Electric power system double-layer optimization scheduling method considering electricity-carbon joint demand response

Abstract

An electric power system double-layer optimization scheduling method considering electricity-carbon joint demand response belongs to the technical field of electric power system optimization scheduling, and comprises the following steps: firstly, providing a CNN-BiGRU-Attention hybrid neural network, and carrying out wind and light generation power prediction; secondly, establishing an intelligent building cluster carbon emission model comprising an electric vehicle, a reducible load, a transferable load and stored energy, and realizing fine quantification of intelligent building cluster carbon emission; then, a power system double-layer optimization model considering the electricity-carbon joint demand response is established, an upper-layer power grid operator takes the minimum operation cost as the target, a lower-layer intelligent building cluster takes the minimum total cost as the target, loop iteration solution is carried out between the upper layer and the lower layer through node carbon potential, electricity price and electricity demand, and the power system double-layer optimization model is established; a more reasonable scheduling scheme considering both economical efficiency and low-carbon property is provided for the day-ahead optimization scheduling of the power system; an example result verifies that the proposed method can provide a reference for the optimization scheduling work of the power system.

Description

Technical Field

[0001] This invention belongs to the field of power system optimization scheduling technology, and in particular relates to a two-level optimization scheduling method for power systems that considers the combined electricity and carbon demand response. Background Technology

[0002] With the widespread integration of high-proportion renewable energy sources into the power system, high-precision wind and solar power generation forecasting has become a key link in improving the reliability and economy of power system operation. However, using a single neural network for wind and solar power generation forecasting has certain limitations. Research on low-carbon optimal dispatching of the power system mainly focuses on the low-carbon potential of the source side, with little research on low-carbon dispatching between the grid side and smart building clusters. With the continuous acceleration of urbanization in China, it is of great significance to fully explore and effectively utilize the huge energy-saving and carbon-reduction potential of smart building clusters. Therefore, a two-level optimal dispatching method for the power system that comprehensively considers wind and solar power generation forecasting and electricity-carbon joint demand response has important practical significance. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a two-layer optimal scheduling method for power systems that considers the combined demand response of electricity and carbon. By deeply exploring the flexible adjustment capabilities of the load side and giving full play to the role of source-load coordination and carbon trading mechanisms in reducing carbon emissions, this method solves the current technical problem that it is difficult to incentivize load-side carbon reduction solely by relying on electricity price signals.

[0004] A two-level optimal dispatch method for power systems considering combined electricity and carbon demand response includes the following steps, which are performed sequentially:

[0005] Step 1: Establish a prediction model for wind and solar power generation based on CNN-BiGRU-Attention neural network, and output the predicted wind and solar power generation power including timestamps through this model;

[0006] Step 2: Establish a carbon emission model for intelligent building clusters based on carbon emission flow theory, and obtain the carbon potential of each node through this model;

[0007] Step 3: Calculate the carbon emissions of the smart building cluster using the carbon potential of each node, establish a carbon emission model for the smart building cluster, and obtain the total carbon emissions generated by the smart building cluster for each node; the carbon emission model for the smart building cluster includes a carbon emission model for electric vehicles, a carbon emission model for load reduction, a carbon emission model for load transfer, and a carbon emission model for energy storage.

[0008] Step 4: Obtain the carbon trading cost of the smart building cluster;

[0009] Step 5: Establish a two-layer optimal dispatch model for the power system that considers the joint demand response of electricity and carbon. This two-layer optimal dispatch model includes an upper-layer economic dispatch model for grid operators and a lower-layer intelligent building cluster response model. The economic dispatch model for grid operators aims to minimize the operating cost of grid operators, while the lower-layer intelligent building cluster response model aims to minimize the total cost. The upper and lower layers are solved iteratively through node carbon potential, electricity price, and electricity demand to obtain a two-layer optimal dispatch plan for the power system that considers the joint demand response of electricity and carbon.

[0010] The specific steps for establishing the wind and solar power generation prediction model based on the CNN-BiGRU-Attention neural network in step one are as follows:

[0011] Step 1: The prediction model for wind and solar power generation based on CNN-BiGRU-Attention neural network includes a convolutional neural network (CNN) layer, a bidirectional gated recurrent unit (BiGRU) layer, an attention layer, a fully connected layer, and an output layer connected in sequence; wherein, the CNN layer includes a convolutional layer I, a pooling layer I, a convolutional layer II, and a pooling layer II connected in sequence.

[0012] Extract the original historical data of wind power and photovoltaic power, parse the timestamps, extract the time features of each month, each weekend, and each hour, normalize the time features and power data respectively, fill in the missing values, and divide the dataset into training set and validation set.

[0013] Step 2: Input the training set into the CNN layer, where convolutional layer I and convolutional layer II are used to extract local spatiotemporal joint feature information of wind power and photovoltaic power data and time features, and pooling layer I and pooling layer II are used to select key features. After two feature extractions and key feature selections by the CNN layer, the most significant feature combination is retained to form high-order spatiotemporal features.

[0014] Step 3: Pass the high-order spatiotemporal features extracted by the CNN layer to the BiGRU layer, and use the bidirectional gated recurrent unit structure of the BiGRU layer to learn the long-term dependencies of the feature sequence, thus expanding the local spatial features into dynamic features with temporal dependencies.

[0015] Step 4: In the attention layer, the attention mechanism is used to adaptively assign weights to the time-dependent dynamic features output by BiGRU, focusing on key time step information, and the weighted feature vector is output to the fully connected layer;

[0016] Step 5: The feature vectors weighted by the attention mechanism are nonlinearly combined and dimensionality reduced through a fully connected layer to map the high-dimensional feature space to the prediction dimension. Finally, inverse normalization is performed to output the predicted power of wind and solar power generation including timestamps.

[0017] Step 6: After validation on the validation set, the prediction model for wind and solar power generation based on the CNN-BiGRU-Attention neural network is completed.

[0018] The formulas for calculating the carbon potential of each node in step two are as follows:

[0019]

[0020] In the formula: e m,t Let ρ be the nodal carbon potential of node m; l,t The carbon flux density of branch l at time t is equal to the nodal carbon potential of the first segment of the branch; P l.t Let be the active power of branch l at time t; Let P be the carbon emission intensity of the i-th generator unit; m,t L represents the unit output value at node m at time t; m This is the set of lines connected to node m.

[0021] The process of establishing the carbon emission model for electric vehicles in step three is as follows:

[0022] The intelligent building cluster collects real-time charging demand data from each electric vehicle and implements centralized scheduling and coordinated control. The collected charging demand data includes the arrival and departure times of the vehicles, the initial state of charge (SOC) when connected to the grid, and the upper and lower limits of SOC that meet the battery's operational safety boundary and the user's charging requirements.

[0023] To simplify the computational complexity of the model, the travel patterns of all electric vehicle users are assumed to be stable, and the daily arrival and departure times are set to fixed values. Within the same charging and discharging time window, the initial SOC of each electric vehicle when it connects to the smart building cluster is modeled using a normal distribution with a mean μ = 40% and a standard deviation σ = 0.14.

[0024] The carbon emission model for electric vehicles is shown in equation (2):

[0025]

[0026] In the formula: Let P be the total carbon emissions of all electric vehicles at time t; t EV N represents the total charging and discharging power of all electric vehicles at time t; EV The total number of electric vehicles within the smart building cluster; These represent the charging and discharging power of the nth electric vehicle at time t; e m,t Let m be the nodal carbon potential of node m;

[0027] The constraints for electric vehicles are shown in equation (3):

[0028]

[0029] In the formula: Let be the 0-1 state variables of the electric vehicle charging and discharging at time t; Let be the battery charge of the nth electric vehicle at time t; and Δt represents the charging and discharging efficiency of the nth electric vehicle, respectively; Δt is the length of a unit time period. These represent the upper and lower limits of the charging and discharging power of the nth electric vehicle, respectively. These represent the upper and lower limits of the battery capacity of the nth electric vehicle, respectively. The minimum battery level of the nth electric vehicle when it leaves;

[0030] The carbon emission model for the load reduction is as follows:

[0031]

[0032] In the formula: The amount of carbon emissions reduced by the load at time t; P t cl The power that can be reduced by the load at time t;

[0033] The constraint condition for load reduction is shown in equation (5):

[0034]

[0035] In the formula: Let t be the 0-1 state variable representing the power reduction at time t; To reduce the upper limit of power; These are the start and end times of the power reduction, respectively; To reduce the total power response time; To reduce the upper limit of the response time for power reduction;

[0036] The carbon emission model for the transferable load is as follows:

[0037]

[0038] In the formula: P represents the carbon emissions of the transferable load at time t. t tl,in P t tl,out These represent the transfer-in and transfer-out power of the transferable load at time t, respectively.

[0039] The constraints for transferable loads are shown in equation (7):

[0040]

[0041] In the formula: These are the 0-1 state variables representing the load transfer in and out at time t; These are the upper limits for the power transferred in and out, respectively. These are the start and end times of the power transfer, respectively. These are the start and end times of the power transfer, respectively. These represent the total time periods for power transfer in and out, respectively; Δt represents the unit scheduling time period.

[0042] The carbon emission model for the energy storage is as follows:

[0043]

[0044] In the formula: Let t be the carbon emissions from energy storage; These represent the 0-1 state variables of the ES charging and discharging at time t; P t ch P t dis These are the charging and discharging power of the energy storage, respectively.

[0045] The constraints for energy storage are shown in equation (9):

[0046]

[0047] In the formula: These are the upper limits for energy storage charging and discharging power, respectively. η represents the stored energy at time t+1 and time t, respectively; κ is the self-discharge coefficient of the stored energy; η is the energy stored at time t+1 and time t, respectively. ch and η dis These refer to the energy storage charging and discharging efficiencies, respectively. These are the upper and lower limits of the energy storage capacity, respectively.

[0048] In summary, the total carbon emissions generated by node m in the power grid when connected to the smart building cluster are shown in equation (10):

[0049]

[0050] In the formula: E D The total carbon emissions of the smart building during the scheduling cycle; Let t be the total carbon emissions of the intelligent building cluster at time t; T is the total scheduling period.

[0051] The carbon trading costs for the smart building cluster in step four are as follows:

[0052]

[0053] In the formula: λ represents the carbon trading cost of the smart building cluster; λ is the benchmark carbon price; l is the unit length of the carbon emission range; α is the carbon trading price growth coefficient; E D The total carbon emissions of a smart building during the scheduling cycle.

[0054] The two-level optimal scheduling model for the power system that considers the combined electricity and carbon demand response in step five includes:

[0055] ① Economic dispatch model of upper-level power grid operators

[0056] a. The objective function for the operating cost of the power grid operator is shown in equation (12):

[0057]

[0058] In the formula: C U For the operating costs of power grid operators; C g Operating costs of thermal power units; C is the cost of starting and stopping thermal power units; w C v The costs of wind and solar power generation are respectively; Carbon trading costs for grid operators; N g This refers to the number of thermal power units; a i b i and c i These are the coal consumption cost coefficients for the i-th thermal power unit; The output of the i-th thermal power unit at time t; f is the start-up / shutdown state variable of the i-th thermal power unit at time t; qt,i The start-up and shutdown cost of the i-th thermal power unit; s w s v These represent the unit power generation costs of wind power and solar power, respectively; P t w P t v The output of wind power and photovoltaic power at time t are respectively. Let N be the carbon emission intensity of the i-th thermal power unit; δ be the carbon quota coefficient; N g This refers to the number of thermal power units.

[0059] b. Constraints:

[0060] Thermal power unit constraints:

[0061]

[0062] In the formula: P i g,max P i g,min These represent the maximum and minimum output values ​​of the i-th thermal power unit, respectively; P iu P i d These are the upper and lower limits of the ramp power of the i-th thermal power unit, respectively; T represents the operating and shutdown times of the i-th thermal power unit at time t-1, respectively; i on,min T i off,min These are the shortest operating and downtime times for the i-th thermal power unit, respectively.

[0063] Wind power and solar power output constraints:

[0064]

[0065] In the formula: P w,max P v,max These are the maximum output values ​​for wind power and solar power, respectively.

[0066] Node power balance constraints:

[0067]

[0068] In the formula: Let m be the load value of node m; These represent the power values ​​flowing into and out of line l connected to node m, respectively.

[0069] Line transmission capacity constraints:

[0070] P l min ≤P l ≤P l max (16);

[0071] In the formula: P l max and P l min These are the upper and lower limits of power transmission for line l, respectively;

[0072] ② Lower-level intelligent building cluster response model

[0073] a. The objective function is shown in equation (17):

[0074]

[0075] In the formula: C D C represents the total cost of the intelligent building cluster; buy Indicates the cost of electricity purchase; C cl This indicates that demand response costs can be reduced; C tl Indicates the demand response cost of transferable load; C evd Indicates the cost of electric vehicle discharge subsidies; Ces This indicates the cost of charging and discharging energy storage. The electricity price is based on time-of-use pricing. The load compensation factor can be reduced; The transferable load is the compensation coefficient; This is the compensation coefficient for transferable load. These are the cost coefficients for energy storage charging and discharging, respectively.

[0076] b. Constraints:

[0077] Power balance constraints:

[0078]

[0079] Other constraints:

[0080] The constraints include electric vehicle constraints, load reduction constraints, load transfer constraints, and energy storage constraints, which are respectively shown in equations (3), (5), (7), and (9).

[0081] Through the above design scheme, the present invention can bring the following beneficial effects:

[0082] This invention proposes a two-layer optimal dispatching method for power systems considering joint electricity-carbon demand response. First, a CNN-BiGRU-Attention hybrid neural network is proposed, which is used to establish prediction models suitable for wind and solar power generation forecasting and to perform power prediction. Second, using the proportional-sharing carbon emission flow theory, a carbon emission model for intelligent building clusters, incorporating electric vehicles, load reduction, load transfer, and energy storage, is established to achieve precise quantification of carbon emissions from intelligent building clusters. Then, a two-layer optimal model for the power system considering joint electricity-carbon demand response is established. The upper layer is the economic dispatching model of the grid operator, and the lower layer is the response model of the intelligent building cluster. The upper-layer grid operator aims to minimize operating costs, while the lower-layer intelligent building cluster aims to minimize total costs. The upper and lower layers iteratively solve the problem using node carbon potential, electricity price, and electricity demand. Finally, a two-layer optimal dispatching plan for the power system considering joint electricity-carbon demand response is obtained, providing a more reasonable dispatching scheme that balances economy and low carbon emissions for day-ahead optimal dispatching of the power system. The numerical examples verify that the proposed method can provide a reference for optimal dispatching of the power system. Attached Figure Description

[0083] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0084] Figure 1 The diagram shows the CNN-BiGRU-Attention neural network in the two-layer optimal scheduling method for power systems that considers combined electricity and carbon demand response in this invention.

[0085] Figure 2 This is a diagram of the two-layer optimal scheduling framework for the combined electricity and carbon demand response in the power system two-layer optimal scheduling method of the present invention.

[0086] Figure 3 This is a flowchart of the solution process for the two-level optimal scheduling model in the two-level optimal scheduling method for power systems that considers combined electricity and carbon demand response in this invention.

[0087] Figure 4 This is a schematic diagram of the bisection method iteration in the two-layer optimal scheduling method for power systems that considers combined electricity and carbon demand response in this invention.

[0088] Figure 5 This invention improves the IEEE 30-bus system topology in a two-layer optimal scheduling method for power systems that considers combined electricity and carbon demand response.

[0089] Figure 6 This is a graph showing the carbon potential variation of each smart building cluster access node in the two-layer optimal scheduling method for power systems that considers combined electricity and carbon demand response in this invention.

[0090] Figure 7 The wind power prediction curve in the two-layer optimal dispatch method for power systems that considers the combined electricity and carbon demand response of this invention;

[0091] Figure 8 The photovoltaic power prediction curve in the two-layer optimal dispatch method for power systems that considers combined electricity and carbon demand response in this invention;

[0092] Figure 9 This invention provides the output results of each generating unit of the grid operator in the two-layer optimal dispatching method for power systems that considers combined electricity and carbon demand response.

[0093] Figure 10 This invention presents the response results of each smart building cluster and the carbon potential and time-of-use electricity price diagram of the two-layer optimal scheduling method for power systems that considers the combined electricity and carbon demand response.

[0094] Figure 11 This is a comparison of the response results of each intelligent building cluster and the node carbon potential and time-of-use electricity price diagrams in the embodiments of the present invention, using method 2. Detailed Implementation

[0095] A two-level optimal scheduling method for power systems considering joint electricity-carbon demand response includes: establishing a prediction model for wind and solar power generation based on a CNN-BiGRU-Attention neural network; establishing a smart building cluster carbon emission model based on carbon emission flow theory; and establishing a two-level optimal scheduling model for power systems considering joint electricity-carbon demand response. The specific content is as follows:

[0096] 1) Establish a prediction model for wind and solar power generation based on CNN-BiGRU-Attention neural network.

[0097] The wind and solar power generation prediction model based on CNN-BiGRU-Attention neural network includes a CNN layer, a Bidirectional Gated Recurrent Unit (BiGRU) layer, an attention layer, a fully connected layer, and an output layer connected in sequence; wherein, the CNN layer includes a convolutional layer I, a pooling layer I, a convolutional layer II, and a pooling layer II connected in sequence.

[0098] A wind and solar power generation prediction model built using a CNN-BiGRU-Attention neural network significantly improves the prediction accuracy of wind and solar power generation through a three-level collaborative mechanism of multi-scale feature extraction, bidirectional temporal modeling, and dynamic weight focusing, providing reliable decision support for the safe dispatch and economical operation of the power system. The specific prediction process is as follows:

[0099] Step 1: Input the original historical data of wind power and photovoltaic power, parse the timestamps, extract the time features of month, weekend, and hour, normalize the time features and power data respectively, fill in the missing values ​​and divide the dataset into training set and validation set;

[0100] Step 2: Input the training set into the CNN layer, use the convolutional layer to extract the local spatiotemporal joint feature information of wind power and photovoltaic power data and time features, and then use the pooling layer to select key features and retain the most significant feature combination.

[0101] Step 3: Pass the high-order spatiotemporal features extracted by the CNN layer to the BiGRU neural network, and use its bidirectional gated recurrent unit structure to learn the long-term dependencies of the feature sequence, thus expanding the local spatial features into dynamic features with temporal dependencies.

[0102] Step 4: Adaptively assign weights to the time-dependent dynamic features output by the BiGRU layer using the Attention mechanism, focusing on key time step information, and output the weighted feature vector to the fully connected layer;

[0103] Step 5: The feature vectors weighted by the attention mechanism are nonlinearly combined and dimensionality reduced through a fully connected layer to map the high-dimensional feature space to the prediction dimension. Finally, inverse normalization is performed to output the predicted power of wind and solar power generation including timestamps.

[0104] Step 6: If the evaluation index reaches the set threshold after validation with the validation set, the prediction model for wind and solar power generation based on the CNN-BiGRU-Attention neural network is established.

[0105] 2) Establish a carbon emission model for intelligent building clusters based on carbon emission flow theory.

[0106] Using the carbon emission flow theory based on the proportional sharing principle, starting from the power supply side, the carbon potential of each node is calculated based on the DC power flow model, and then the node carbon emission is quantified by combining the node load power, thus establishing a precise dynamic carbon emission tracking method from power supply to load.

[0107] According to the proportional-sharing carbon emission flow theory, the carbon potential of node m at time t is determined by two types of carbon sources: one is the carbon emissions generated by the generators directly connected to the node, and the other is the carbon flow transmitted from other nodes to the node. The calculation formula is shown in equation (1):

[0108]

[0109] In the formula: e m,t Let ρ be the nodal carbon potential of node m; l,t The carbon flux density of branch l at time t is equal to the nodal carbon potential of the first segment of the branch; P l.t Let be the active power of branch l at time t; Let P be the carbon emission intensity of the i-th generator unit; m,t L represents the unit output value at node m at time t; m For the set of lines connected to node m;

[0110] After calculating the carbon potential of each node using carbon emission flow theory, the carbon emissions of the intelligent building cluster are calculated using the node carbon potential, thus establishing a carbon emission model for the intelligent building cluster. Assuming the intelligent building cluster is connected to node m in the system, the carbon emission model is as follows:

[0111] ① Carbon emission model of electric vehicles

[0112] Once an individual electric vehicle user and the smart building cluster reach an incentive compensation agreement, the building cluster has the right to uniformly schedule the charging and discharging behavior of each connected electric vehicle within the agreed time period and operating conditions. The smart building cluster implements centralized scheduling and coordinated control by collecting real-time charging demand data from each electric vehicle. The collected information includes the vehicle's arrival and departure times, its initial state of charge (SOC) upon grid connection, and the upper and lower limits of SOC constraints that meet battery operating safety boundaries and user charging requirements.

[0113] To simplify the computational complexity of the model, it is assumed that the travel patterns of all electric vehicle users are stable, and the daily arrival and departure times are set to fixed values. However, even within the same charging and discharging time window, the initial SOC of each electric vehicle when connecting to the smart building cluster will vary due to factors such as mileage and driving habits. To accurately characterize this randomness, a normal distribution is used to model the initial SOC, with a mean μ = 40% and a standard deviation σ = 0.14.

[0114] The carbon emission model for electric vehicles is shown in equation (2):

[0115]

[0116] In the formula: Let P be the total carbon emissions of all electric vehicles at time t; t EV N represents the total charging and discharging power of all electric vehicles at time t; EV The total number of electric vehicles within the smart building cluster; Let be the charging and discharging power of the nth electric vehicle at time t, respectively.

[0117] The constraints that electric vehicles must meet are shown in equation (3):

[0118]

[0119] In the formula: Let be the 0-1 state variables of the electric vehicle charging and discharging at time t; Let be the battery charge of the nth electric vehicle at time t; and Δt represents the charging and discharging efficiency of the nth electric vehicle, respectively; Δt is the length of a unit time period. These represent the upper and lower limits of the charging and discharging power of the nth electric vehicle, respectively. These represent the upper and lower limits of the battery capacity of the nth electric vehicle, respectively. The minimum battery level of the nth electric vehicle when it leaves;

[0120] ② Carbon emission models that can reduce load

[0121] Reduceable loads refer to a type of load that can reduce power consumption according to dispatch requirements during peak load periods in the power system. This type of load actively reduces its electricity demand by responding to control commands, which helps maintain the balance between power grid supply and demand and ensures stable system operation. The carbon emission model is shown in equation (4):

[0122]

[0123] In the formula: The amount of carbon emissions reduced by the load at time t; P tcl The power that can be reduced by the load at time t;

[0124] The constraints that the load can be reduced are shown in equation (5):

[0125]

[0126] In the formula: Let t be the 0-1 state variable representing the power reduction at time t; To reduce the upper limit of power; These are the start and end times of the power reduction, respectively; To reduce the total power response time; To reduce the upper limit of the response time for power reduction;

[0127] ③ Carbon emission model for transferable loads

[0128] Transferable loads are a type of load that can be flexibly adjusted in time, and their electricity consumption can be shifted from one time period to another. Their carbon emission model is shown in equation (6):

[0129]

[0130] In the formula: P represents the carbon emissions of the transferable load at time t. t tl,in P t tl,out These represent the transfer-in and transfer-out power of the transferable load at time t, respectively.

[0131] The constraints that the transferable load must satisfy are shown in equation (7):

[0132]

[0133] In the formula: These are the 0-1 state variables representing the load transfer in and out at time t; These are the upper limits for the power transferred in and out, respectively. These are the start and end times of the power transfer, respectively. These are the start and end times of the power transfer, respectively. These represent the total time periods for power transfer in and out, respectively.

[0134] ④ Carbon emission models for energy storage

[0135] To ensure the power supply reliability of intelligent building clusters and improve the absorption capacity of renewable energy, a certain capacity of energy storage is usually configured within the cluster. Its carbon emission model is shown in equation (8):

[0136]

[0137] In the formula: Let t be the carbon emissions from energy storage; These represent the 0-1 state variables of the ES charging and discharging at time t; P t ch P t dis These are the charging and discharging power of the energy storage, respectively.

[0138] The constraints that energy storage must meet are shown in equation (9):

[0139]

[0140] In the formula: The charging and discharging power of energy storage are respectively limit; η represents the stored energy at time t+1 and time t, respectively; κ is the self-discharge coefficient of the stored energy; η is the energy stored at time t+1 and time t, respectively. ch and η dis These refer to the energy storage charging and discharging efficiencies, respectively. These are the upper and lower limits of the energy storage capacity, respectively.

[0141] In summary, if node m in the power grid is connected to the smart building cluster, the total carbon emissions generated by the smart building cluster at that node are shown in equation (10):

[0142]

[0143] In the formula: E D The total carbon emissions of the smart building during the scheduling cycle; Let T be the total carbon emissions of the intelligent building cluster at time t; T is the total scheduling period.

[0144] To effectively constrain system carbon emissions and incentivize smart building clusters to actively participate in emission reduction, this paper designs a tiered carbon trading model based on node carbon potential. This model calculates carbon emissions based on the smart building cluster carbon emission model and sets multiple carbon emission ranges accordingly: as the actual emissions increase, the applicable carbon trading price increases progressively, thus reflecting the carbon cost constraint effect of differentiated pricing. The carbon trading cost of the smart building cluster is shown in equation (11):

[0145]

[0146] In the formula: λ represents the carbon trading cost of the smart building cluster; λ represents the benchmark carbon price; l represents the unit length of the carbon emission range; and α represents the carbon trading price growth coefficient.

[0147] 3) Establish a two-level optimal dispatch model for the power system that considers the combined electricity and carbon demand response.

[0148] ① Economic dispatch model of upper-level power grid operators

[0149] a. Objective function:

[0150] The grid operator's optimization objective is to minimize system operating costs, including the generation and start-up / shutdown costs of thermal power units, the costs of wind and solar power generation, and carbon trading costs. The objective function is shown in equation (12):

[0151]

[0152] In the formula: C U For the operating costs of power grid operators; C g Operating costs of thermal power units; C is the cost of starting and stopping thermal power units; w C v The costs of wind and solar power generation are respectively; Carbon trading costs for grid operators; N g This refers to the number of thermal power units; a i b i and c i These are the coal consumption cost coefficients for the i-th thermal power unit; The output of the i-th thermal power unit at time t; f is the start-up / shutdown state variable of the i-th thermal power unit at time t; qt,i The start-up and shutdown cost of the i-th thermal power unit; s w s v These represent the unit power generation costs of wind power and solar power, respectively; P t w P t v The output of wind power and photovoltaic power at time t are respectively. Let N be the carbon emission intensity of the i-th thermal power unit; δ be the carbon quota coefficient; N g This refers to the number of thermal power units.

[0153] b. Constraints:

[0154] Thermal power unit constraints:

[0155]

[0156] In the formula: P i g,max P i g,min These represent the maximum and minimum output values ​​of the i-th thermal power unit, respectively; P i u P i d These are the upper and lower limits of the ramp power of the i-th thermal power unit, respectively; T represents the operating and shutdown times of the i-th thermal power unit at time t-1, respectively; i on,min T i off,min These are the shortest operating and downtime times for the i-th thermal power unit, respectively.

[0157] Wind power and solar power output constraints:

[0158]

[0159] In the formula: P w,max P v,max These are the maximum output values ​​for wind power and solar power, respectively.

[0160] Node power balance constraints:

[0161]

[0162] In the formula: Let m be the load value of node m; These represent the power values ​​flowing into and out of line l connected to node m, respectively.

[0163] Line transmission capacity constraints:

[0164] P l min ≤P l ≤P l max (16);

[0165] In the formula: P l max and P l min These are the upper and lower limits of power transmission for line l, respectively;

[0166] ② Lower-level intelligent building cluster response model

[0167] a. Objective function:

[0168] The lower-level intelligent building cluster aims to minimize total cost, including electricity purchase cost, demand response cost, and carbon trading cost from the upper level. The objective function is shown in equation (17):

[0169]

[0170] In the formula: C D C represents the total cost of the intelligent building cluster; buy Indicates the cost of electricity purchase; C cl This indicates that demand response costs can be reduced; C tl Indicates the demand response cost of transferable load; C evd Indicates the cost of electric vehicle discharge subsidies; Ces This indicates the cost of charging and discharging energy storage. The electricity price is based on time-of-use pricing. The load compensation factor can be reduced; The transferable load is the compensation coefficient; This is the compensation coefficient for transferable load. These are the cost coefficients for energy storage charging and discharging, respectively.

[0171] b. Constraints:

[0172] Power balance constraints:

[0173]

[0174] Other constraints:

[0175] The constraints include electric vehicle constraints, load reduction constraints, load transfer constraints, and energy storage constraints, as shown in equations (3), (5), (7), and (9), respectively.

[0176] Example:

[0177] The following uses appendix Figures 1-11 The invention will be further illustrated by the embodiments.

[0178] A two-level optimal scheduling method for power systems considering joint electricity-carbon demand response includes: establishing a prediction model for wind and solar power generation based on a CNN-BiGRU-Attention neural network; establishing a smart building cluster carbon emission model based on carbon emission flow theory; and establishing a two-level optimal scheduling model for power systems considering joint electricity-carbon demand response. The specific content is as follows:

[0179] 1) Parameter setting and wind and solar power prediction

[0180] The parameters of the thermal power unit and other parameters are shown in Tables 1-7;

[0181] Table 1 Relevant parameters of electric vehicles

[0182]

[0183] Table 2 Parameters of Reduced Load Incentive Contracts

[0184]

[0185] Table 3 Transferable Load Incentive Contract Parameters

[0186]

[0187] Table 4 Energy Storage Related Parameters

[0188]

[0189] Table 5 Time-of-use Electricity Price List

[0190]

[0191]

[0192] Table 6 Parameter Table for Thermal Power Units

[0193]

[0194] Table 7 Other parameters

[0195]

[0196] Figure 2 and Figure 3 The diagrams show the framework of the two-layer optimal scheduling for electricity-carbon joint demand response and the flowchart of the solution process for the two-layer optimal scheduling model. The upper-layer grid operator's economic scheduling model aims to minimize the operating cost of the power system, optimizes the unit generation plan, and calculates the carbon potential of each node using CEF theory. The carbon potential signal and time-of-use electricity price information are then sent to the corresponding smart building cluster. The lower-layer smart building cluster optimizes its electricity consumption behavior based on the received electricity-carbon price signal, with the goal of minimizing the total cost, and feeds back the updated electricity demand to the upper layer. The upper layer readjusts the generation plan and updates the node carbon potential based on the feedback. This process is iterated until the convergence condition is met, ultimately achieving collaborative optimal scheduling.

[0197] Figure 4 This is a schematic diagram of the bisection method iteration. To enhance the stability of the algorithm and avoid oscillations, the bisection method is introduced to constrain the range of load demand variation.

[0198] Figure 5 To improve the topology of the IEEE 30-node system, nodes 1, 2, 5, and 8 are each connected to a thermal power unit, node 11 is connected to a wind power station, and node 13 is connected to a photovoltaic power station.

[0199] Figure 6 The graph shows the carbon potential changes of the access nodes of each smart building cluster. To verify the universality of this invention, three smart building clusters were accessed at nodes 26, 14, and 24 respectively. The carbon potential of node 26 did not change much in the morning, noon, and evening. The carbon potential of node 14 was high in the morning and evening and low at noon. The carbon potential of node 24 was high in the morning and evening and low at noon.

[0200] Figure 7 and Figure 8Tables 8 and 9 show the prediction curves of the prediction model, the prediction curve of the CNN-BiGRU-Attention neural network, and the actual power curve, respectively. The wind power installed capacity is 500MW, and the photovoltaic installed capacity is 450MW. Comparisons of wind power prediction accuracy and photovoltaic power prediction accuracy are shown in Tables 8 and 9, respectively. Compared to CNN-GRU-Attention, the prediction method of this invention achieves higher R... 2 These figures represent increases of 3.43% and 3.34% respectively; MAE decreased by 27.85% and 8.98% respectively; and RMSE decreased by 28.17% and 7.63% respectively, thus improving forecast accuracy.

[0201] Table 8 Comparison of Wind Power Prediction Accuracy

[0202]

[0203] Table 9 Comparison of Photovoltaic Power Prediction Accuracy

[0204]

[0205] 2) Solve the two-level optimal dispatch model of the power system considering the combined electricity and carbon demand response.

[0206] Three different scheduling methods were designed and compared. The design methods are as follows:

[0207] This invention sets up three methods for comparative analysis to verify the correctness and effectiveness of the scheduling method proposed in this invention;

[0208] (1) The method of the present invention: a two-layer optimization scheduling strategy that considers time-of-use electricity pricing and tiered carbon trading to guide intelligent building clusters to conduct joint electricity-carbon demand response;

[0209] (2) Comparison with Method 1: Considering fixed electricity prices, the smart building cluster does not participate in carbon trading and does not implement demand response optimization scheduling strategies;

[0210] (3) Comparison Method 2: Consider the time-of-use pricing to guide smart building clusters to respond to electricity price-based demand, and optimize the scheduling strategy in which each smart building cluster does not participate in carbon trading. Figures 7-11 To optimize the power system using the day-ahead optimization scheduling method based on the evolved virtual net load, and the optimization scheduling methods 1, 2, 3 and 4 respectively, the output curves of each unit are obtained.

[0211] Figure 9 The method of this invention represents the power output results of each generating unit of the power grid operator. Figure 10 The diagram shows the response results of each intelligent building cluster according to the method of this invention, along with the node carbon potential and time-of-use electricity price. Figure 11To compare the response results of each smart building cluster with the node carbon potential and time-of-use electricity price map of Method 2. Figure 10 and Figure 11 In the diagram, (a) represents intelligent building cluster 1 (abbreviated as cluster 1), (b) represents intelligent building cluster 2 (abbreviated as cluster 2), and (c) represents intelligent building cluster 3 (abbreviated as cluster 3). The electricity-carbon joint demand response proposed in this invention not only reduces electricity consumption during peak-hour electricity price periods and increases electricity consumption during off-peak hours for each intelligent building cluster, scheduling electric vehicles and energy storage to discharge during peak hours and store energy during off-peak hours, thus reducing electricity purchase costs; but also adjusts the electricity consumption behavior of intelligent building clusters according to carbon potential, shifting the electricity consumption behavior of each intelligent building cluster to periods with lower electricity prices and carbon potential, and shifting the discharge behavior to periods with higher electricity prices and carbon potential, thus balancing the system's economic efficiency and low-carbon characteristics.

[0212] Table 10 compares the costs and carbon emissions of each smart building cluster after optimization using different scheduling methods. The method of this invention performs electricity-carbon joint demand response. Compared with comparative method 1 and comparative method 2, although it increases carbon trading costs, the carbon emissions are the lowest among the three methods. Moreover, the cost of purchasing electricity is almost the same as that of comparative method 2, which can simultaneously take into account both low carbon emissions and economic efficiency.

[0213] Table 10 Comparison of Costs and Carbon Emissions for Different Smart Building Clusters Using Different Methods

[0214]

[0215] The above comparative analysis shows that:

[0216] In comparison method 1, each smart building cluster does not engage in demand response or participate in carbon trading, and does not regulate the flexible load on the demand side, resulting in the highest carbon emissions and electricity purchase costs for each cluster. In comparison method 2, each smart building cluster accepts time-of-use pricing guidance to engage in price-based demand response. Although the electricity purchase cost of comparison method 2 is lower, the carbon emissions are much higher than those of the method of this invention.

[0217] The method of this invention performs a combined electricity-carbon demand response. Compared with comparative method 1 and comparative method 2, although it increases carbon trading costs, the carbon emissions are the lowest among the three methods, and the cost of purchasing electricity is almost the same as that of comparative method 2. It can simultaneously take into account both low carbon emissions and economic efficiency.

[0218] The proposed CNN-BiGRU-Attention neural network forms a three-level collaborative mechanism of multi-scale feature extraction, bidirectional temporal modeling, and dynamic weight focusing, which can effectively cope with the strong nonlinearity, high volatility, and complex temporal dependence characteristics in wind and solar power generation prediction, and improve the prediction accuracy of wind and solar power generation. Based on carbon emission flow theory, a carbon emission model of intelligent building clusters including electric vehicles, load reduction, load transfer, and energy storage is established, which transfers carbon emission responsibility to the load side. Through the joint scheduling of electricity price and carbon price, intelligent building clusters can carry out electricity-carbon joint demand response, which can effectively reduce carbon emissions and reduce electricity purchase costs, taking into account both economic efficiency and low carbon emissions. At the same time, it effectively solves the problem that it is difficult to incentivize load side carbon reduction by relying solely on electricity price signals.

[0219] The specific embodiments used in this invention have provided a detailed description of the invention, but are not limited to these embodiments. Any obvious modifications made by those skilled in the art based on the teachings of this invention are within the scope of protection of this invention.

Claims

1. A two-level optimal dispatch method for power systems considering combined electricity and carbon demand response, characterized in that: Includes the following steps, And the following steps are performed in sequence: Step 1: Establish a prediction model for wind and solar power generation based on CNN-BiGRU-Attention neural network, and output the predicted wind and solar power generation power including timestamps through this model; Step 2: Establish a carbon emission model for intelligent building clusters based on carbon emission flow theory, and obtain the carbon potential of each node through this model; Step 3: Calculate the carbon emissions of the smart building cluster using the carbon potential of each node, establish a carbon emission model for the smart building cluster, and obtain the total carbon emissions generated by the smart building cluster for each node; the carbon emission model for the smart building cluster includes a carbon emission model for electric vehicles, a carbon emission model for load reduction, a carbon emission model for load transfer, and a carbon emission model for energy storage. Step 4: Obtain the carbon trading cost of the smart building cluster; Step 5: Establish a two-layer optimal dispatch model for the power system that considers the joint demand response of electricity and carbon. This two-layer optimal dispatch model includes an upper-layer economic dispatch model for grid operators and a lower-layer intelligent building cluster response model. The economic dispatch model for grid operators aims to minimize the operating cost of grid operators, while the lower-layer intelligent building cluster response model aims to minimize the total cost. The upper and lower layers are solved iteratively through node carbon potential, electricity price, and electricity demand to obtain a two-layer optimal dispatch plan for the power system that considers the joint demand response of electricity and carbon.

2. The two-level optimal dispatching method for power systems considering combined electricity and carbon demand response as described in claim 1, characterized in that: The specific steps for establishing the wind and solar power generation prediction model based on the CNN-BiGRU-Attention neural network in step one are as follows: Step 1: The prediction model for wind and solar power generation based on CNN-BiGRU-Attention neural network includes a convolutional neural network (CNN) layer, a bidirectional gated recurrent unit (BiGRU) layer, an attention layer, a fully connected layer, and an output layer connected in sequence; wherein, the CNN layer includes a convolutional layer I, a pooling layer I, a convolutional layer II, and a pooling layer II connected in sequence. Extract the original historical data of wind power and photovoltaic power, parse the timestamps, extract the time features of each month, each weekend, and each hour, normalize the time features and power data respectively, fill in the missing values, and divide the dataset into training set and validation set. Step 2: Input the training set into the CNN layer, where convolutional layer I and convolutional layer II are used to extract local spatiotemporal joint feature information of wind power and photovoltaic power data and time features, and pooling layer I and pooling layer II are used to select key features. After two feature extractions and key feature selections by the CNN layer, the most significant feature combination is retained to form high-order spatiotemporal features. Step 3: Pass the high-order spatiotemporal features extracted by the CNN layer to the BiGRU layer, and use the bidirectional gated recurrent unit structure of the BiGRU layer to learn the long-term dependencies of the feature sequence, thus expanding the local spatial features into dynamic features with temporal dependencies. Step 4: In the attention layer, the attention mechanism is used to adaptively assign weights to the time-dependent dynamic features output by BiGRU, focusing on key time step information, and the weighted feature vector is output to the fully connected layer; Step 5: The feature vectors weighted by the attention mechanism are nonlinearly combined and dimensionality reduced through a fully connected layer to map the high-dimensional feature space to the prediction dimension. Finally, inverse normalization is performed to output the predicted power of wind and solar power generation including timestamps. Step 6: After validation on the validation set, the prediction model for wind and solar power generation based on the CNN-BiGRU-Attention neural network is completed.

3. The two-level optimal dispatching method for power systems considering combined electricity and carbon demand response as described in claim 1, characterized in that: The formulas for calculating the carbon potential of each node in step two are as follows: In the formula: e m,t Let ρ be the nodal carbon potential of node m; l,t The carbon flux density of branch l at time t is equal to the nodal carbon potential of the first segment of the branch; P l.t Let be the active power of branch l at time t; Let P be the carbon emission intensity of the i-th generator unit; m,t L represents the unit output value at node m at time t; m This is the set of lines connected to node m.

4. The two-level optimal dispatching method for power systems considering combined electricity and carbon demand response as described in claim 3, characterized in that: The process of establishing the carbon emission model for electric vehicles in step three is as follows: The intelligent building cluster collects real-time charging demand data from each electric vehicle and implements centralized scheduling and coordinated control. The collected charging demand data includes the arrival and departure times of the vehicles, the initial state of charge (SOC) when connected to the grid, and the upper and lower limits of SOC that meet the battery's operational safety boundary and the user's charging requirements. To simplify the computational complexity of the model, the travel patterns of all electric vehicle users are assumed to be stable, and the daily arrival and departure times are set to fixed values. Within the same charging and discharging time window, the initial SOC of each electric vehicle when it connects to the smart building cluster is modeled using a normal distribution with a mean μ = 40% and a standard deviation σ = 0.

14. The carbon emission model for electric vehicles is shown in equation (2): In the formula: Let P be the total carbon emissions of all electric vehicles at time t; t EV N represents the total charging and discharging power of all electric vehicles at time t; EV The total number of electric vehicles within the smart building cluster; These represent the charging and discharging power of the nth electric vehicle at time t; e m,t Let m be the nodal carbon potential of node m; The constraints for electric vehicles are shown in equation (3): In the formula: Let be the 0-1 state variables of the electric vehicle charging and discharging at time t; Let be the battery charge of the nth electric vehicle at time t; and Δt represents the charging and discharging efficiency of the nth electric vehicle, respectively; Δt is the length of a unit time period. These represent the upper and lower limits of the charging and discharging power of the nth electric vehicle, respectively. These represent the upper and lower limits of the battery capacity of the nth electric vehicle, respectively. The minimum battery level of the nth electric vehicle when it leaves; The carbon emission model for the load reduction is as follows: In the formula: The amount of carbon emissions reduced by the load at time t; P t cl The power that can be reduced by the load at time t; The constraint condition for load reduction is shown in equation (5): In the formula: Let t be the 0-1 state variable representing the power reduction at time t; To reduce the upper limit of power; These are the start and end times of the power reduction, respectively; To reduce the total power response time; To reduce the upper limit of the response time for power reduction; The carbon emission model for the transferable load is as follows: In the formula: P represents the carbon emissions of the transferable load at time t. t tl,in P t tl,out These represent the transfer-in and transfer-out power of the transferable load at time t, respectively. The constraints for transferable loads are shown in equation (7): In the formula: These are the 0-1 state variables representing the load transfer in and out at time t; These are the upper limits for the power transferred in and out, respectively. These are the start and end times of the power transfer, respectively. These are the start and end times of the power transfer, respectively. These represent the total time periods for power transfer in and out, respectively. The carbon emission model for the energy storage is as follows: In the formula: Let t be the carbon emissions from energy storage; These represent the 0-1 state variables of the ES charging and discharging at time t; P t ch P t dis These are the charging and discharging power of the energy storage, respectively. The constraints for energy storage are shown in equation (9): In the formula: These are the upper limits for energy storage charging and discharging power, respectively. η represents the stored energy at time t+1 and time t, respectively; κ is the self-discharge coefficient of the stored energy; η is the energy stored at time t+1 and time t, respectively. ch and η dis These refer to the energy storage charging and discharging efficiencies, respectively. These are the upper and lower limits of the energy storage capacity, respectively. In summary, the total carbon emissions generated by node m in the power grid when connected to the smart building cluster are shown in equation (10): In the formula: E D The total carbon emissions of the smart building during the scheduling cycle; Let t be the total carbon emissions of the intelligent building cluster at time t; T is the total scheduling period.

5. The two-level optimal dispatching method for power systems considering combined electricity and carbon demand response as described in claim 4, characterized in that: The carbon trading costs for the smart building cluster in step four are as follows: In the formula: λ represents the carbon trading cost of the smart building cluster; λ is the benchmark carbon price; l is the unit length of the carbon emission range; α is the carbon trading price growth coefficient; E D The total carbon emissions of a smart building during the scheduling cycle.

6. The two-level optimal dispatching method for power systems considering combined electricity and carbon demand response as described in claim 5, characterized in that: The two-level optimal scheduling model for the power system that considers the combined electricity and carbon demand response in step five includes: ① Economic dispatch model of upper-level power grid operators a. The objective function for the operating cost of the power grid operator is shown in equation (12): In the formula: C U For the operating costs of power grid operators; C g Operating costs of thermal power units; C is the cost of starting and stopping thermal power units; w C v The costs of wind and solar power generation are respectively; Carbon trading costs for grid operators; N g This refers to the number of thermal power units; a i b i and c i These are the coal consumption cost coefficients for the i-th thermal power unit; The output of the i-th thermal power unit at time t; f is the start-up / shutdown state variable of the i-th thermal power unit at time t; qt,i The start-up and shutdown cost of the i-th thermal power unit; s w s v These represent the unit power generation costs of wind power and solar power, respectively; P t w P t v The output of wind power and photovoltaic power at time t are respectively. Let N be the carbon emission intensity of the i-th thermal power unit; δ be the carbon quota coefficient; N g This refers to the number of thermal power units. b. Constraints: Thermal power unit constraints: In the formula: P i g,max P i g,min These represent the maximum and minimum output values ​​of the i-th thermal power unit, respectively; P i u P i d These are the upper and lower limits of the ramp power of the i-th thermal power unit, respectively; T represents the operating and shutdown times of the i-th thermal power unit at time t-1, respectively; i on ,min T i off,min These are the shortest operating and downtime times for the i-th thermal power unit, respectively. Wind power and solar power output constraints: In the formula: P w,max P v,max These are the maximum output values ​​for wind power and solar power, respectively. Node power balance constraints: In the formula: Let m be the load value of node m; These represent the power values ​​flowing into and out of line l connected to node m, respectively. Line transmission capacity constraints: P l min ≤P l ≤P l max (16); In the formula: P l max and P l min These are the upper and lower limits of power transmission for line l, respectively; ② Lower-level intelligent building cluster response model a. The objective function is shown in equation (17): In the formula: C D C represents the total cost of the intelligent building cluster; buy Indicates the cost of electricity purchase; C cl This indicates that demand response costs can be reduced; C tl Indicates the demand response cost of transferable load; C evd Indicates the cost of electric vehicle discharge subsidies; C es This indicates the cost of charging and discharging energy storage. The electricity price is based on time-of-use pricing. The load compensation factor can be reduced; The transferable load is the compensation coefficient; This is the compensation coefficient for transferable load. These are the cost coefficients for energy storage charging and discharging, respectively. b. Constraints: Power balance constraints: Other constraints: The constraints include electric vehicle constraints, load reduction constraints, load transfer constraints, and energy storage constraints, which are respectively shown in equations (3), (5), (7), and (9).

Images