combined with real-time load demand, the carbon difference gradient between multi-energy supply schemes is analyzed, the correlation between load, carbon potential and equipment energy efficiency is explored, and a set of alternative energy supply paths guided by carbon emission reduction gains is constructed. Based on the set of alternative power supply paths, a multi-objective convex optimization model is constructed, which is then transformed into executable control commands and executed through a cloud-edge collaborative architecture to obtain the control execution results. Based on the results of regulation implementation, the effectiveness of regulation is evaluated and anomalies are diagnosed. Combined with user load characteristics, the power marketing strategy is iterated to form a closed-loop optimization.

[0070] Alternatively, when the processor executes the computer program, it implements the functions of each module in the above system, for example: The graph fingerprint module 1 is used to acquire terminal-side energy node, energy flow, carbon flow and load operation data, construct an electricity-carbon-energy knowledge graph and initialize load carbon fingerprint features, and obtain energy-carbon correlation mapping relationship and equipment-level carbon characteristic perception results; Coupled modeling module 2 is used to construct a dynamic electric carbon coupling model based on the energy-carbon correlation mapping relationship and the device-level carbon characteristic perception results, and integrate deep learning and thermodynamic mechanisms to obtain the instantaneous carbon emission intensity of each energy supply path. The path construction module 3 is used to analyze the carbon difference gradient between multiple energy supply schemes based on the instantaneous carbon emission intensity of each energy supply path and combined with real-time load demand, explore the correlation between load, carbon potential and equipment energy efficiency, and construct a set of alternative energy supply paths guided by carbon emission reduction gains. The regulation and execution module 4 is used to construct a multi-objective convex optimization model based on the set of alternative power supply paths, convert it into executable control commands and execute them through the cloud-edge collaborative architecture to obtain the regulation and execution results; Evaluation Iteration Module 5 is used to evaluate the effectiveness of regulation and diagnose anomalies based on the results of regulation execution, and iterate the power marketing strategy in combination with user load characteristics to form a closed-loop optimization.

[0071] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the mobile terminal.

[0072] For example, the computer program can be divided into a graph fingerprint module 1, a coupling modeling module 2, a path construction module 3, a control execution module 4, and an evaluation and iteration module 5; The specific functions of each module are as follows: The graph fingerprint module 1 is used to acquire terminal-side energy node, energy flow, carbon flow and load operation data, construct an electricity-carbon-energy knowledge graph and initialize load carbon fingerprint features, and obtain energy-carbon correlation mapping relationship and equipment-level carbon characteristic perception results; Coupled modeling module 2 is used to construct a dynamic electric carbon coupling model based on the energy-carbon correlation mapping relationship and the device-level carbon characteristic perception results, and integrate deep learning and thermodynamic mechanisms to obtain the instantaneous carbon emission intensity of each energy supply path. The path construction module 3 is used to analyze the carbon difference gradient between multiple energy supply schemes based on the instantaneous carbon emission intensity of each energy supply path and combined with real-time load demand, explore the correlation between load, carbon potential and equipment energy efficiency, and construct a set of alternative energy supply paths guided by carbon emission reduction gains. The regulation and execution module 4 is used to construct a multi-objective convex optimization model based on the set of alternative power supply paths, convert it into executable control commands and execute them through the cloud-edge collaborative architecture to obtain the regulation and execution results; Evaluation Iteration Module 5 is used to evaluate the effectiveness of regulation and diagnose anomalies based on the results of regulation execution, and iterate the power marketing strategy in combination with user load characteristics to form a closed-loop optimization.

[0073] The mobile terminal can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The mobile terminal may include, but is not limited to, a processor and memory.

[0074] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the mobile terminal, connecting various parts of the mobile terminal via various interfaces and lines.

[0075] The memory can be used to store the computer program and / or module. The processor implements various functions of the mobile terminal by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0076] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function (such as sound playback, image playback, etc.); the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). Furthermore, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, SmartMediaCards (SMC), Secure Digital (SD) cards, FlashCards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0077] Example 4 The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for online collaborative optimization and active response of multi-energy complementary loads.

[0078] If the modules / units integrated in the mobile terminal are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0079] Based on this understanding, all or part of the processes in the above method can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the above-described aggregated reinforcement learning resource scheduling method. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate form.

[0080] The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0081] It should be noted that the content contained in the computer-readable medium may be appropriately added to or subtracted from the content as required by the legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium may not include electrical carrier signals and telecommunication signals.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for online collaborative optimization and active response of multi-energy complementary loads, characterized in that, include: Acquire terminal-side energy node, energy flow, carbon flow and load operation data, construct an electricity-carbon-energy knowledge graph and initialize load carbon fingerprint features to obtain energy-carbon correlation mapping relationship and equipment-level carbon characteristic perception results; Based on the energy-carbon correlation mapping relationship and the device-level carbon characteristic perception results, a dynamic electric-carbon coupling model is constructed by integrating deep learning and thermodynamic mechanisms to obtain the instantaneous carbon emission intensity of each energy supply path. Based on the instantaneous carbon emission intensity of each energy supply path and combined with real-time load demand, the carbon difference gradient between multi-energy supply schemes is analyzed, the correlation between load, carbon potential and equipment energy efficiency is explored, and a set of alternative energy supply paths guided by carbon emission reduction gains is constructed. Based on the set of alternative power supply paths, a multi-objective convex optimization model is constructed, which is then transformed into executable control commands and executed through a cloud-edge collaborative architecture to obtain the control execution results. Based on the results of regulation and control, the effectiveness of regulation and control is evaluated and anomalies are diagnosed. Combined with user load characteristics, the power marketing strategy is iterated to form a closed-loop optimization.

2. The method for online collaborative optimization and active response of multi-energy complementary loads according to claim 1, characterized in that, The specific process of constructing a knowledge graph of electricity, carbon, and energy, initializing load carbon fingerprint features, and obtaining the energy-carbon correlation mapping relationship and device-level carbon characteristic perception results includes: By introducing knowledge graph technology and non-intrusive load monitoring algorithms, an explicit characterization of the implicit carbon footprint at the terminal side is obtained; When constructing a knowledge graph, define the graph. ,in Represents an energy node; Represents a connection of energy flow or carbon flow. Represents the mapping rules between nodes; the Laplacian feature mapping formula is introduced as follows: ,in, The adjacency matrix is ​​an energy-carbon relational matrix. The degree matrix is ​​used to identify the bottleneck node with the highest carbon emission intensity in the energy network by calculating the eigenvalues ​​of the matrix. During load carbon fingerprint initialization, the NILM formula based on instantaneous power decomposition is used, assuming the bus current signal is... Harmonic feature vectors of each device are extracted using discrete Fourier transform. Using the residual minimization decomposition formula: in, For the first The on / off status of this type of equipment; n is the equipment serial number; n is the total number of all electrical devices in the system; For harmonic characteristic mapping terms; The sum of squared residuals is used to measure the magnitude of the current fitting error. To perform current superposition for all load devices; The bus-side mixed carbon flow is separated into individual device carbon traces to complete the full perception of micro-load carbon characteristics. Based on the full perception of micro-load carbon characteristics, the energy-carbon correlation mapping relationship and device-level carbon characteristic perception results are obtained.

3. The method for online collaborative optimization and active response of multi-energy complementary loads according to claim 1, characterized in that, The specific process of constructing a dynamic electrocarbon coupling model includes: A dynamic electro-carbon coupling model for energy conversion is constructed, and a bidirectional long short-term memory network is used to process time-series energy data to capture the impact of historical load inertia and future carbon potential prediction on the current state. For electro-thermal conversion equipment, a thermodynamic exergy efficiency-corrected carbon intensity model is introduced, with the following formulas: in, for The equivalent carbon emission intensity of the electro-thermal conversion equipment at any given time; for The marginal emission factor of the upstream power grid reflects the proportion of clean energy on the grid side in real time; This is the efficiency coefficient for electrothermal conversion; Carnot factor, Absolute temperature for environmental reference; This is the absolute temperature at the heat output terminal; for Random noise disturbance term of carbon emissions from the power grid at any given time; Introducing a weighted formula for attention mechanisms: in, The weight representing the contribution of carbon intensity fluctuations at different times to the prediction target. and They are respectively The hidden layer state vector and global context vector at each time step represent the feature abstraction extracted by the deep learning model; It is a natural exponential function, which maps attention scores to non-negative values; This is an attention scoring function used to calculate the degree of association and matching between a single temporal hidden state and the global context; For any number The hidden layer state vector at each time step is obtained by iterating through all time-series nodes; ∑ is the summation operation, which accumulates the exponential scores of all time-series nodes. A dynamic electrocarbon coupling model is constructed by deeply coupling a data-driven deep learning model with a thermodynamic physical mechanism model.

4. The method for online collaborative optimization and active response of multi-energy complementary loads according to claim 1, characterized in that, The process involves exploring the correlation between load, carbon potential, and equipment energy efficiency to construct a set of alternative energy supply paths guided by carbon emission reduction gains. Using the Apriori association rule algorithm, the correlation between load, carbon potential, and equipment energy efficiency is further explored, and the carbon emission gradient for different energy path combinations is calculated in real time. The specific process includes: Distributed energy carbon differential analysis introduces a carbon emission comparison discriminant formula: in, This represents the difference in instantaneous carbon emission reduction gradient. This represents the current real-time total load demand power; For the first Real-time output power of the internal distributed energy source; For the first Carbon emission coefficient per unit of electricity generated from a particular energy form; For the first The transmission loss correction factor for various energy supply paths; Multi-energy complementary path construction, introducing information gain decision criteria: In the formula, It is the entropy value of the system's original carbon intensity state, representing a measure of the system's uncertainty before regulation; In order to perform specific regulatory actions Under the influence of this effect, the proportion and weight of subset samples to the total sample; The number of samples in the v-th state subset after being divided by the control action A; The total number of samples in the total sample set S; Using carbon difference as the target variable, the contribution of each control variable to carbon emission reduction is evaluated, and a multi-level decision search tree is automatically constructed, with carbon emission reduction gain as the primary search driver.

5. The method for online collaborative optimization and active response of multi-energy complementary loads according to claim 1, characterized in that, Based on a set of alternative power supply paths, a multi-objective convex optimization model is constructed, which is then transformed into executable control commands and executed through a cloud-edge collaborative architecture to obtain the control execution results. The specific process includes: A second-order cone programming algorithm is introduced to transform the nonlinear power flow constraint into a convex constraint, ensuring the online real-time performance and convergence of the control action; Construct a multi-objective optimization function, as shown in the following formula: in, These are the weighting coefficients for each objective. Let t be the time-of-use electricity price. For the predicted real-time carbon emission intensity, This represents the system's adjustment increment for grid frequency offset; The summation is performed over time, where t is the scheduling time and T is the total scheduling duration. The economic cost weighting coefficient represents the priority of the electricity cost control target in the overall optimization. Let t be the electrical power exchanged between the system and the upper-level power grid at time t, with positive values ​​representing purchased electricity and negative values ​​representing sold electricity; The weighting coefficient for low-carbon emission reduction is used to adjust the priority of carbon emission targets in the overall optimization. This is the power grid frequency regulation weighting coefficient, which characterizes the priority of the system in participating in the power grid frequency stability regulation objective; The frequency adjustment increment squared term is used to weaken the influence of positive and negative deviations by adopting a squared form, thereby achieving smooth suppression of frequency fluctuations. We introduce the Lagrange multiplier method to construct the augmented objective function, as shown below: in, To augment the Lagrangian function, the constrained optimization problem is transformed into an unconstrained optimization problem; Let be the Lagrange multiplier corresponding to the j-th node, representing the th node. The marginal carbon cost and power deviation cost of each additional unit of load at each node; Indicates the first The network loss power of each node; transforming the constrained optimization problem into an unconstrained optimization problem; The objective function for system comprehensive optimization includes economic cost, carbon emission cost, and frequency regulation cost; To sum over all M power grid / energy network nodes in the system sequentially; The real-time power output of the distributed power source at node j; The equivalent power of charging and discharging for the energy storage device at node j, where discharging is positive and charging is negative; The real-time power output of the distributed power source at node j; Adopting a cloud-edge collaborative architecture, the cloud formulates global strategies at the minute level, and the edge controller performs frequency tracking at the second level; for flexible loads, pulse width modulation technology is used to fine-tune the input power, so that the load carbon footprint curve is fitted with the photovoltaic or wind power output curve in the region in real time, and active carbon footprint offsetting is carried out.

6. The method for online collaborative optimization and active response of multi-energy complementary loads according to claim 1, characterized in that, Based on the results of regulation implementation, the efficiency of regulation is evaluated and anomalies are diagnosed. The specific process of iterating electricity marketing strategies in conjunction with user load characteristics to form a closed-loop optimization includes: The actual carbon reduction after each load adjustment Response speed Incremental economic costs and grid carbon potential prediction bias Composition of high-dimensional feature vectors Utilizing abnormal scoring Calculate the anomaly score for each control sample. When s(x,n) approaches 1, it indicates that the control action has caused an abnormal deviation. The cause of the anomaly is located by the correlation analysis backtracking algorithm. At the marketing closed-loop level, a collaborative filtering recommendation model is introduced. Based on users' historical carbon reduction performance, low-carbon benchmark enterprises with similar load characteristics are identified through cosine similarity. Optimized electricity marketing packages are then pushed to users with poor performance, forming a closed loop of model correction and marketing strategy optimization.

7. The method for online collaborative optimization and active response of multi-energy complementary loads according to claim 6, characterized in that, The abnormal score The formula is as follows: in, The average path height of the sample across all isolation trees. The average constant of the search path; The cosine similarity formula is as follows: in, This represents the rating index for user u in the i-th control dimension; Let v be the quantitative scoring index for user v under the i-th control dimension.

8. A system for online collaborative optimization and active response of multi-energy complementary loads, characterized in that, include: The graph fingerprint module is used to acquire terminal-side energy node, energy flow, carbon flow and load operation data, construct an electricity-carbon-energy knowledge graph and initialize load carbon fingerprint features, and obtain energy-carbon correlation mapping relationship and equipment-level carbon characteristic perception results; The coupling modeling module is used to construct a dynamic electric carbon coupling model based on the energy-carbon correlation mapping relationship and the device-level carbon characteristic perception results, and integrate deep learning and thermodynamic mechanisms to obtain the instantaneous carbon emission intensity of each energy supply path. The path construction module is used to analyze the carbon difference gradient between multiple energy supply schemes based on the instantaneous carbon emission intensity of each energy supply path and combined with real-time load demand, explore the correlation between load, carbon potential and equipment energy efficiency, and construct a set of alternative energy supply paths guided by carbon emission reduction gains. The regulation and execution module is used to construct a multi-objective convex optimization model based on the set of alternative power supply paths, transform it into executable control commands, and execute them through a cloud-edge collaborative architecture to obtain the regulation and execution results. The evaluation and iteration module is used to evaluate the effectiveness of regulation and diagnose anomalies based on the results of regulation execution, and to iterate the power marketing strategy in combination with user load characteristics to form a closed-loop optimization.

9. A mobile terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for online collaborative optimization and active response of multi-energy complementary loads as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for online collaborative optimization and active response of multi-energy complementary loads as described in any one of claims 1-7.

Images