Wind power system demand response low-carbon economic dispatching method based on carbon emission flow

Abstract

The invention relates to the technical field of power dispatching, and discloses a wind power system demand response low-carbon economic dispatching method based on carbon emission flow. In the first stage, an economic dispatching model is constructed by taking the minimum total operation cost of a wind power system as a target, and a dispatching strategy of each unit is solved by adopting a TD3 algorithm. And in the second stage, on the basis of a carbon emission flow theory, node carbon potential is calculated by utilizing the scheduling scheme solved in the first stage, and a load side demand response low-carbon model taking the minimum sum of the carbon emission cost and the demand response cost as a target is established, so that the carbon emission responsibility of the power system is reduced from the power generation side to the load side. According to the two-stage demand response method, load distribution in the whole scheduling period can be optimized, a load curve is effectively smoothed, the wind power utilization rate is improved, and energy conservation and emission reduction of the load side are actively guided, so that the total amount of carbon emission is reduced, and the carbon reduction potential of the load side is fully excavated.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power dispatching, and in particular to a low-carbon economic dispatching method for demand response of a wind power system based on carbon emission flow. Background Art

[0002] The power generation side generates electricity by burning non-renewable resources such as coal and natural gas, which produces a large amount of carbon dioxide greenhouse gas and forms carbon emissions. Now assume that these carbon dioxide are not emitted into the atmosphere, but flow from the power generation side to the load side following the power system flow, forming a virtual carbon emission flow. In the process from power generation to electricity consumption, users consume electricity, and carbon emissions follow the power flow to the load side accordingly. However, in traditional low-carbon dispatch research involving wind power systems, carbon emissions are mainly closely related to the power generation side, and fixed carbon emission coefficients are used to calculate carbon emissions. However, this method simplifies carbon emission behavior and cannot reflect the temporal and spatial differences in carbon emission accounting.

[0003] In order to track carbon emissions from the power generation side to the load side, the power system carbon emission flow method was born. The power system carbon emission flow theory is a theoretical framework for evaluating the generation, transmission and consumption of carbon emissions in the entire power system. This theory studies the correlation between carbon emission levels and the components of the power system. The power system carbon emission flow theory determines the distribution and overall level of carbon emissions based on the results of power system flow calculations. The carbon emission flow theory can guide the optimization of user electricity consumption behavior, assist in the evaluation and settlement of carbon emission reduction, and facilitate the formulation of carbon emission management measures on the source-load side of the power system. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a low-carbon economic dispatching method for wind power system demand response based on carbon emission flow, which can optimize the load distribution in the whole dispatching cycle, effectively smooth the load curve, improve the utilization rate of wind power, actively guide the energy conservation and emission reduction on the load side, thereby reducing the total carbon emissions and fully tapping the carbon reduction potential on the load side. The technical solution is as follows: The low-carbon economic dispatch method for demand response of a wind power system based on carbon emission flow includes the following steps: Step 1: Collect the initial values ​​of the power system, including unit parameters, electric load data and wind power data, as input data; Step 2: A one-stage economic dispatch model is constructed with the goal of minimizing the overall operating cost of the wind power system generation side, and the TD3 algorithm is used to solve the dispatch strategy of each unit, including unit output data and power flow distribution data; Step 3: Based on the unit output data and power flow distribution data obtained from the one-stage economic dispatch model, the carbon potential and carbon emissions of each node are calculated using the carbon emission flow theory; Step 4: Take the node carbon potential and carbon emissions calculated in step 3 as the basic data input, take the original load data and carbon price as the basic parameters input, and use the carbon price as a signal to establish a two-stage demand response low-carbon optimization model. Use the carbon price as a price signal to guide users to respond to demand, so that the carbon emission responsibility of the power system is transferred from the power generation side to the load side. Step 5: Feedback the load responded in step 4 to the one-stage economic dispatch model for solution, output the recalculated optimal economic dispatch result, and calculate the total carbon emissions and carbon emission costs on the load side at this time.

[0005] Furthermore, the step 2 specifically includes: Step 2.1: Construct a one-stage economic dispatch model with the goal of minimizing the overall operating cost of the power generation side of the wind power system: ; in, , and They are wind power cost, conventional thermal power generation cost and energy storage operation cost respectively; Step 2.2: Get the current time Corresponding status Status information: ; in, They are Load power at all times, energy storage charging and discharging status, voltage at each node, wind turbine power and conventional thermal power unit power; Step 2.3: The Markov decision process of the one-stage economic dispatch model takes the real-time charging and discharging power of the energy storage battery as the action of the intelligent agent. The action space of the intelligent agent is It is expressed as: ; in, for Moment The charging and discharging power of each energy storage device; is the number of energy storage devices; Step 2.4: Convert the one-stage economic scheduling problem into the cumulative reward maximization form of deep reinforcement learning for solution, and set the reward function in the process of interaction between the agent and the environment as: ; Step 2.5: Use the TD3 algorithm to solve the one-stage economic dispatch model. In the algorithm, the policy network takes the current state as input and the dispatch decision as the action output; the intelligent agent is trained by interacting with the wind power distribution network environment to obtain trajectory data, and the trajectory data is placed in the experience replay pool, and m samples are extracted from it for the optimization of the policy network and the value network; the policy network updates parameters by maximizing the cumulative expected return; the value network updates parameters by minimizing the difference between the estimated value and the target value.

[0006] Furthermore, the step 2.5 specifically includes: Step 2.5.1: By optimizing the policy network parameters , to maximize the policy function The purpose of strategy function The equation is expressed as: ; Where: Strategy function Indicates the expected return that can be obtained; Indicates the current state in the strategy The distribution under Indicates in status Follow the strategy The value function of the action; is the expected function; the gradient descent method is used to update the strategy function The parameters in are as follows: ; Where: is the policy function The gradient of is the action value function The gradient of is the value network parameter; For action, Status strategy; are the policy network parameters; Step 2.5.2: Value Network Update: The value network evaluates the actions taken by the policy network, the action-value function It is expressed by the Bellman equation: ; Where: is the discount factor, The state of the next moment; For the next state Actions; is the reward value at the current moment; The value network is parameterized using a neural network and the action-value function is approximated by minimizing the Bellman residual: ; Where: and Represent the parameters of the current value network and the target value network respectively; for Output of the moment value network Q value; The goal of the moment value network output Q The values ​​are: ; Approximate the action-value function loss function using Monteroit sampling , update the parameters of the current value network by minimizing the loss function , the loss function is: ; Where: is the number of training samples; For the Samples in The target action-state value at the moment; For the Samples in The status information at all times includes load power, energy storage charging and discharging status, voltage of each node, power of wind turbines and status information of conventional thermal power units; For the Energy storage samples in The charging and discharging power at the time; A dual-target value network and target strategy smoothing regularization are used, that is, two sets of target value networks with the same network architecture are used, and the minimum value between the two is selected as the target value to suppress the over-estimation problem of the value network. The specific expression is: ; Where: Network parameters for two different target value networks; is the serial number of the value network; Target policy smoothing regularization uses target actions to calculate the target Q Value, so in the next state The actions are: ; Where: is random noise; Step 2.5.3: Parameter delayed update: The policy network and target value network parameters are updated in the form of soft updates: ; Where: is the learning rate, are the network parameters of different current value networks, are policy network parameters, The target policy network.

[0007] Furthermore, step 3 specifically includes: Step 3.1: Define the carbon emission flow density of a branch of the power system is the ratio of the carbon emission flow rate to the active power of the branch: ; Where: is the carbon emission flow density of a branch; is the active power of a branch; is the carbon emission flow rate of a branch; Step 3.2: Define the The ratio of the carbon flow into a power generation or consumption node to the active power flowing through the node is the node carbon potential, which is expressed as express: ; Where: The number of branches connected to the node that inject power is represented as the set of all branches; is the node number, Injection node Active power of Injection node Carbon emission flow density of the branch; For Node The output power of the generator; is the carbon potential of the generating unit node, i.e., the carbon emission intensity of the generating unit; Step 3.3: Use the branch power flow distribution matrix to describe the active power distribution of the power system. The branch power flow distribution matrix is order square matrix, using express: ; Where: When the node With Node When the active power flow between is positive, , =0; otherwise, , = ; For all diagonal elements, we have ; Use each unit to inject the distribution matrix It is used to describe the active power injected when the generator set is connected to the power system. The matrix is ​​as follows: ; In the formula: If Generators at the node Connect to the system and inject active power into the system When , in other cases is 0; Node active flux matrix It is expressed as: ; ; Where: For Node To Node The branch power of According to known and The calculation results are: ; Where: For elements with 1 Order row vector; is an auxiliary variable; is the node active flux matrix The order of for the order of the auxiliary variables; Step 3.4: The carbon emission flow density of a branch Substitute it with the carbon potential of the node at the beginning of the branch and rewrite it into matrix form as follows: ; Where: is the carbon emission intensity vector of the generator set, For the unit row vector with elements equal to 1, is the node carbon potential matrix; is the transpose symbol; It is a set of generators connected to the node and having output; The node carbon potential matrix is: ; ; Where: is the load distribution matrix, which is constructed in the same way as the unit injection distribution matrix Consistent, R is the carbon emission matrix of the load nodes.

[0008] Furthermore, step 4 specifically includes: The scheduling objective is to minimize the sum of carbon emission cost and demand response cost: ; Where: is the carbon price at different times; for Carbon emissions of system load nodes at each moment; is the demand response amount; for demand response costs; is the number of load nodes; During demand response, the load balancing constraints are as follows: ; Where: To respond to the preload power; is the power after response; The load variation constraints are as follows: ; ; Where: and represent the upper and lower limits of the transferable load factor respectively; The carbon emission constraints are as follows: ; Where: for Carbon potential of system load node at moment.

[0009] Compared with the prior art, the beneficial effects of the present invention are: in the first stage of the scheduling method of the present invention, an economic scheduling model is constructed with the goal of minimizing the overall operating cost of the wind power system, and the TD3 algorithm is used to solve the scheduling strategy of each unit; in the second stage, based on the carbon emission flow theory, the node carbon potential is calculated using the scheduling scheme solved in the first stage, and a load-side demand response low-carbon model is established with the goal of minimizing the sum of carbon emission cost and demand response cost, so as to transfer the carbon emission responsibility of the power system from the power generation side to the load side; the two-stage demand response method can optimize the load distribution within the entire scheduling cycle, effectively smooth the load curve, improve the utilization rate of wind power, and actively guide energy conservation and emission reduction on the load side, thereby reducing the total carbon emissions. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is an overall schematic diagram of the low-carbon economic dispatching method for demand response of a wind power system based on carbon emission flow according to the present invention.

[0011] Figure 2 A schematic diagram of economic dispatch based on the TD3 algorithm in one stage of the method of the present invention.

[0012] Figure 3 This is a comparison chart of the algorithm training before and after demand response.

[0013] Figure 4 This is the load change diagram of the algorithm before and after demand response.

[0014] Figure 5 This is the change diagram of wind turbine after demand response training.

[0015] Figure 6 This is the change diagram of thermal power units after demand response training.

[0016] Figure 7 This is a diagram showing the changes in energy storage after demand response training.

[0017] Figure 8 is the carbon emission intensity of each node before demand response.

[0018] Fig. 9 is the carbon emission intensity of each node after demand response.

[0019] Fig.10 This is the carbon emission change curve before and after demand response. DETAILED DESCRIPTION

[0020] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] In one embodiment of the present invention, Figure 1 As shown, the present invention proposes a low-carbon economic intelligent dispatching method for demand response of a wind power system based on carbon emission flow theory, comprising the following steps: Step S1: Input the initial values ​​of the power system, including basic data such as unit parameters, power load data, and wind power data.

[0022] Step S2: A one-stage economic dispatch model is constructed with the goal of minimizing the overall operating cost of the wind power system on the power generation side, and the TD3 algorithm is used to solve the dispatch strategy of each unit, including unit output data and power flow distribution data.

[0023] Construct a one-stage economic dispatch model with the goal of minimizing the overall operating cost of the system power generation side: ; in, , and They are wind power cost, conventional thermal power generation cost and energy storage operation cost respectively; Get the current time Corresponding status information ,in , They are Load power at all times, energy storage charging and discharging status, voltage at each node, wind turbine power, conventional thermal power unit power and time.

[0024] The Markov decision process of the one-stage economic dispatch model takes the real-time charging and discharging power of the energy storage battery as the action of the intelligent agent. The action space of the intelligent agent is It can be expressed as: ; in, for Moment The charging and discharging power of an energy storage device.

[0025] The one-stage economic scheduling problem is transformed into the cumulative reward maximization form of DRL (Deep Reinforcement Learning) for solution, and the reward function in the process of interaction between the agent and the environment is set as: ; The TD3 algorithm is used to solve the one-stage economic dispatch model, such as Figure 2 As shown in the lower part, the policy network in the algorithm takes the current state as input and outputs the dispatch decision as action. The agent is trained by interacting with the wind power distribution network environment to obtain trajectory data, which is placed in the experience replay pool and m samples are extracted from it for policy and value optimization. The policy network updates parameters by maximizing the cumulative expected return; the value network parameters are updated by minimizing the difference between the estimated value and the target value.

[0026] The specific process is as follows: 1) By optimizing the policy network parameters , to maximize the policy function The purpose of this equation can be expressed as: ; Where: Indicates the current state in the strategy The distribution under Indicates in status Follow the strategy The value of the action; Indicates the expected return that can be obtained.

[0027] Update the policy function using gradient descent method Parameters in: ; 2) Value network update: The value network evaluates the actions taken by the policy network, the action-value function The Bellman equation can be expressed as: ; The value network is parameterized using a neural network and the action-value function is approximated by minimizing the Bellman residual: ; Where: , Represents the current Q Network and Target Q Parameters of the network. The output of the moment value network is ,exist The output at the moment is Approximate the action-value function loss function using Monteroit sampling , update the parameters of the current network by minimizing the loss function , the loss function is: ; Where: is the number of training samples; For the goal Q value.

[0028] Since the value network has problems such as overestimation during training and large variance of target estimation caused by deterministic strategies, the TD3 algorithm uses a dual-target value network and target strategy smoothing regularization to solve the problem. A dual-target value network refers to a target value network with two sets of the same network architecture, and the minimum value between the two is selected as the target value, thereby suppressing the overestimation problem of the value network, which is specifically expressed as: ; Where: For two different goals Q Network parameters of the network; for Q The serial number of the network.

[0029] Target policy smoothing regularization uses the target action to calculate the target Q value, which is conducive to the smooth estimation of the target, so in the next state The actions are: ; Where: is random noise. In order to make the target action closer to the original action, a normal distribution is generally used and the sampling noise is truncated , is the variance, are the upper and lower limits of the noise.

[0030] 3) Delayed parameter update: In order to reduce Q To prevent errors caused by updating the policy network before the network is stable, the policy network parameter update should be delayed and the frequency of value network parameter update should be increased. The policy network parameters should be updated after the value network converges to reduce the cumulative error and thus the variance. The policy network and target value network parameters are updated in the form of soft updates: ; Where: is the learning rate, are the network parameters of the network with different current values, is the current policy network parameter, The target policy network.

[0031] Step S3: output the unit output data and power flow distribution data of the first-stage scheduling model, and calculate the carbon potential and carbon emissions of each node according to the carbon emission flow theory; The carbon flow rate that passes through a branch of the power system in unit time is called the carbon emission flow rate, which is usually expressed as Expressed in tCO2 / h: ; Where: F It represents carbon flow, and its physical meaning is the cumulative amount of carbon emissions generated by a branch or node in the power system within a given period of time, in units of tCO2.

[0032] The carbon emission flow of the branch in the power system is affected by the active power flow. In order to clarify the correlation between the carbon emission flow rate and the active power flow, the branch carbon emission flow density is used to represent it. is the ratio of the carbon emission flow to the active power of the branch: ; Where: is the active power of a branch. The unit of branch carbon flow density is tCO2 / (MW·h).

[0033] The branch carbon flow density reveals the inherent relationship between branch carbon emissions and power flow. However, in the power system, power generation and consumption are mostly carried out in the form of nodes. The ratio of the carbon flow into a node to the active power flowing through the node is the node carbon potential, which is expressed as express: ; Where: The number of branches connected to the node that inject power is represented as the set of all branches; is the node number, Injection node Active power of Injection node Carbon emission flow density of the branch; For Node The output power of the generator; is the carbon potential of the generating unit node, that is, the carbon emission intensity of the generating unit.

[0034] Power flow calculation is the basis for carbon emission flow calculation. After the power flow calculation is completed, the corresponding branch flow, power generation output, carbon emission intensity of the generator set and other parameters can be obtained. Using these parameters and according to the carbon emission flow theory, the carbon potential of each node in the power system can be obtained. Based on the node carbon potential, the carbon emission responsibility of the power generation side can be effectively allocated to the load side, and then the nodes with higher carbon potential can be treated with low carbon.

[0035] In this embodiment, the power flow distribution matrix of each branch describes the active power distribution of the power system. The matrix is order square matrix, using express: ; Where: When the node With Node When the active power flow between is positive, , =0; otherwise, , = ; For all diagonal elements, we have .

[0036] In this embodiment, each unit injects a distribution matrix It is used to describe the active power injected when the generator set is connected to the power system. The matrix is ​​as follows: ; In the formula: If Generators at the node Connect to the system and inject active power into the system When , in other cases is 0.

[0037] In this embodiment, it can be known from Kirchhoff's current law that at any time, for any node, the absolute values ​​of all currents flowing into and out of the node are equal. However, in the carbon emission flow theory, the node carbon potential is only related to the active power injected into the node, and has nothing to do with the active power flowing out of the node.

[0038] Node active flux matrix It can be expressed as: ; ; Where: For Node To Node of branch power.

[0039] In this embodiment, it can be further deduced based on the above formulas: ; Where: For elements with 1 Order row vector; is an auxiliary variable. The above formula shows that when , In known cases, It can be directly obtained by calculation.

[0040] In this embodiment, the branch carbon flow density It can be replaced by the carbon potential of the node at the beginning of the branch and rewritten into the matrix form as follows: ; Where: is the carbon emission intensity vector of the generator set, For the unit row vector with elements equal to 1, is the node carbon potential matrix. According to the definition of the node active flux matrix, we can further obtain: ; In this embodiment, due to is a diagonal matrix, and the formula Expanding to the entire dimension gives: ; In this embodiment, the formula After deformation, the node carbon potential matrix is ​​obtained: ; ; Where: is the load distribution matrix, which is constructed in the same way as the unit injection distribution matrix Consistent.

[0041] Step S4: Input the node carbon potential and carbon emissions calculated in step S3 into the two-stage model as basic data; input the original load data, carbon price and other basic parameters, and establish a two-stage demand response low-carbon optimization model with carbon price as the signal. The two-stage model uses carbon price as a price signal to guide users to respond to demand.

[0042] In this embodiment, the scheduling target is to minimize the sum of carbon emission cost and demand response cost: ; Where: is the carbon price at different times; for Carbon emissions of system load nodes at each moment; is the demand response amount; is the demand response cost.

[0043] During the demand response process, the load balancing constraint ensures that the load response will not cause changes in the total system load within the dispatch cycle, but instead transfers the transferable load from periods with higher carbon emission coefficients to periods with lower carbon emission coefficients.

[0044] ; Where: To respond to the preload power; is the power after response.

[0045] The load variation constraints are as follows: ; ; Where: and represents the upper and lower limits of the transferable load factor, which are 0.8 and 1.2 respectively.

[0046] The carbon emission constraints are as follows: ; Where: for Carbon potential of system load node at moment.

[0047] Step S5: Feedback the load after response in step S4 to the first stage for solution, output the recalculated optimal economic dispatch result, and calculate the total carbon emissions and carbon emission costs on the load side at this time.

[0048] This experiment simulates the actual operating environment by calculating the power flow, using Tensorflow2.6.0 as the framework and Python3.8 as the programming environment. The simulation is implemented on a computer equipped with an Inter core i9-13900HX CPU and an NVIDIA RTX4070 graphics card. The experiment uses the improved IEEE33 node power distribution system as an example for simulation. The detailed structure is as follows Figure 2As shown in the upper part, in the improved IEEE33 node, a traditional thermal power generator is connected at node 1, a wind turbine is connected at node 9, and 10 distributed energy storage units with a capacity of 0.24MW are equipped at node 14.

[0049] Based on the two-stage low-carbon economic dispatch model of the present invention, the economic performance of the wind power system before and after demand response is compared in the experiment. The training results are as follows: Figure 3 shown. Figure 4 The paper shows the changes in system load before and after demand response within a scheduling cycle. The low-carbon demand response model based on node carbon emission flow theory transfers part of the peak load to the valley period, achieving the effect of "peak shaving and valley filling" on the load side. Figure 5 The changes in wind power output before and after demand response are shown. The wind power output after demand response is greater than before response in the night, early morning and afternoon periods. The main reason is that the load of the system is adjusted after the response.

[0050] Figure 6 to Figure 7 The output of conventional units before and after the response and the energy storage charging and discharging power after the response are shown respectively. Due to the anti-peak regulation characteristics of wind power, during the night time period when the load is relatively low, in order to improve the economy of the system, the intelligent agent controls the energy storage to adaptively adapt to the wind power output and system power load for charging.

[0051] Figures 8 to 9 The change in carbon emission intensity of each node before and after demand response is shown. Overall, the carbon emission intensity of the system is more stable due to the response of the system load. Fig.10 The carbon emission change curve of the system before and after the response is shown. It can be seen from the figure that the carbon emission of the system before and after the response is basically consistent with the output change trend of the conventional unit. After the response, the overall carbon emission of the system was reduced by 78.74t, which effectively verified the effectiveness of the proposed low-carbon model.

[0052] In summary, the verification results of the improved IEEE33 node simulation example show that the two-stage demand response method proposed in this invention can optimize the load distribution within the entire scheduling cycle, effectively smooth the load curve, improve the utilization rate of wind power, and actively guide energy conservation and emission reduction on the load side, thereby reducing the total carbon emissions.

Claims

1. A low-carbon economic dispatch method for demand response of wind power system based on carbon emission flow, characterized in that: The following steps are involved: Step 1: Collect the initial values ​​of the power system, including unit parameters, electric load data and wind power data, as input data; Step 2: A one-stage economic dispatch model is constructed with the goal of minimizing the overall operating cost of the wind power system generation side, and the TD3 algorithm is used to solve the dispatch strategy of each unit, including unit output data and power flow distribution data; Step 3: Based on the unit output data and power flow distribution data obtained from the one-stage economic dispatch model, the carbon potential and carbon emissions of each node are calculated using the carbon emission flow theory; Step 4: Take the node carbon potential and carbon emissions calculated in step 3 as the basic data input, take the original load data and carbon price as the basic parameters input, and use the carbon price as a signal to establish a two-stage demand response low-carbon optimization model. Use the carbon price as a price signal to guide users to respond to demand, so that the carbon emission responsibility of the power system is transferred from the power generation side to the load side. Step 5: Feedback the load responded in step 4 to the one-stage economic dispatch model for solution, output the recalculated optimal economic dispatch result, and calculate the total carbon emissions and carbon emission costs on the load side at this time.

2. The low-carbon economic dispatch method for demand response of wind power system based on carbon emission flow according to claim 1 is characterized in that: The step 2 specifically includes: Step 2.1: Construct a one-stage economic dispatch model with the goal of minimizing the overall operating cost of the power generation side of the wind power system: ； in, , and They are wind power cost, conventional thermal power generation cost and energy storage operation cost; Step 2.2: Get the current time Corresponding status Status information: ； in, They are Load power at all times, energy storage charging and discharging status, voltage at each node, wind turbine power and conventional thermal power unit power; Step 2.3: The Markov decision process of the one-stage economic dispatch model takes the real-time charging and discharging power of the energy storage battery as the action of the intelligent agent. The action space of the intelligent agent is It is expressed as: ； in, for Moment The charging and discharging power of each energy storage device; is the number of energy storage devices; Step 2.4: Convert the one-stage economic scheduling problem into the cumulative reward maximization form of deep reinforcement learning for solution, and set the reward function in the process of interaction between the agent and the environment as: ； Step 2.5: Use the TD3 algorithm to solve the one-stage economic dispatch model. In the algorithm, the policy network takes the current state as input and the dispatch decision as the action output; the intelligent agent is trained by interacting with the wind power distribution network environment to obtain trajectory data, and the trajectory data is placed in the experience replay pool, and m samples are extracted from it for the optimization of the policy network and the value network; the policy network updates parameters by maximizing the cumulative expected return; the value network updates parameters by minimizing the difference between the estimated value and the target value.

3. The low-carbon economic dispatch method for demand response of wind power system based on carbon emission flow according to claim 2 is characterized in that: The step 2.5 specifically includes: Step 2.5.1: By optimizing the policy network parameters , to maximize the policy function The purpose of strategy function The equation is expressed as: ； Where: Strategy function Indicates the expected return that can be obtained; Indicates the current state in the strategy The distribution under Indicates in status Follow the strategy The value function of the action; is the expected function; the gradient descent method is used to update the strategy function The parameters in are as follows: ； Where: is the policy function The gradient of is the action value function The gradient of is the value network parameter; For action, Status strategy; are the policy network parameters; Step 2.5.2: Value Network Update: The value network evaluates the actions taken by the policy network, the action-value function It is expressed by the Bellman equation: ； Where: is the discount factor, The state of the next moment; For the next state Actions; is the reward value at the current moment; The value network is parameterized using a neural network and the action-value function is approximated by minimizing the Bellman residual: ； Where: and Represent the parameters of the current value network and the target value network respectively; for Output of the moment value network Q value; The goal of the moment value network output Q The values ​​are: ； Approximate the action-value function loss function using Monteroit sampling , update the parameters of the current value network by minimizing the loss function , the loss function is: ； Where: is the number of training samples; For the Samples in The target action-state value at the moment; For the Samples in The status information at all times includes load power, energy storage charging and discharging status, voltage of each node, power of wind turbines and status information of conventional thermal power units; For the Energy storage samples in The charging and discharging power at the time; A dual-target value network and target strategy smoothing regularization are used, that is, two sets of target value networks with the same network architecture are used, and the minimum value between the two is selected as the target value to suppress the over-estimation problem of the value network. The specific expression is: ； Where: Network parameters for two different target value networks; is the serial number of the value network; Target policy smoothing regularization uses target actions to calculate the target Q Value, so in the next state The actions are: ； Where: is random noise; Step 2.5.3: Parameter delayed update: The policy network and target value network parameters are updated in the form of soft updates: ； Where: is the learning rate, are the network parameters of different current value networks, are policy network parameters, The target policy network.

4. The low-carbon economic dispatch method for demand response of wind power system based on carbon emission flow according to claim 1 is characterized in that: Step 3 specifically includes: Step 3.1: Define the carbon emission flow density of a branch of the power system is the ratio of the carbon emission flow rate to the active power of the branch: ； Where: is the carbon emission flow density of a branch; is the active power of a branch; is the carbon emission flow rate of a branch; Step 3.2: Define the The ratio of the carbon flow into a power generation or consumption node to the active power flowing through the node is the node carbon potential, which is expressed as express: ； Where: The number of branches connected to the node that inject power is represented as the set of all branches; is the node number, Injection node Active power of Injection node Carbon emission flow density of the branch; For Node The output power of the generator; is the carbon potential of the generating unit node, i.e., the carbon emission intensity of the generating unit; Step 3.3: Use the branch power flow distribution matrix to describe the active power distribution of the power system. The branch power flow distribution matrix is order square matrix, using express: ； Where: When the node With Node When the active power flow between is positive, , =0; otherwise, , = ; For all diagonal elements, we have ; Use each unit to inject the distribution matrix It is used to describe the active power injected when the generator set is connected to the power system. The matrix is ​​as follows: ； In the formula: If Generators at the node Connect to the system and inject active power into the system When , in other cases is 0; Node active flux matrix It is expressed as: ； ； Where: For Node To Node The branch power of According to known and The calculation results are: ； Where: For elements with 1 Order row vector; is an auxiliary variable; is the node active flux matrix The order of for the order of the auxiliary variables; Step 3.4: The carbon emission flow density of a branch Substitute it with the carbon potential of the node at the beginning of the branch and rewrite it into matrix form as follows: ； Where: is the carbon emission intensity vector of the generator set, For the unit row vectors with elements equal to 1, is the node carbon potential matrix; is the transpose symbol; It is a set of generators connected to the node and having output; The node carbon potential matrix is: ； ； Where: is the load distribution matrix, which is constructed in the same way as the unit injection distribution matrix Consistent, R is the carbon emission matrix of the load nodes.

5. The low-carbon economic dispatch method for demand response of wind power system based on carbon emission flow according to claim 1 is characterized in that: Step 4 specifically includes: The scheduling objective is to minimize the sum of carbon emission cost and demand response cost: ； Where: is the carbon price at different times; for Carbon emissions of system load nodes at each moment; is the demand response amount; for demand response costs; is the number of load nodes; During demand response, the load balancing constraints are as follows: ； Where: To respond to the preload power; is the power after response; The load variation constraints are as follows: ； ； Where: and represent the upper and lower limits of the transferable load factor respectively; The carbon emission constraints are as follows: ； Where: for Carbon potential of system load node at moment.

Images