Low-carbon demand response potential prediction method based on generative mobile interactive neural network

Through the method based on the generative mobile interactive neural network, combined with energy flow and carbon emission flow model, the accurate prediction of the response potential of the low-carbon demand of comprehensive energy systems is achieved, and the poor overfitting and generalization capabilities in the existing technology are solved. The problem improves the accuracy of prediction and the low-carbon scheduling potential of the system.

CN118917641AInactive Publication Date: 2024-11-08STATE GRID ZHEJIANG ELECTRIC POWER COMPANY TAIZHOU POWER SUPPLY

Patent Information

Application Number
CN202411412061.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-11
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art has problems of overfitting and generalization capabilities when predicting the response potential of low-carbon demand in integrated energy systems, making it difficult to accurately model and predict the response behavior of energy systems in low-carbon operating mode.

Method used

Using a low-carbon demand response potential prediction method based on a generative mobile interactive neural network, an optimal economic scheduling model for energy flow is established by obtaining the historical electricity consumption data of energy users and the operation data of the integrated energy system, combining the carbon emission flow model and storage The carbon state of the energy system is proposed, and the demand response mechanism of the carbon perspective is proposed, and the prediction is made using the generated mobile interactive neural network framework.

Benefits of technology

Achieve the prediction of the accurate demand response potential of the low-carbon operation mode of the integrated energy system, fully tap the low-carbon scheduling potential of both energy storage and energy release, solve the problem of fitting the existing method, and improve the accuracy of the prediction and Generalization ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917641A_ABST
    Figure CN118917641A_ABST
Patent Text Reader

Abstract

The invention provides a low-carbon demand response potential prediction method based on a generative mobile interactive neural network, and the method comprises the steps: considering the influence of the charging and discharging states of an energy storage system on the carbon emission flow of the system, and the dynamic relation between the carbon emission of the energy storage system and the energy of a whole scheduling period; combining the carbon state with a traditional carbon emission flow model, and proposing an extended carbon emission flow model considering the energy storage system; a demand response mechanism of a carbon perspective is provided, the cost for reducing carbon emission and saving is used as an excitation signal, the relation between the energy consumption behavior and the load node carbon potential is established, and low-carbon operation of the comprehensive energy system is achieved; and finally, proposing a short-term demand response prediction framework based on the generated mobile interactive neural network according to features provided by the model, and realizing a function of predicting the low-carbon demand response potential. According to the method, the potential of collaborative low-carbon scheduling of energy storage and energy release of the integrated energy system is fully excavated, and the fitting problem of an existing demand response potential prediction method is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of power system and optimization algorithm modeling, and specifically relates to a method for predicting low-carbon demand response potential based on a generative mobile interactive neural network. Background Art

[0002] The clean and low-carbon transformation of energy systems is the primary focus of curbing global greenhouse gas emissions. However, the volatility and uncertainty of intermittent renewable energy generation have brought new challenges to the low-carbon operation of energy systems. In the integrated energy system (IES), the coupling characteristics and mutual conversion capabilities of electricity, heat and natural gas provide an effective way to enhance the adaptability of intermittent renewable energy generation. Fully tapping the low-carbon potential of IES in production, transmission, transformation, distribution and storage is an important means to popularize the uncertainty of intermittent renewable energy generation units and achieve low-carbon scheduling. Under demand response events, the response behavior of the integrated energy system is affected by multiple coupling factors such as the incentive signal, the coupling characteristics of electricity, heat and natural gas, and it is difficult to accurately model and predict it. Therefore, it is of great significance to reasonably simulate the response behavior of the integrated energy system under the low-carbon operation mode and accurately predict the available demand response potential.

[0003] Neural network algorithms are currently a hot topic in the research of solving the problem of short-term demand response potential prediction. Existing methods based on search improvement process-chaos optimization-particle swarm optimization-elite retention strategy, as well as improved trigeneration strategy based on quadratic scaling, can find the optimal value of network parameters through genetic coding and evolutionary operations; the sparrow algorithm combining ant colony algorithm and chaotic mapping can introduce the heuristic search strategy of nature into neural network optimization; however, the above existing technologies still have the problem of overfitting and poor generalization ability. Summary of the invention

[0004] The technical problem to be solved by the present invention is: to provide a method for predicting the low-carbonization demand response potential based on a generative mobile interactive neural network, which is used to predict the low-carbonization demand response potential.

[0005] The technical solution adopted by the present invention to solve the above technical problems is: a method for predicting the low-carbon demand response potential based on a generative mobile interactive neural network, comprising the following steps: S1: Obtain historical electricity consumption data and demand response data of energy users, as well as the operation data of the integrated energy system IES; S2: Optimize the distribution of energy flow based on the data in step S1 and establish a plan for the optimal economic dispatch of energy flow of the integrated energy system IES; S3: Based on the results of the optimal economic dispatch of energy flow, an extended carbon emission flow model ECEF is established to map the relationship between energy flow and carbon emission flow, calculate the carbon emission flow and obtain the node carbon intensity NCI of the load node, implement a low-carbon demand response mechanism guided by the node carbon intensity NCI, and obtain the carbon emission reduction of energy users; S4: A demand response potential prediction framework based on a generative mobile interactive neural network GMINN is proposed, the carbon emission reduction in step S3 is used as a feature input framework, and the optimal demand response potential prediction model is output.

[0006] According to the above scheme, in step S1, the operating data includes the operating data of the power system, the thermal system and the natural gas system; in step S2, the energy flow includes electricity, thermal energy flow and natural gas.

[0007] Furthermore, in step S2, it is assumed that , and are the operating costs of the power system, thermal system and natural gas system in time period t, is the carbon trading cost within the dispatch scope; the objective function of the optimal economic dispatch of energy flow is to minimize the total operating cost F of the integrated energy system IES: ; The specific calculation steps are as follows: S21: Assume that the transportation and storage cost of CO2 is , the excess cost of carbon quota is , the CO2 purchase cost of P2G equipment is ; , and They are the cost coefficients for CO2 transportation and storage, CO2 trading, and CO2 purchase; is the CO2 captured by the CCS from coal-fired unit k during period t in scenario n; is the CO2 consumed by P2G device u in time period t; is the total carbon emissions of the integrated energy system IES in period t; , , and are the carbon emission allowable coefficients of coal-fired unit g, coal-fired unit k with carbon capture and storage device, combined heat and power unit c, and gas turbine r, respectively; is the carbon emission allowable coefficient of the external power grid o; , , , are the power outputs of coal-fired unit g, carbon capture and storage unit k, combined heat and power unit c and gas turbine r in time period t, respectively; is the amount of electricity purchased from the external grid o; is the heat output of CHP unit c in period t; is the conversion efficiency of P2G device u; is the energy consumption of P2G device u; K, U, G, C, R and O are the collections of coal-fired units including carbon capture and storage devices, P2G devices, coal-fired units, CHP units, gas turbines and external grids, respectively; T is the set of time periods; Carbon trading costs are calculated on a daily basis for: ; S22: Set and are the operating cost coefficients of the coal-fired unit g and turbine r, respectively; and are the charging cost coefficient and discharging cost coefficient of the energy storage system e respectively; and are the penalty factors for load shedding and wind curtailment, respectively; is the electricity price during period t; is the load reduction power at node i in period t; is the curtailed wind power at node w in period t; is the predicted wind power in period t; and are the charging active power and discharging active power of the energy storage system e in period t respectively; is the wind power that can be accommodated in period t; I is the set of all nodes; W is the set of wind turbines; E is the collection of energy storage systems; Then the operating cost of the power system in time period t is for: , ; S23: Set is the penalty factor for heat load reduction; is the heat load reduction at node z in time period t; Z is the set of nodes in the heat network; then the operating cost of the thermal system in time period t for: ; S24: Set is the gas cost of source s; is the penalty factor for natural gas load reduction; M is the set of natural gas load reduction points; is the gas load reduction at node m in time period t, is the amount of natural gas exported from source s in time period t; then the operating cost of the natural gas system in time period t for: .

[0008] According to the above scheme, in step S2, the constraints for optimal economic dispatch of energy flow include power system operation constraints, thermal system operation constraints, natural gas system operation constraints and energy coupling device constraints; The specific calculation steps are as follows: S25: Power system operation constraints include power balance constraints, operation constraints of carbon capture and storage devices of coal-fired power generation units, branch power flow constraints and unit start-stop constraints; set up represents the device connected to node i; is the power consumption of pump q in time t; is the net power output of coal-fired unit k equipped with carbon capture and storage equipment in period t; is the input electrical power of P2G device u in time period t; and are the powers flowing through branch (i, j) and branch (j, i) in period t, respectively; is the predicted power load at node i in time period t; is the load reduction power at node i in period t in scenario n; the power balance constraint is: ; set up is the net power output of coal-fired unit k equipped with carbon capture and storage equipment in period t; is the fixed power consumption of coal-fired unit k with CCS during time period t; is the operating power consumption of coal-fired unit k with carbon capture and storage during period t; is the energy consumption required to capture each unit of CO2 in the CCS device; and are the carbon capture efficiency and upper limit of the carbon capture and storage device, respectively; is the total carbon emission of coal-fired unit k using carbon capture and storage devices in period t; the operation constraint of the carbon capture and storage device of the coal-fired power generation unit is: ; set up is the power flowing through branch (i, j) in period t; , are the phase angles of node i and node j in period t respectively; is the reactance of branch (i, j); is the power upper limit of branch (i,j), and are the upper and lower limits of the phase angle of node i; L is the branch set; the branch power flow constraint is: ; set up / , / , / , / are the upper and lower limits of the output of coal-fired unit g, coal-fired unit k, gas turbine r, and cogeneration unit c respectively; the start and stop constraints of the units are: ; S26: Thermal system operation constraints include thermal station constraints and heat exchange station constraints; thermal station constraints are as follows: set up is the heat capacity of the liquid in the pipe network of the thermal power station; and are the mass flow rates of the liquids of CHP unit c and pump q, respectively; and are the return temperature and supply temperature of CHP unit c in period t, respectively; and are the supply pressure and return pressure of pump q in period t, respectively; is the working efficiency of pump q; is the density of the liquid d in the pipe; , ; set up is the set of pipelines starting from node z, is the set of pipelines ending at node z; and are the outlet temperatures of the water supply pipe and the return pipe bh in time period t respectively; and are the inlet temperatures of the water supply pipe and the return pipe bh in the time period t respectively; and are the supply and return temperatures of the thermal network node z in time period t, respectively; is the ambient temperature during period t; is the thermal conductivity of the pipe bh, is the length of pipe bh; is the mass flow rate of the return pipe bh in time period t; , ; set up is the predicted value of the heat demand response potential of node z within time t; V is the set of nodes z; is the mass flow rate of the liquid at the thermal network node z; is the penalty factor for the heat demand response potential of node z within time t; the heat exchange station constraint is: ; S27: In the deterministic scenario n, the natural gas system operation constraints include node flow balance constraints and pipeline operation constraints; set up Indicates connection to natural gas network node a; is the output of natural gas source s in time t; is the gas production of P2G device u in period t; is the gas flow rate of natural gas pipeline network ab in time t; is the gas load reduction of node a in period t; is the predicted gas load of node a in time t; is the gas consumption of the cogeneration unit c in time period t; is the gas consumption of turbine r in period t; S is the setting of gas source; A is the set of gas network nodes; the node flow balance constraint is: ; set up is the physical coefficient of gas pipeline ab; and are the pressures at nodes a and b in period t, respectively; and are the upper and lower limits of the gas flow rate in pipelines ab, respectively; / and / are the upper and lower limits of the pressure at nodes a and b respectively; the pipeline operation constraints are: , , , ; S28: Energy coupling device constraints include P2G operation constraints, gas turbine operation constraints and combined heat and power operation constraints; set up is the efficiency of P2G device u; is the amount of CO2 required per unit of electrical power consumed by P2G device u; is the carbon emission of the P2G equipment in period t; the P2G operation limit is: ; set up is the efficiency of the gas turbine r; is the natural gas input to the gas turbine r; the gas turbine operating constraints are: ; set up is the efficiency of the CHP unit c; the CHP operation limit is: .

[0009] According to the above scheme, in the step S3, based on the carbon emission flow theory, it is assumed that the CO2 generated on the source side is transferred to the load side through the branch, and a virtual carbon flow accompanies the branch flow; an extended carbon emission flow model ECEF is established by establishing a network carbon emission flow model, a CEF model of the energy coupling device and a CEF model of the energy storage device.

[0010] Furthermore, in step S3, the specific steps of establishing the network carbon emission flow model are: set up is the carbon emissions flowing through a branch or node; the carbon emissions per unit time passing through a branch or node is the carbon emissions flow rate CEFR: ; Carbon intensity CI represents the carbon emissions per unit of energy; Carbon intensity CI includes power generation carbon intensity GCI, branch carbon intensity BCI, gateway carbon intensity PCI and node carbon intensity NCI; , and are the carbon emission intensities of node n, branch i and unit g in period t respectively; is the set of branches that inject power into node n; Indicates that unit g is connected to node n; is the power flowing through branch i in period t; is the power output of unit g in period t; the node carbon intensity NCI represents the average carbon emission per unit energy injected into the node, which is equal to the ratio of the total carbon emission intensity to the total energy injected into the node: ; The inflow at node n includes the inflow from the branch connected to node n and the inflow from the generator connected to node n; when node n is not connected to a generator, the node carbon intensity NCI is: ; set up is the carbon emission of node i in time period t; is the load power at node i; is the time interval; represents the node carbon intensity of node i in period t; the carbon emission of node i in time period t is: .

[0011] Furthermore, in step S3, the specific steps of establishing the CEF model of the energy coupling device are: The energy coupling device includes a single-input single-output SISO energy coupling device and a single-input multiple-output SIMO energy coupling device; For a single-input single-output SISO energy coupling device, and are the input PCI and output PCI of the energy coupling device p in period t, respectively; and are the input and output powers of the energy coupling device p in period t respectively; according to the carbon conservation principle, all carbon emissions related to the input energy are allocated to the output energy, the total carbon emission flow CEFR of the input port is equal to the total carbon emission flow CEFR of the output port, and the carbon intensity PCI of the input and output ports is expressed as: ; Assume that the conversion efficiency of the single-input single-output SISO energy coupling device is , then the relationship between energy input and output is: , ; For single-input multiple-output SIMO energy coupling devices, according to the principle of carbon conservation, all carbon emissions related to the input energy are allocated to the output port; From the perspective of energy conversion, for back pressure CHP, and are the electrical output and thermal output of the CHP unit c in time period t, respectively; is the natural gas input to CHP unit c during time period t; and are the gas-to-electricity conversion efficiency and gas-to-heat conversion efficiency of the cogeneration unit c in time period t, respectively; the output power and heat energy levels are proportional to the input natural gas energy, expressed as: ; From the perspective of carbon emissions, and are the threshold carbon intensity PCI of the electrical output port and the thermal output port of the cogeneration unit c in time period t, respectively; is the threshold carbon intensity PCI of the input port of CHP unit c in time period t; the total carbon emission flow CEFR of the input port is equal to the total carbon emission flow CEFR of the output port, and the threshold carbon intensity PCI of the input port and the output port is expressed as: ; Assume that the PCI of the electrical output port and the thermal output port is inversely proportional to the conversion efficiency: .

[0012] Furthermore, in step S3, the specific steps of establishing the CEF model of the energy storage device are: The carbon state of the energy storage system is used to characterize the relationship between the energy and carbon emissions of the energy storage system. is the carbon emissions flowing into the energy storage system e in period t; is the charging power of the energy storage system e in period t; is the NCI of the node where the energy storage system e is located in period t; when the energy storage system is charged, carbon emissions flow into the energy storage system along with the electric energy, as shown in the following formula: ; set up is the carbon emissions released by the energy storage system e in period t; is the discharge power of the energy storage system e in period t; is the GCI of the energy storage system e in period t; is the discharge efficiency of the energy storage system e; is the carbon state of the energy storage system e in the t-1 period; when the energy storage system is discharged, carbon emissions are released from the energy storage system along with the electrical energy, as shown in the following formula: ; set up , are the available capacity of the energy storage system e in time periods t and t-1; the carbon state of the energy storage system Defined as: .

[0013] Furthermore, in step S3, it is assumed that Daily carbon emission reduction for power users, and are the power reduction and increase of the load connected to node i during time period t, , are binary variables for load increase and load decrease states, is the load proportion in the power system that can participate in low-carbon demand response; the low-carbon demand response mechanism is: , , , .

[0014] According to the above scheme, in step S4, set is the target sequence loaded over a period of time; input prediction feature sequence , the total time length is T, the total number of features is d, each of which Represents an input feature; Represents the time series of the kth feature, using represents the feature vector at time t; the corresponding set of demand response potential prediction values ​​is The specific steps of the prediction process of the demand response potential prediction framework based on the generative mobile interactive neural network GMINN are: S41: Use data generation components to expand the training data set; feature sequence After data preprocessing, a training sample set is formed ; Enhance the training sample set to generate an enhanced training sample set ; Using loss function The average value over the data distribution P represents the expected risk to measure the predictive value With actual value The difference between the expected risk and the expected risk is defined as follows: ; set up represents the empirical distribution, and the training data set Each sample in In line with the empirical distribution ; Use empirical risk to approximate expected risk. The empirical risk is defined as follows: ; A hybrid decision-making method based on the principle of minimizing neighbor risk is adopted; using prior knowledge to construct neighbor samples from the training set distribution and obtain a data set ; All data in follows the neighbor distribution , the empirical adjacent risk is: ; set up and ; is a hyperparameter that controls the number of linear interpolation times, ; and are two training samples randomly selected from the training data set; are new samples generated; they are mixed from the neighborhood distribution using linear interpolation A sample of: ; The lower the value of , the lower the degree of linear interpolation; when When it is equal to 1, the distribution becomes uniform; S42: Use genetic algorithm GA to optimize hyperparameters; S43: Moving average filter MAF is selected as the built-in decomposition component of the framework to decompose the input feature sequence into a trend sequence that captures the overall trend and a residual sequence containing fine details; S44: Use sample convolution and interactive learning SCINet to model trend series and residual series respectively, extract time dependencies at multiple time resolutions, and obtain the demand response potential prediction value of the sample; S45: Searching for the optimal network structure with the help of genetic algorithm GA; when the termination condition is met, outputting the optimal demand response potential prediction model.

[0015] The beneficial effects of the present invention are: 1. A method for predicting the low-carbon demand response potential based on a generative mobile interactive neural network in the present invention first considers the impact of the charging and discharging state of the energy storage system on the system carbon emission flow, as well as the dynamic relationship between the carbon emission of the energy storage system and the energy of the entire scheduling cycle; then combines the carbon state (SOCB) with the traditional carbon emission flow (CEF) model, and proposes an extended carbon emission flow model (ECEF) considering the energy storage system (ESS); and proposes a "carbon perspective" demand response (DR) mechanism, which uses the cost saved by reducing carbon emissions as an incentive signal, establishes the relationship between energy consumption behavior and the carbon potential of load nodes, and realizes the low-carbon operation of the integrated energy system (IES); finally, according to the characteristics provided by the established model, a short-term demand response prediction framework based on a generative mobile interactive neural network is proposed, which realizes the function of predicting the low-carbon demand response potential.

[0016] 2. The present invention fully explores the potential of coordinated low-carbon scheduling of energy storage and release in the integrated energy system, and solves the fitting problem of existing demand response potential prediction methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0019] See also Figure 1 , an embodiment of the present invention comprises the following steps: Step 1: Initialization, obtain the historical electricity consumption data, demand response data of industrial park load users, and the operating data of the power system, thermal system, and natural gas system in the integrated energy system.

[0020] Step 2: Establish economic dispatch based on energy flow. Optimize the distribution of electricity, natural gas and thermal energy flows, and formulate economically optimal dispatch plans for energy producers and energy users.

[0021] (1) Objective function The goal of low-carbon economic dispatch of energy systems is to minimize operating costs and carbon emission costs while taking into account the operating constraints of the power system, thermal system and natural gas system. The operating costs of cogeneration and gas turbines are integrated into the purchase cost of natural gas.

[0022]

[0023] Where F is the total operating cost of IES. , and are the operating costs of the electricity system, heating system and natural gas system in time period t, respectively. is the carbon emission cost within the dispatch scope.

[0024] 1) Carbon trading costs Carbon trading costs Usually calculated on a daily basis, including CO2 transportation and storage costs , Excess Cost of Carbon Quota and CO2 purchase cost of P2G equipment , the calculation formula is as follows

[0025] In the formula, , and They are the cost coefficients for CO2 transportation and storage, CO2 trading, and CO2 purchase. is the CO2 captured from coal-fired unit k by the carbon capture and storage facility in scenario n during period t. is the CO2 consumed by P2G device u in time period t. is the total carbon emissions of IES in period t. , , and They are the carbon emission allowable coefficients of coal-fired unit g, coal-fired unit k with carbon capture and storage device, cogeneration unit c and gas turbine r. is the carbon emission allowance coefficient of the external power grid o. , , , are the power outputs of coal-fired unit g, carbon capture and storage device coal-fired unit k, cogeneration unit c and gas turbine r in time period t. is the amount of electricity purchased from the external grid o. is the heat output of the CHP unit c in period t. is the conversion efficiency of P2G device u. is the energy consumption of P2G device u. K, U, G, C, R and O are the set of coal-fired units including carbon capture and storage device, P2G device, coal-fired unit, CHP unit, gas turbine and external grid respectively. T is the set of time periods.

[0026] 2) Power system operating costs

[0027]

[0028] In the formula, and are the operating cost coefficients of the coal-fired unit g and the turbine r respectively. is the charging / discharging cost coefficient of ESSe. and are the penalty factors for load shedding and wind curtailment, respectively. is the electricity price during period t. is the load reduction power at node i during period t. is the curtailed wind power at node w in period t. is the predicted wind power in period t. and are the charging active power and discharging active power of the energy storage system e in period t respectively. is the wind power that can be accommodated in time period t. I is the set of all nodes. W is the set of wind turbines. E is the set of energy storage systems.

[0029] 3) Thermal system operating costs

[0030] In the formula, is the penalty factor for heat load reduction. is the heat load reduction at node z during time period t. Z is the set of nodes in the heat network.

[0031] 4) Natural gas system operating costs

[0032] In the formula, is the gas cost of source s. is the penalty factor for natural gas load reduction. M is the set of natural gas load reduction points. is the gas load reduction at node m in time period t, is the amount of natural gas exported from source s during time period t.

[0033] (2) Constraints The constraints in the IES low-carbon economic dispatch model include: power system operation constraints, thermal system operation constraints, natural gas system operation constraints and energy coupling device constraints.

[0034] 1) Power system It mainly includes power balance constraints, operation constraints of carbon capture and storage devices of coal-fired power generation units, branch flow constraints and unit start and stop constraints.

[0035] Power balance constraints:

[0036] In the formula, Represents the device connected to node i. is the power consumption of pump q in time t. is the net power output of coal-fired unit k with carbon capture and storage in period t. is the input electrical power of P2G device u in time period t. and are the powers flowing through branch (i, j) and branch (j, i) in period t, respectively. is the predicted power load at node i in time period t. is the load reduction power at node i during period t in scenario n.

[0037] Operating constraints for coal-fired power plants with carbon capture and storage:

[0038] In the formula, is the net power output of coal-fired unit k with carbon capture and storage in period t. is the fixed power consumption of coal-fired unit k with carbon capture and storage during time period t. is the operating power consumption of coal-fired unit k with carbon capture and storage during period t. is the energy consumption required to capture each unit of CO2 in a carbon capture and storage device. and are the carbon capture efficiency and upper limit of the carbon capture and storage device, respectively. is the total carbon emissions of coal-fired unit k using carbon capture and storage equipment in period t.

[0039] Branch power flow constraints:

[0040] In the formula, is the power flowing through branch (i, j) in period t. , are the phase angles of node i and node j in period t respectively. is the reactance of branch (i,j). is the power upper limit of branch (i,j), and are the upper and lower limits of the phase angle of node i. L is the branch set.

[0041] Unit operating restrictions:

[0042] In the formula, / , / , / , / They are the upper and lower limits of the output of coal-fired unit g, coal-fired unit k, gas turbine r, and cogeneration unit c respectively.

[0043] 2) Heating system Thermal station restrictions:

[0044]

[0045] In the formula, It is the heat capacity of the liquid in the pipe network of the thermal power station. and are the mass flow rates of the liquid of CHP unit c and pump q respectively. and are respectively the return temperature and the supply temperature of the CHP unit c during period t. and are the supply pressure and return pressure of pump q in cycle t respectively. is the working efficiency of the pump q. is the density of the liquid d in the pipe.

[0046]

[0047]

[0048] In the formula, is the set of pipelines starting from node z, is the set of pipelines ending at node z. and are the outlet temperatures of the supply pipe and return pipe bh in time period t respectively. and are the inlet temperatures of the supply pipe and return pipe bh in time period t respectively. and are respectively the supply temperature and the return temperature of the thermal network node z in time period t. is the ambient temperature during period t. is the thermal conductivity of the pipe bh, is the length of pipe bh. is the mass flow rate of the return pipe bh during time period t.

[0049] Heat exchange station constraints:

[0050] In the formula, is the predicted value of the heat demand response potential of node z in time t. V is the set of nodes z; is the mass flow rate of the liquid at the thermal network node z; is the penalty factor for the thermal demand response potential of node z at time t.

[0051] 3) Natural gas system In the deterministic scenario n, the natural gas system constraints include node flow balance constraints and pipeline operation constraints.

[0052] Node traffic balance constraints:

[0053] In the formula, Indicates connection to natural gas network node a. is the production of natural gas source s in time t. is the gas production of P2G device u in period t. is the gas flow rate of natural gas pipeline network ab in time t. is the gas load reduction of node a in period t. is the predicted gas load of node a within time t. is the gas consumption of the cogeneration unit c in time period t. is the gas consumption of turbine r in period t. S is the setting of gas source. A is the set of gas network nodes.

[0054] Natural gas pipeline operation constraints:

[0055]

[0056]

[0057]

[0058] In the formula, is the physical coefficient of gas pipeline ab. and are the pressures at nodes a and b in period t, respectively. and They are the upper and lower limits of the gas flow in pipelines ab respectively. / and / are the upper / lower limits of the pressure at nodes a and b respectively.

[0059] 4) Energy coupling device constraints P2G operation restrictions:

[0060] In the formula, is the efficiency of P2G device u. is the amount of CO2 required per unit of electrical power consumed by P2G device u. is the carbon emission of the P2G device in period t.

[0061] Gas turbine operating constraints:

[0062] In the formula, is the efficiency of the gas turbine r. is the natural gas input to the gas turbine r.

[0063] Cogeneration operation restrictions:

[0064] In the formula, is the efficiency of the combined heat and power unit c.

[0065] Step 3: Propose a method for calculating carbon emission flow. Based on the results of the optimal economic dispatch of energy flow, an ECEF model is established to map the relationship between energy flow and carbon emission flow, and the NCI of the load node is obtained. Then, a low-carbon demand response guided by NCI is implemented to minimize carbon emissions on the demand side.

[0066] (1) ECEF model considering energy consumption Based on the carbon emission flow theory, it is assumed that the CO2 generated on the source side is transferred to the load side through the branch (line and transformer), and a virtual "carbon flow" accompanies the branch flow. The proposed "carbon flow" can intuitively indicate the flow direction of carbon emissions during system operation, providing a new perspective for the low-carbon economic dispatch of IES.

[0067] 1) Network carbon emission flow model Carbon emission flow: Carbon emission flow rate (CEFR) is used to characterize the amount of carbon emissions per unit time passing through a branch or node.

[0068]

[0069] in is the amount of carbon emissions flowing through a branch or node.

[0070] Carbon intensity: Carbon intensity (CI) is used to characterize the carbon emissions per unit of energy. CI is usually divided into generation carbon intensity (GCI), branch carbon intensity (BCI), gateway carbon intensity (PCI) and node carbon intensity (NCI). BCI represents the carbon intensity of energy flowing along a branch. PCI represents the carbon intensity associated with the input or output energy of an energy coupling device. NCI reflects the superposition effect of carbon emission intensity, that is, the carbon emission intensities of various branches connected to the same node should be aggregated. NCI represents the average carbon emissions per unit of energy injected into the node, which is equal to the ratio of total carbon emission intensity to the total energy injected into the node.

[0071]

[0072] In the formula, , and are the carbon emission intensities of node n, branch i and unit g in period t respectively. is the set of branches that inject power into node n. Indicates that unit g is connected to node n. is the power flowing through branch i during period t. is the power output of unit g in period t. The inflow at node n includes two components: one originating from the branch connected to node n and the other originating from the generator connected to node n. At the same time, the value of NCI depends only on the power and carbon emissions flowing into the node and is not affected by the power and carbon emissions flowing out of the node. Therefore, NCI is not affected by the load connection. When there is no generator connected to node n, the mathematical expression of NCI can be expressed by the formula as follows:

[0073] The carbon emissions of node i in time period t can be calculated by the following formula:

[0074] In the formula, is the carbon emission of node i in time period t; is the load power at node i; is the time interval. represents the node carbon intensity of node i in period t.

[0075] 2) CEF model of energy coupling device In the process of energy coupling equipment converting between different types of energy, carbon emissions will be transferred in different energy systems. The CEF model of energy coupling equipment can be used to analyze the carbon emission transfer characteristics in the energy conversion process. Energy coupling devices are divided into single-input single-output (SISO) devices (such as gas turbines, P2G) and single-input multiple-output (SIMO) devices (such as cogeneration), and their carbon emission flow models are established respectively.

[0076] SISO Energy Coupling Devices: According to the principle of carbon conservation, all carbon emissions related to the input energy of the SISO equipment are allocated to the output energy, that is, the total CEFR of the input port is equal to the total CEFR of the output port, which can be expressed as:

[0077] In the formula, and are the input PCI and output PCI of the energy coupling device p in period t, respectively. and are the input and output powers of the energy coupling device p in period t, respectively.

[0078] If the conversion efficiency p of the SISO energy coupling device is , then the relationship between energy input and output can be expressed as:

[0079]

[0080] SIMO Energy Coupling Devices: For SIMO energy coupling devices, the principle of carbon conservation still holds true, and all carbon emissions associated with the input energy need to be allocated to the output port. From the perspective of energy conversion, for a typical back-pressure CHP, the output electrical and thermal energy levels are proportional to the input natural gas energy, which can be expressed as

[0081] In the formula, and are the electrical output and thermal output of the CHP unit c in time period t, respectively. is the natural gas input to CHP unit c during time period t. and are the gas-to-electricity conversion efficiency and gas-to-heat conversion efficiency of the cogeneration unit c in time period t, respectively.

[0082] From the perspective of carbon emissions, the CEFR of the input port is equal to the sum of the CEFR of the output port. The PCI of the input port and the output port can be expressed as

[0083] In the formula, and are the PCIs of the electrical output port and the thermal output port of the cogeneration unit c in time period t. is the input port PCI of CHP unit c in time period t.

[0084] Assume that the PCI of the electrical output port and the thermal output port is inversely proportional to the conversion efficiency:

[0085] 3) CEF model of energy storage device The equipment in IES can be divided into three categories according to the way of energy utilization: energy release equipment, such as generators. Energy consumption equipment such as various loads, and energy coupling equipment such as P2G. The energy storage system has two states: energy release and energy consumption. When the energy storage system is in the energy consumption state, it is equivalent to a special load that can absorb part of the carbon emissions. In the energy release state, it is equivalent to a special power generation device that will release part of the carbon emissions. The diversity of ESS states increases the complexity of the related carbon emission flow model. For this reason, the concept of "carbon state" of electric energy storage system is proposed to characterize the relationship between energy and carbon emissions of energy storage system.

[0086] When the ESS is charged, carbon emissions flow into the ESS along with the electrical energy, as shown in the following equation:

[0087] in, is the carbon emissions flowing into ESSe in period t. is the charging power of ESSe in period t. is the NCI of the node where ESSe is located in period t.

[0088] When the ESS discharges, carbon emissions are released from the ESS along with the electrical energy, as shown in the following equation:

[0089] In the formula, is the carbon emission released by ESSe in period t. is the discharge power of ESSe in period t. is the GCI of ESSe in period t. is the discharge efficiency of ESSe. is the SOCB of ESSe in period t-1. The SOCB of ESS is defined as

[0090] In the formula, , are the available capacities of ESSe in time periods t and t-1, respectively.

[0091] (2) Low-carbon demand response mechanism

[0092]

[0093]

[0094]

[0095] In the formula, Daily carbon emission reduction for power users, and are the power reduction and increase of the load connected to node i during time period t, , are binary variables for load increase and load decrease states, is the proportion of load in the power system that can participate in low-carbon demand response.

[0096] Step 4: A short-term demand response forecasting framework based on generative mobile interactive neural networks is proposed to achieve the prediction of low-carbon demand response potential according to the features selected for building the model.

[0097] First, let’s define the mathematical symbols used in the following sections. is the target sequence of load over a period of time. Given a prediction feature sequence , the total time length is T, the total number of features is d, each of which Represents an input feature. To represent the time series of the kth feature, use to represent the feature vector at time t.

[0098] Obtained through the demand response potential prediction framework Assume that the corresponding set of predicted values ​​of power load is given by To illustrate the problem definition, the present invention uses the first sample As an example, its definition is as follows:

[0099] In the formula, is the predicted value, It is a mapping relationship.

[0100] This paper proposes a high-performance deep learning network framework, Generative and Mobile Interactive Neural Network (GMINN). The process of GMINN is as follows: (1) Use the data generation component to expand the training dataset.

[0101] The data generation component can transform and expand the original data set using various techniques to increase the data volume, improve the data quality, and enhance the generalization ability of the model. Model performance based on training data sets with short statistical periods is usually low and only a small amount of data is available. Generate new data by linearly interpolating the original data, increase the data volume and diversity, and improve the generalization ability and robustness of the prediction model.

[0102] Neural networks usually use a loss function To measure the predicted value and actual value The expected risk is the average of the loss function over the data distribution P:

[0103] However, the data distribution P is usually unknown in practice. The empirical risk is used to approximate the expected risk. The empirical risk is defined as follows:

[0104] In the formula, represents the empirical distribution and the training data set Each sample in In line with the empirical distribution Neural networks tend to overfit, which leads to poor performance on test samples despite high accuracy on training samples. To address this problem, a hybrid decision method based on the neighbor risk minimization principle is proposed. This principle involves using prior knowledge to construct neighbor samples from the training set distribution and obtaining a dataset . All data in follows the neighbor distribution , which leads to empirical ortho risk.

[0105]

[0106] In order to ensure the predictive performance and generalization performance of the neural network, it is necessary to minimize not only the empirical risk but also the empirical neighborhood risk. A sample is as follows:

[0107] In the formula, and . is a hyperparameter that controls the number of linear interpolation times, . The lower the value of , the less linear interpolation is performed. When it is equal to 1, the distribution becomes uniform. is the new sample generated; and are two training samples randomly selected from the training dataset.

[0108] In GMINN, the feature sequence After data preprocessing, a training sample set is formed The training sample set is then enhanced to generate an enhanced training sample set Then, several decomposition components and SCINet with different structures are initialized with GA and trained using the enhanced training sample set. To illustrate the time series decomposition module, the first input sample is used As an example.

[0109] (2) Use a genetic algorithm (GA) to randomly initialize N chromosomes, each chromosome represents a set of model hyperparameters, and perform hyperparameter optimization.

[0110] Genetic algorithm is a new global optimization algorithm. In GMINN, complex network structures are required to simulate complex non-stationary load sequences. However, the total number of network structures grows exponentially with the complexity of the network, making it impractical to evaluate all possible structures. Therefore, in order to overcome this challenge and maintain the versatility of the model, a genetic algorithm is used to achieve self-optimization of neural network hyperparameters. The main steps of the genetic algorithm are initialization, fitness calculation, selection, crossover, mutation, and termination condition judgment.

[0111] 1) Initialization: After constructing the input sample, the hyperparameters to be optimized in the framework are encoded into a specific bit string, i.e., chromosome. Each bit of the chromosome corresponds to a hyperparameter. Then, N chromosomes are randomly initialized to form a population. Each chromosome is instantiated as a specific neural network, and all neural networks are trained using the same training set.

[0112] 2) Fitness calculation: In the real world, the fitness of a species to its environment determines its ability to survive. Similarly, genetic algorithms use a fitness function to measure the adaptability of an individual, called fitness. In GMINN, the mean square error (MSE) is chosen as the fitness function.

[0113]

[0114] In the formula, is the predicted value, is the actual value, and n is the total number of samples in the test set.

[0115] 3) Selection: Based on the fitness of each chromosome, use roulette wheel selection to select and obtain Selection is an important step in the genetic algorithm, which determines whether a chromosome participates in the reproduction process. Individuals with higher fitness have a greater chance of being selected, so that the superior genes can be inherited by the next generation.

[0116] 4) Crossover: Next, The chromosomes in the pair are paired and a single-point crossover is performed. That is, a crossover point is randomly selected on the chromosome and the genetic information after that point is exchanged to generate a population of offspring. .

[0117] 5) Mutation: Mutation allowed The chromosomes in the GA randomly change the value of the gene with a certain mutation probability, but within the effective range of the hyperparameters. Crossover only operates on the genetic information generated at initialization, not the hyperparameters generated at initialization. They cannot be added to the subsequent evolution process, which may cause the GA to converge to a local optimum. However, mutation can generate new genetic information, thereby ensuring the diversity of the population and preventing it from falling into a local optimum.

[0118] 6) Termination condition judgment: Evaluate the population generated by the mutation The chromosomes in the test set are checked to see if the termination condition is met. Since the optimal adaptation range of the neural network on the test set cannot be determined, the number of iterations is selected as the termination condition of GA. When the iteration counter t is less than the maximum number of iterations MAX, use As the initial population of the t+1th generation , and returns to step 3 of GMINN. When the iteration counter t reaches MAX, the GA process ends and the optimal network structure is output.

[0119] (3) The input sequence is decomposed into a trend sequence and a residual sequence through a moving average filter (MAF).

[0120] The present invention selects the moving average filter (MAF) as the built-in decomposition component of the framework to avoid data leakage. In MAF, the input sample H1 is averaged and merged To obtain a trend series defined as :

[0121] Where m represents the size of the average pooling layer, m = 2k + 1, t>k.

[0122]

[0123] The decomposition process consists of three steps: 1) Use Pad at both ends of the original input sequence H1 to ensure that the length of the sequence s remains unchanged after average pooling; 2) Use Perform average pooling to obtain trend series ; 3) Subtract the trend sequence from the input sequence H1 , and obtain the residual sequence , the mathematical expression is as follows:

[0124] A MAF-based decomposition component is used to split the feature sequence into two subsequences: a trend sequence that captures the overall trend and a residual sequence that contains finer details. Both subsequences have more predictable patterns than the original sequence. Two separate models are applied to model the subsequences, and then a combiner is used to combine their predictions to obtain the final result. This approach simplifies the prediction task and allows the model to learn more meaningful features from the subsequences, thereby improving prediction accuracy.

[0125] (4) Sample convolution and interactive learning (SCINet) are used to model the two sequences separately and extract temporal dependencies for prediction at multiple temporal resolutions.

[0126] The Sampled Convolutional and Interaction Network (SCINet) is a recursive hierarchical structure that combines downsampling, convolution, and interaction. It extracts and exchanges information from time series at different temporal resolutions. By integrating rich features from multiple resolutions, SCINet can effectively model complex temporal dynamics in time series.

[0127] 1) SCI-Block: The basic module of SCINet encoder is SCI-Block, which consists of a process, which converts the trend sequence The two subsequences retain most of the characteristics of the original sequence. The process can be expressed by the following formula:

[0128] The hidden state is obtained by a one-dimensional convolution Applied to even subsequences , followed by the exponential function The convolution kernel also captures the temporal dependency of the process. Then, to compensate for the possible information loss due to downsampling, SCI-Block uses an interactive learning strategy to With odd subsequence Multiply the corresponding elements of to obtain . This operation can be viewed as a scaling of the odd subsequence Hodd, as shown in the following process:

[0129] In the formula, is the Hadamard product or element-wise multiplication. The odd subsequence undergoes the same operation to obtain .

[0130]

[0131] Then, the scaled features are transformed into Project to a new hidden state, then add or subtract it from combination. Similarly, Finally, SCI-Block outputs two updated subsequences and These subsequences contain essential information extracted from the original sequence, which reflects local and global temporal dependencies. The formula for this operation is as follows:

[0132] 2) SCINet: SCINet is a tree-structured network composed of SCI-Blocks. The network has L layers, each with SCI-Blocks. SCINet for trend sequences We perform continuous downsampling and use SCI-Blocks to learn their temporal dependencies at different temporal resolutions.

[0133] Trend Series After passing through L layers of SCI-Blocks, 2L+1 subsequences are generated. These subsequences are rearranged in the original parity order to form a new sequence Then, through the residual connection, Added to the original sequence to generate an augmented sequence with higher predictability, and finally a simple fully connected network decodes the augmented sequence and produces output.

[0134]

[0135] By adding the output of the trend sequence to the residual sequence in SCINet, the demand response potential prediction value of the sample is obtained. Then, the optimal network structure is continuously searched with the help of genetic algorithm. (5) The fitness function is used to calculate the fitness of each chromosome, and GA is used to perform selection, crossover, and mutation operations.

[0136] (6) Determine whether GA meets the termination condition. If yes, output the optimal demand response potential prediction model; if not, return to step (3).

[0137] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0138] To verify the effectiveness and superiority of the method of the present invention, the performance of GMINN was compared with eight machine learning models: GRU, LSTM, ANN, K-nearest neighbor (KNN), Adaboost, Xgboost, LightGBM and random forest. Four advanced sequence prediction models were also compared: generalized learning system (BLS), SCINet, Nbeats and temporal convolutional network (TCN). Three performance indicators were used to measure the difference between the actual observations and the model predictions: root mean square error (RMSE), mean absolute percentage error (MAPE) and mean absolute error (MAE). The calculation formula is as follows:

[0139]

[0140]

[0141] Table 1 gives the results of the experimental dataset. The best result of each indicator is shown in bold, and the second best result is underlined. The experimental results show that the performance of the proposed model is better than other models.

[0142] Table 1 Comparison of prediction index results between the method of the present invention and the existing method Model RMSE MAE MAPE GRU 3874.763 2912.120 4.004 LSTM 4565.208 3731.241 5.193 ANN 5988.829 5075.859 7.384 daboost 4567.582 3648.055 5.304 KNN 6544.003 5515.540 7.766 LightGBM 4595.713 3642.017 5.211 RandomForest 4600.231 3718.946 5.352 Xgboost 10349.933 9135.410 13.754 SCINet 3094.402 <![CDATA[ 2439.579 ]]> <![CDATA[ 3.447 ]]> TCN <![CDATA[ 3081.878 ]]> 2471.922 3.621 Nbeats 3160.244 2485.251 3.501 BLS 21994.941 15272.763 21.467 GMINN 2920.071 2342.494 3.317 The above embodiments are only used to illustrate the design ideas and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, any equivalent changes or modifications made based on the principles and design ideas disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A method for predicting low-carbon demand response potential based on a generative mobile interactive neural network, characterized by: The following steps are involved: S1: Obtain historical electricity consumption data and demand response data of energy users, as well as the operation data of the integrated energy system IES; S2: Optimize the distribution of energy flow based on the data in step S1 and establish a plan for the optimal economic dispatch of energy flow of the integrated energy system IES; S3: Based on the results of the optimal economic dispatch of energy flow, an extended carbon emission flow model ECEF is established to map the relationship between energy flow and carbon emission flow, calculate the carbon emission flow and obtain the node carbon intensity NCI of the load node, implement a low-carbon demand response mechanism guided by the node carbon intensity NCI, and obtain the carbon emission reduction of energy users; S4: A demand response potential prediction framework based on a generative mobile interactive neural network GMINN is proposed, the carbon emission reduction in step S3 is used as a feature input framework, and the optimal demand response potential prediction model is output.

2. According to claim 1, a method for predicting low-carbon demand response potential based on a generative mobile interactive neural network is characterized in that: In step S1, the operation data includes the operation data of the power system, the thermal system and the natural gas system; In step S2, the energy flow includes electricity, thermal energy flow and natural gas.

3. The method for predicting low-carbon demand response potential based on a generative mobile interactive neural network according to claim 2 is characterized in that: In the step S2, it is assumed that , and are the operating costs of the power system, thermal system and natural gas system in time period t, is the carbon trading cost within the dispatch scope; the objective function of the optimal economic dispatch of energy flow is to minimize the total operating cost F of the integrated energy system IES: ; The specific calculation steps are as follows: S21: Assume that the transportation and storage cost of CO2 is , the excess cost of carbon quota is , the CO2 purchase cost of P2G equipment is ; , and They are the cost coefficients for CO2 transportation and storage, CO2 trading, and CO2 purchase; is the CO2 captured by the CCS from coal-fired unit k during period t in scenario n; is the CO2 consumed by P2G device u in time period t; is the total carbon emissions of the integrated energy system IES in period t; , , and are the carbon emission allowable coefficients of coal-fired unit g, coal-fired unit k with carbon capture and storage device, combined heat and power unit c, and gas turbine r, respectively; is the carbon emission allowable coefficient of the external power grid o; , , , are the power outputs of coal-fired unit g, carbon capture and storage unit k, combined heat and power unit c and gas turbine r in time period t, respectively; is the amount of electricity purchased from the external grid o; is the heat output of CHP unit c in period t; is the conversion efficiency of P2G device u; is the energy consumption of P2G device u; K, U, G, C, R and O are the collections of coal-fired units including carbon capture and storage devices, P2G devices, coal-fired units, CHP units, gas turbines and external grids, respectively; T is the set of time periods; Carbon trading costs are calculated on a daily basis for: ; S22: Set and are the operating cost coefficients of the coal-fired unit g and turbine r, respectively; and are the charging cost coefficient and discharging cost coefficient of the energy storage system e respectively; and are the penalty factors for load shedding and wind curtailment, respectively; is the electricity price during period t; is the load reduction power at node i in period t; is the curtailed wind power at node w in period t; is the predicted wind power in period t; and are the charging active power and discharging active power of the energy storage system e in period t respectively; is the wind power that can be accommodated in period t; I is the set of all nodes; W is the set of wind turbines; E is the collection of energy storage systems; Then the operating cost of the power system in time period t is for: , ; S23: Set is the penalty factor for heat load reduction; is the heat load reduction at node z in time period t; Z is the set of nodes in the heat network; then the operating cost of the thermal system in time period t for: ; S24: Set is the gas cost of source s; is the penalty factor for natural gas load reduction; M is the set of natural gas load reduction points; is the gas load reduction at node m in time period t, is the amount of natural gas exported from source s in time period t; then the operating cost of the natural gas system in time period t for: 。 4. The method for predicting low-carbon demand response potential based on a generative mobile interactive neural network according to claim 1 is characterized in that: In step S2, the constraints for optimal economic dispatch of energy flow include power system operation constraints, thermal system operation constraints, natural gas system operation constraints and energy coupling device constraints; The specific calculation steps are as follows: S25: Power system operation constraints include power balance constraints, operation constraints of carbon capture and storage devices of coal-fired power generation units, branch power flow constraints and unit start-stop constraints; set up represents the device connected to node i; is the power consumption of pump q in time t; is the net power output of coal-fired unit k equipped with carbon capture and storage equipment in period t; is the input electrical power of P2G device u in time period t; and are the powers flowing through branch (i, j) and branch (j, i) in period t, respectively; is the predicted power load at node i in time period t; is the load reduction power at node i in period t in scenario n; the power balance constraint is: ; set up is the net power output of coal-fired unit k equipped with carbon capture and storage equipment in period t; is the fixed power consumption of coal-fired unit k with CCS during time period t; is the operating power consumption of coal-fired unit k with carbon capture and storage during period t; is the energy consumption required to capture each unit of CO2 in the CCS device; and are the carbon capture efficiency and upper limit of the carbon capture and storage device, respectively; is the total carbon emission of coal-fired unit k using carbon capture and storage devices in period t; the operation constraint of the carbon capture and storage device of the coal-fired power generation unit is: ; set up is the power flowing through branch (i, j) in period t; , are the phase angles of node i and node j in period t respectively; is the reactance of branch (i, j); is the power upper limit of branch (i,j), and are the upper and lower limits of the phase angle of node i; L is the branch set; the branch power flow constraint is: ; set up / , / , / , / are the upper and lower limits of the output of coal-fired unit g, coal-fired unit k, gas turbine r, and cogeneration unit c respectively; the start and stop constraints of the units are: ; S26: Thermal system operation constraints include thermal station constraints and heat exchange station constraints; thermal station constraints are as follows: set up is the heat capacity of the liquid in the pipe network of the thermal power station; and are the mass flow rates of the liquids of CHP unit c and pump q, respectively; and are the return temperature and supply temperature of CHP unit c in period t, respectively; and are the supply pressure and return pressure of pump q in period t, respectively; is the working efficiency of pump q; is the density of the liquid d in the pipe; , ; set up is the set of pipelines starting from node z, is the set of pipelines ending at node z; and are the outlet temperatures of the water supply pipe and the water return pipe bh in time period t respectively; and are the inlet temperatures of the water supply pipe and the return pipe bh in the time period t respectively; and are the supply and return temperatures of the thermal network node z in time period t, respectively; is the ambient temperature during period t; is the thermal conductivity of the pipe bh, is the length of pipe bh; is the mass flow rate of the return pipe bh in time period t; , ; set up is the predicted value of the heat demand response potential of node z within time t; V is the set of nodes z; is the mass flow rate of the liquid at the thermal network node z; is the penalty factor for the heat demand response potential of node z within time t; the heat exchange station constraint is: ; S27: In the deterministic scenario n, the natural gas system operation constraints include node flow balance constraints and pipeline operation constraints; set up Indicates connection to natural gas network node a; is the output of natural gas source s in time t; is the gas production of P2G device u in period t; is the gas flow rate of natural gas pipeline network ab in time t; is the gas load reduction of node a in period t; is the predicted gas load of node a in time t; is the gas consumption of the cogeneration unit c in time period t; is the gas consumption of turbine r in period t; S is the setting of gas source; A is the set of gas network nodes; the node flow balance constraint is: ; set up is the physical coefficient of gas pipeline ab; and are the pressures at nodes a and b in period t, respectively; and are the upper and lower limits of the gas flow rate in pipelines ab, respectively; / and / are the upper and lower limits of the pressure at nodes a and b respectively; the pipeline operation constraints are: , , , ; S28: Energy coupling device constraints include P2G operation constraints, gas turbine operation constraints and combined heat and power operation constraints; set up is the efficiency of P2G device u; is the amount of CO2 required per unit of electrical power consumed by P2G device u; is the carbon emission of the P2G equipment in period t; the P2G operation limit is: ; set up is the efficiency of the gas turbine r; is the natural gas input to the gas turbine r; the gas turbine operating constraints are: ; set up is the efficiency of the CHP unit c; the CHP operation limit is: 。 5. The method for predicting low-carbon demand response potential based on a generative mobile interactive neural network according to claim 1, characterized in that: In the step S3, based on the carbon emission flow theory, it is assumed that the CO2 generated on the source side is transferred to the load side through the branch, and a virtual carbon flow accompanies the branch flow; an extended carbon emission flow model ECEF is established by establishing a network carbon emission flow model, a CEF model of an energy coupling device, and a CEF model of an energy storage device.

6. The method for predicting low-carbon demand response potential based on a generative mobile interactive neural network according to claim 5, characterized in that: In step S3, the specific steps of establishing the network carbon emission flow model are: set up is the carbon emissions flowing through a branch or node; the carbon emissions per unit time passing through a branch or node is the carbon emissions flow rate CEFR: ; Carbon intensity CI represents the carbon emissions per unit of energy; Carbon intensity CI includes power generation carbon intensity GCI, branch carbon intensity BCI, gateway carbon intensity PCI and node carbon intensity NCI; , and are the carbon emission intensities of node n, branch i and unit g in period t respectively; is the set of branches that inject power into node n; Indicates that unit g is connected to node n; is the power flowing through branch i in period t; is the power output of unit g in period t; the node carbon intensity NCI represents the average carbon emission per unit energy injected into the node, which is equal to the ratio of the total carbon emission intensity to the total energy injected into the node: ; The inflow at node n includes the inflow from the branch connected to node n and the inflow from the generator connected to node n; when node n is not connected to a generator, the node carbon intensity NCI is: ; set up is the carbon emission of node i in time period t; is the load power at node i; is the time interval; represents the node carbon intensity of node i in period t; the carbon emission of node i in time period t is: 。 7. The method for predicting low-carbon demand response potential based on a generative mobile interactive neural network according to claim 6, characterized in that: In step S3, the specific steps of establishing the CEF model of the energy coupling device are: The energy coupling device includes a single-input single-output SISO energy coupling device and a single-input multiple-output SIMO energy coupling device; For a single-input single-output SISO energy coupling device, and are the input PCI and output PCI of the energy coupling device p in period t, respectively; and are the input and output power of the energy coupling device p in period t respectively; according to the principle of carbon conservation, all carbon emissions related to the input energy are allocated to the output energy, the total carbon emission flow CEFR of the input port is equal to the total carbon emission flow CEFR of the output port, and the carbon intensity PCI of the input and output ports is expressed as: ; Assume that the conversion efficiency of the single-input single-output SISO energy coupling device is , then the relationship between energy input and output is: , ; For single-input multiple-output SIMO energy coupling devices, according to the principle of carbon conservation, all carbon emissions related to the input energy are allocated to the output port; From the perspective of energy conversion, for back pressure CHP, and are the electrical output and thermal output of the CHP unit c in time period t, respectively; is the natural gas input to CHP unit c during time period t; and are the gas-to-electricity conversion efficiency and gas-to-heat conversion efficiency of the cogeneration unit c in time period t, respectively; the output power and heat energy levels are proportional to the input natural gas energy, expressed as: ; From the perspective of carbon emissions, and are the threshold carbon intensity PCI of the electrical output port and the thermal output port of the cogeneration unit c in time period t; is the threshold carbon intensity PCI of the input port of CHP unit c in time period t; the total carbon emission flow CEFR of the input port is equal to the total carbon emission flow CEFR of the output port, and the threshold carbon intensity PCI of the input port and the output port is expressed as: ; Assume that the PCI of the electrical output port and the thermal output port is inversely proportional to the conversion efficiency: 。 8. The method for predicting low-carbon demand response potential based on a generative mobile interactive neural network according to claim 7, characterized in that: In step S3, the specific steps of establishing the CEF model of the energy storage device are: The carbon state of the energy storage system is used to characterize the relationship between the energy and carbon emissions of the energy storage system. is the carbon emissions flowing into the energy storage system e in period t; is the charging power of the energy storage system e in period t; is the NCI of the node where the energy storage system e is located in period t; when the energy storage system is charged, carbon emissions flow into the energy storage system along with the electric energy, as shown in the following formula: ; set up is the carbon emissions released by the energy storage system e in period t; is the discharge power of the energy storage system e in period t; is the GCI of the energy storage system e in period t; is the discharge efficiency of the energy storage system e; is the carbon state of the energy storage system e in the t-1 period; when the energy storage system is discharged, carbon emissions are released from the energy storage system along with the electrical energy, as shown in the following formula: ; set up , are the available capacities of the energy storage system e in time periods t and t-1, respectively; The carbon status of energy storage systems Defined as: 。 9. The method for predicting low-carbon demand response potential based on a generative mobile interactive neural network according to claim 8, characterized in that: In the step S3, it is assumed that Daily carbon emission reduction for power users, and are the power reduction and increase of the load connected to node i during time period t, , are binary variables for load increase and load decrease states, is the proportion of load in the power system that can participate in low-carbon demand response; The low-carbon demand response mechanism is: , , , 。 10. The method for predicting low-carbon demand response potential based on a generative mobile interactive neural network according to claim 1, characterized in that: In the step S4, set up is the target sequence loaded over a period of time; input prediction feature sequence , the total time length is T, the total number of features is d, each of which Represents an input feature; use Represents the time series of the kth feature, using represents the feature vector at time t; the corresponding set of demand response potential prediction values ​​is The specific steps of the prediction process of the demand response potential prediction framework based on the generative mobile interactive neural network GMINN are: S41: Use data generation components to expand the training data set; feature sequence After data preprocessing, a training sample set is formed ; Enhance the training sample set to generate an enhanced training sample set ; Using loss function The average value over the data distribution P represents the expected risk to measure the predictive value With actual value The difference between the expected risk and the expected risk is defined as follows: ; set up represents the empirical distribution, and the training data set Each sample in In line with the empirical distribution ; Use empirical risk to approximate expected risk. The empirical risk is defined as follows: ; A hybrid decision-making method based on the principle of minimizing neighboring risk is adopted; Use prior knowledge to construct neighboring samples from the training set distribution and obtain a data set ; All data in follows the neighbor distribution , the empirical adjacent risk is: ; set up and ; is a hyperparameter that controls the number of linear interpolation times, ; and are two training samples randomly selected from the training data set; are new samples generated; they are mixed from the neighborhood distribution using linear interpolation A sample of: ; The lower the value of , the lower the degree of linear interpolation; when When it is equal to 1, the distribution becomes uniform; S42: Use genetic algorithm GA to optimize hyperparameters; S43: Moving average filter MAF is selected as the built-in decomposition component of the framework to decompose the input feature sequence into a trend sequence that captures the overall trend and a residual sequence containing fine details; S44: Use sample convolution and interactive learning SCINet to model trend series and residual series respectively, extract time dependencies at multiple time resolutions, and obtain the demand response potential prediction value of the sample; S45: Searching for the optimal network structure with the help of genetic algorithm GA; when the termination condition is met, outputting the optimal demand response potential prediction model.

Citation Information

Patent Citations

  • Comprehensive energy system low-carbon scheduling method and system considering extended carbon emission flow

    CN115859686A

  • Low-carbon scheduling method and system for integrated energy system

    CN116128262A

  • Micro-grid low-carbon economic dispatching method, system, equipment and medium

    CN116780546A

  • Performing interactive digital image operations utilizing modified machine learning models

    US20230120232A1

Cited By

  • Power distribution system source load coordinated regulation and control method and system based on low-carbon demand response

    CN122620529A