A multi-source collaborative power system rolling scheduling method, system, device and storage medium based on bipartite graph-graph convolutional neural network

Abstract

The present application relates to the technical field of power system scheduling, in particular to a multi-source collaborative power system rolling scheduling method, system, device and storage medium based on bipartite graph-graph convolutional neural network. The present application converts the multi-source collaborative power system scheduling model into a mixed integer programming model, first establishes an intra-day planning prediction model and a rolling planning prediction model and trains them, predicts the power generation plan of each generator set through the intra-day planning prediction model, fixes the set predicted to be started as started in the corresponding period of the rolling planning prediction model, realizes the addition of the start period constraint of the set, models the rolling planning model with the added constraint into a bipartite graph, predicts the scheduling plan of the rolling planning model through the rolling planning prediction model, fixes the set predicted to be started as a constraint in the corresponding period of the subsequent rolling planning model, and then outputs the final scheduling plan. The present application improves the speed of obtaining the scheduling plan.

Description

Technical Field

[0001] This invention relates to the field of power system dispatching technology, and in particular to a method, system, device and storage medium for multi-source collaborative power system rolling dispatching based on a bipartite graph-graph convolutional neural network. Background Technology

[0002] The economic dispatch problem of power systems is one of the most fundamental optimization problems in power systems and electricity markets. The main objective of the economic dispatch problem is to find a dispatch plan one day in advance for the on / off states of generating units. This dispatch plan achieves the minimum generation cost while ensuring compliance with various constraints related to generation, transmission, and system stability.

[0003] Researchers are currently striving to shorten the execution time of various power system economic dispatch problems to obtain additional security and economic benefits. This includes having finer time granularity and a longer field of view in model formulas, as well as opportunities to explore scenario-based methods. Therefore, there is a need for a multi-source collaborative rolling dispatch method, system, equipment, and storage medium for power systems based on bipartite graph-graph convolutional neural networks. Summary of the Invention

[0004] To address the issue of low solution speed for the economic dispatch problem of power systems in existing technologies, this invention provides a method, system, equipment, and storage medium for multi-source cooperative rolling dispatch of power systems based on a bipartite graph-graph convolutional neural network. The specific technical solution is as follows:

[0005] A rolling dispatch method for multi-source cooperative power systems based on bipartite graph-graph convolutional neural networks includes the following steps:

[0006] Step S1: Establish a multi-source collaborative power system dispatch model, including an intraday planning model with a time period of 1 hour and a cycle of 1 day, and a rolling planning model with a time period of 15 minutes and a cycle of 3 hours; and convert the intraday planning model and the rolling planning model into mixed integer programming models respectively; and model the intraday planning model that needs to be solved into a bipartite graph.

[0007] Step S2: Establish an intraday planning forecasting model and a rolling planning forecasting model. Collect historical intraday planning load data and historical rolling planning load data, and generate datasets to train the intraday planning forecasting model and the rolling planning forecasting model, obtaining trained intraday planning forecasting model and trained rolling planning forecasting model, which are used to predict the scheduling plan of the intraday planning model and the scheduling plan of the rolling planning model, respectively; both the intraday planning forecasting model and the rolling planning forecasting model adopt graph convolutional neural networks;

[0008] Step S3: Predict the scheduling plan of the intraday planning model through the trained intraday planning prediction model to obtain the power generation plan of each generator unit. Fix the generator units predicted to be started in the corresponding time period of the rolling planning prediction model to be started, which is achieved by adding the start-up time period constraint of the generator units. No operation is performed on the generator units predicted to be shut down.

[0009] Step S4: Model the rolling plan model with added unit start-up time constraints into a bipartite graph, predict the scheduling plan of the rolling plan model through the trained rolling plan prediction model, and fix the corresponding time period of the predicted start-up units in the subsequent rolling plan model as constraints.

[0010] Step S5: Output the final scheduling plan.

[0011] Preferably, the constraints of the multi-source coordinated power system dispatch model include operational constraints for wind, solar, hydro, thermal, nuclear, and energy storage power sources, and the objective function is as follows:

[0012]

[0013] Where T is the total number of time periods in the scheduling cycle, T is 24 for the intraday planning model and 12 for the rolling planning model; These represent the costs of wind and solar power curtailment during time period t. Let i be the power generation cost and start-up cost of thermal power unit i in time period t; Let i be the cost of decommissioning nuclear power unit i during time period t;

[0014] The costs of curtailing wind, solar, and nuclear power are calculated as follows:

[0015]

[0016]

[0017]

[0018] Where, δ W δ PV δ N These are the unit costs for wind curtailment, solar curtailment, and nuclear curtailment, respectively.

[0019] These represent the predicted available power for wind power and solar power in time period t, respectively.

[0020] P t W P t PV These represent the actual dispatched power of wind power and solar power during time period t, respectively.

[0021] P represents the actual dispatched power of nuclear power unit i during time period t;i N,max This is the maximum allowable output of nuclear power unit i.

[0022] Preferably, the power generation cost of a thermal power unit is calculated as follows:

[0023]

[0024] Among them, a i b i c i The correlation coefficient for the operating costs of thermal power units. This indicates the output of thermal power unit i during time period t;

[0025] right Perform a projection transformation to obtain

[0026]

[0027] Among them, P i T,max P i T,min The upper and lower limits of the output of thermal power unit i; u i,t Let i be the state variable of thermal power unit i during time period t, where "1" indicates operation and "0" indicates shutdown.

[0028] but Represented as:

[0029]

[0030]

[0031]

[0032]

[0033] Start-up cost of thermal power units

[0034] in, The hot start cost of thermal power unit i; The portion of the start-up cost of thermal power unit i that exceeds the hot start cost; v i,t This represents the startup status of thermal power unit i during time period t; 1 indicates unit startup and 0 indicates otherwise.

[0035] Daily planned power generation constraints for thermal power units:

[0036]

[0037] Daily planned power generation constraints for nuclear power units:

[0038]

[0039] Among them, T day This refers to the number of time slots planned for the day; These are the daily planned power generation limits for thermal power unit i and nuclear power unit i, respectively. These are the daily planned power limits for thermal power unit i and nuclear power unit i, respectively.

[0040] Preferably, the operational constraints of the wind, solar, hydro, thermal, nuclear, and energy storage power sources are as follows:

[0041] (1) Wind power operation constraints:

[0042]

[0043] (2) Photoelectric operation constraints:

[0044]

[0045] (3) Constraints on hydropower operation:

[0046] a. State constraints of pumped storage power stations:

[0047]

[0048] in, The variable is a binary variable representing the energy storage state of a pumped storage power station. It is 1 when the power station is in the energy storage state and 0 otherwise. The variable is a binary variable representing the power generation status of a pumped storage power station. It is 1 when the power station is generating electricity and 0 otherwise.

[0049] b. Power generation constraints and pumping power constraints of pumped storage power stations:

[0050]

[0051]

[0052] Among them, P t H,in P represents the electricity consumed by a pumped-storage power station during time period t for pumping water. Hin,max P Hin,min These represent the upper and lower limits of the electricity consumption of a pumped storage power station in a single time period; P t H,out P represents the output power of a pumped-storage hydroelectric power station during time period t. Hout,max P Hout,min These represent the upper and lower limits of the electricity output of a pumped storage power station during a single time period.

[0053] c. Reservoir capacity constraints of pumped storage power stations:

[0054] V min ≤V t ≤V max ;

[0055]

[0056] Among them, V t V represents the reservoir capacity of a pumped storage power station during time period t; max V min These represent the upper and lower limits of the reservoir capacity of a pumped storage power station; V t+1 η represents the reservoir capacity of the pumped storage power station at time t+1. in η out These are the pumping efficiency and power generation efficiency of the pumped storage power station's reservoir capacity, respectively.

[0057] (4) Constraints on thermal power plant operation:

[0058] a. Binary variable logical constraints for thermal power units:

[0059] v i,t -w i,t =u i,t -u i,t-1 ;

[0060] Among them, w i,t The shutdown status of thermal power unit i during time period t is 1 if the unit is shut down, and 0 otherwise.

[0061] b. Initial state constraints:

[0062] u i,t =u i,0 , t∈[1,…,U i +L i ];

[0063] U i =[min[T,u i,0 ( T on,i -T i,0 )]] + ;

[0064] L i =[min[T,(1-u i,0 ()( T off,i +T i,0 )]] + ;

[0065] Among them, u i,0 This represents the state of thermal power unit i during the initial period, i.e., the period before the start of the dispatch cycle, where "1" indicates operation and "0" indicates shutdown; U iL represents the time that thermal power unit i still needs to run at the initial moment. i This indicates the time that thermal power unit i still needs to be shut down at the initial moment; T is the total number of scheduling periods, [·] + This represents max(0,·); T on,i This represents the minimum start-up time of thermal power unit i; T off,i T represents the minimum shutdown time of thermal power unit i; i,0 This represents the number of time periods during which thermal power unit i has been continuously started or shut down before the start of the dispatching cycle; T represents the number of time periods during which thermal power unit i has been continuously started before the start of the dispatching cycle. i,0 When T is positive, it remains positive during continuous shutdown. i,0 Negative;

[0066] c. Minimum start-up and shutdown constraints:

[0067]

[0068]

[0069] d. Startup cost constraints:

[0070]

[0071] in, The cold start cost of thermal power unit i; The hot start cost of thermal power unit i; T cold,i f is the cold start time of thermal power unit i; init,i,t It is a constant, if τ- T off,i - T cold,i ≤0 and [-T i,0 ] + <|τ- T off,i - T cold,i -1|+1, then f init,i,t =1, indicating that thermal power unit i does not belong to cold start at time t; otherwise, f init,i,t =0 indicates that thermal power unit i is in a cold start at time t;

[0072] e. Upper and lower limits of thermal power unit output:

[0073] Based on the fundamental data of the multi-source coordinated power system, the upper / lower bounds of unit output are projected to 0 to 1, constructing a continuous variable of projected unit output, and establishing upper and lower limit constraints for unit output:

[0074]

[0075]

[0076]

[0077]

[0078]

[0079] Among them, P i T,up This represents the upward ramp rate of thermal power unit i; P represents the upward climbing rate of thermal power unit i after projection transformation; i T,down This represents the downward ramp rate of thermal power unit i; P represents the downward ramp rate of thermal power unit i after projection transformation; i T,start This represents the minimum output value of thermal power unit i when it is started up; P represents the minimum output value of thermal power unit i at startup after projection transformation; i T,shut This indicates the maximum output value of thermal power unit i when it is shut down; This represents the maximum output value of thermal power unit i after projection transformation when it is shut down; This represents the maximum output value of thermal power unit i; This represents the minimum output value of thermal power unit i;

[0080] f. Gradient constraints for thermal power units:

[0081]

[0082]

[0083] (5) Nuclear power plant operation constraints:

[0084] Divide the nuclear power safety peak shaving depth range into n equal parts k For the k-th level of nuclear power unit i, the peak-shaving depth is:

[0085]

[0086] Among them, P i N,max The minimum output allowed for nuclear power unit i;

[0087] The nuclear power output of nuclear power unit i during the low-power phase at the k-th peak shaving depth is:

[0088]

[0089] u i,k,j,t Let be the power increase / decrease operating state variable of nuclear power unit i in the j-th state of the k-th peak shaving depth during time period t. There are 3 power increase / decrease states under each peak shaving depth, and the corresponding nuclear power output is:

[0090]

[0091] Nuclear power output can be linearly expressed as:

[0092]

[0093] Among them, h i,t Let l be the rated power operating state variable of nuclear power unit i during time period t; i,k,t For nuclear power unit i, the low-power operation state variable during time period t at peak shaving depth k;

[0094] b. Constraints on runtime state variables:

[0095]

[0096] c. Coupling constraints of operating state variables when the power rise / fall time is 2 hours:

[0097] h i,t+1 ≥l i,k,t-1 +u i,k,2,t -1;

[0098] l i,k,t+1 ≥h i,t-1 +u i,k,2,t -1;

[0099] d. Coupling constraints of operating state variables when the power rise / fall time is 3 hours:

[0100] h i,t+1 ≥u i,k,1,t-1 +u i,k,3,t -1;

[0101] l i,k,t+1 ≥u i,k,3,t-1 +u i,k,1,t -1;

[0102] u i,k,1,t+1 ≥h i,t-1 +u i,k,3,t -1;

[0103] u i,k,3,t+1 ≥l i,k,t-1 +u i,k,1,t -1;

[0104] e. Constraints on rated power and low-power operation time of nuclear power units:

[0105]

[0106]

[0107] in, These represent the minimum continuous operating time at full power and the minimum continuous operating time at low power for nuclear power unit i, respectively; h i,x Let l be the rated power operating state variable of nuclear power unit i during time period x; i,k,t Let i be the low-power operation state variable of nuclear power unit i at the peak shaving depth of k during the x-period.

[0108] (6) Operational constraints of battery energy storage power stations:

[0109] a. Charge / discharge state constraints:

[0110] Battery energy storage actually has three states of charge and discharge: charging, discharging, and standby. In any given time period t, battery energy storage can only be in one state of charge and discharge, namely:

[0111] u c,t +u d,t ≤1

[0112] Among them, u c,t u d,t The variables representing the charge and discharge states of a battery energy storage power station are respectively: u during charging. c,t If u is 1, otherwise it is 0; when the battery discharges... d,t It is 1 if it is true, otherwise it is 0.

[0113] b. Charge / discharge power constraints:

[0114] 0≤P c,t ≤u c,t P c,max ;

[0115] 0≤P d,t ≤u d,t P d,max ;

[0116] Among them, P c,t P d,t P represents the charging power and discharging power of the battery energy storage power station during time period t, respectively. c,max P d,max These represent the maximum charging power and maximum discharging power of the battery energy storage power station, respectively.

[0117] c. Energy constraint:

[0118] Overcharging and over-discharging batteries will shorten their cycle life. Therefore, during any scheduling period, the energy of the battery energy storage station should be within its allowable energy range. To ensure the sustainability of the battery energy storage station's scheduling, the stored energy should be restored to its initial value at the end of the scheduling cycle.

[0119] E b,t+1 =E b,t +η c P c,t Δt-P d,t Δt / η d ;

[0120] E b,min ≤E b,t ≤E b,max ;

[0121] E b,T =E b,0 ;

[0122] Among them, E b,t η represents the remaining energy of the battery storage power station during time period t. c η d These represent the charging and discharging efficiency of the battery energy storage power station; Δt is the time length corresponding to the time period of the scheduling plan; E b,max E b,min These represent the upper and lower limits of the remaining energy of the battery energy storage power station; E b,T E represents the remaining energy of the battery storage power station at the end of the dispatch cycle. b,0 This indicates the initial remaining energy of the battery energy storage power station;

[0123] The specific rotational spare constraint is as follows:

[0124]

[0125]

[0126] in, Let t be the system's required value for positive rotational reserve during time period t; The system's negative spinning reserve requirement during time period t;

[0127] The power balance constraint is:

[0128]

[0129] Among them, P t D The system load power prediction value for time period t.

[0130] Preferably, the structure of the graph convolutional neural network is as follows:

[0131] Input layer:

[0132]

[0133] in It is a function containing learnable parameters; n is the number of nodes in the bipartite graph V, and m is the number of nodes in the bipartite graph C. These represent the input data for the i-th variable and the j-th constraint, respectively. Let i and j represent the features of the i-th variable and the j-th constraint in the first layer of the graph convolutional neural network, respectively.

[0134] Second layer:

[0135]

[0136]

[0137] in It is a function that contains learnable parameters; This represents the characteristic of the edge connecting the i-th variable and the j-th constraint;

[0138] Let i and j represent the features of the i-th variable and the j-th constraint in the second layer of the graph convolutional neural network, respectively.

[0139] Third layer:

[0140]

[0141]

[0142] in It is a function that contains learnable parameters; Let i and j represent the features of the i-th variable and the j-th constraint in the third layer of the graph convolutional neural network, respectively.

[0143] Output layer:

[0144]

[0145] sigmoid(x)=(1+e -x ) -1 Used to map the real number space to the range 0-1. Includes learnable parameters for... Reduced to one dimension, as the final output; This represents the feature of the i-th variable in the output layer of the graph convolutional neural network.

[0146] Preferably, the loss function used in step S2 when training the intraday planning prediction model and the rolling planning prediction model is the binary cross-entropy function; specifically as follows:

[0147]

[0148] Where N is the total number of units, T is the total number of time periods; log is a logarithmic function with base e; the optimal scheduling plan is y, y i,t This represents the on / off state of the i-th generator in the optimal scheduling plan during the t-th time period; the predicted scheduling plan is x, x i,t This represents the on / off state of the i-th generator in the predicted scheduling plan during the t-th time period.

[0149] Preferably, steps S3 / S4 further include:

[0150] Feasibility repair ensures the quality of the scheduling plan from the intraday planning model / dynamic planning model obtained through prediction. The specific feasibility repair is as follows:

[0151] Assumption It is the predicted scheduling plan, and u is the on / off status of each unit in the multi-source collaborative power system scheduling model at each time period;

[0152] In the multi-source cooperative power system dispatch model, a new variable α is introduced and constraints are added. The number of variables α is the same as the number of variables u. The specific constraints added are as follows:

[0153]

[0154]

[0155]

[0156] Where k is the size of the neighborhood; N is the total number of units; T is the total number of time periods; α i Indicate u i The changing state, if α i Set the value of k to 1, otherwise set it to 0; initialize the value of k, solve the scheduling plan, and if no feasible solution can be obtained, continuously increase the value of k until a feasible solution is obtained.

[0157] A multi-source cooperative power system rolling dispatching system based on a bipartite graph-graph convolutional neural network, applied to the method described, includes:

[0158] The module for establishing a multi-source collaborative power system dispatch model is used to establish a multi-source collaborative power system dispatch model, including an intraday planning model with a time period of 1 hour and a cycle of 1 day, and a rolling planning model with a time period of 15 minutes and a cycle of 3 hours; and converts the intraday planning model and the rolling planning model into mixed integer programming models respectively; and models the intraday planning model that needs to be solved into a bipartite graph.

[0159] The prediction model building module is used to build an intraday planning prediction model and a rolling planning prediction model. It collects historical intraday planning load data and historical rolling planning load data, generates datasets to train the intraday planning prediction model and the rolling planning prediction model, and obtains trained intraday planning prediction model and trained rolling planning prediction model, which are used to predict the scheduling plan of the intraday planning model and the scheduling plan of the rolling planning model, respectively; both the intraday planning prediction model and the rolling planning prediction model adopt graph convolutional neural networks.

[0160] The intraday planning forecasting module is used to model the intraday planning model to be solved into a bipartite graph. It predicts the scheduling plan of the intraday planning model through the trained intraday planning forecasting model, and obtains the power generation plan of each generator unit. The units predicted to be started are fixed to be started in the corresponding time period of the rolling planning forecasting model by adding the start-up time period constraint of the units. No operation is performed on the units predicted to be shut down.

[0161] The rolling plan prediction module is used to model the rolling plan model with added unit start-up time constraints into a bipartite graph. It predicts the scheduling plan of the rolling plan model through the trained rolling plan prediction model and fixes the corresponding time period of the predicted start-up units as constraints in the subsequent rolling plan model.

[0162] The output module is used to output the final scheduling plan.

[0163] A rolling dispatching device for multi-source cooperative power systems based on a bipartite graph-graph convolutional neural network, the device comprising:

[0164] Memory containing executable program code;

[0165] A processor coupled to the memory;

[0166] The processor calls the executable program code stored in the memory to execute the multi-source cooperative power system rolling scheduling method based on a bipartite graph-graph convolutional neural network.

[0167] A computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to execute the multi-source cooperative power system rolling dispatch method based on a bipartite graph-graph convolutional neural network.

[0168] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention transforms the multi-source coordinated power system scheduling model into a mixed-integer programming model, combining a bipartite graph with a graph convolutional neural network. First, an intraday planning prediction model and a rolling planning prediction model are established and trained. The trained intraday planning prediction model predicts the scheduling plan of the intraday planning model, obtaining the power generation plan of each generator unit. Units predicted to be operational are fixed to be operational in the corresponding time period of the rolling planning prediction model, achieved by adding constraints on the unit's operational time period. The rolling planning model with added unit operational time period constraints is modeled as a bipartite graph. The trained rolling planning prediction model predicts the scheduling plan of the rolling planning model, fixing the corresponding time period of the subsequently rolling planning model as a constraint, and then outputting the final scheduling plan. This invention uses machine learning methods to improve the speed of obtaining scheduling plans, thereby enabling power companies to shorten time intervals and formulate scheduling plans with higher matching degrees to actual loads, thus improving the multi-source coordinated power system's ability to absorb wind and solar power curtailment. Attached Figure Description

[0169] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0170] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0171] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0172] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0173] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0174] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0175] Example 1:

[0176] This embodiment provides a rolling dispatch method for multi-source cooperative power systems based on a bipartite graph-graph convolutional neural network. It combines a bipartite graph with a graph convolutional neural network. A bipartite graph is a special type of graph. Assume that graph G = (U, V) is an undirected graph where nodes U can be divided into two disjoint subsets (A, B). Each edge in the graph has two nodes originating from these two subsets, and nodes within the same subset are not connected by edges. The nodes and constraints of the MIP (Multipartite Power Layout Process) constitute these two subsets. Non-zero coefficients in the constraints form undirected edges between the constraints and variables. If the coefficient of variable v1 in constraint c1 is zero and the coefficient of variable v2 is 3, then there is an undirected edge between constraint c1 and variable v2. The method includes the following steps:

[0177] Step S1: Establish a multi-source collaborative power system dispatch model, including a daily planning model with a 1-hour time period and a 1-day cycle, and a rolling planning model with a 15-minute time period and a 3-hour cycle; and convert the daily planning model and the rolling planning model into mixed integer programming models respectively. Model the daily planning model that needs to be solved into a bipartite graph.

[0178] The constraints of the multi-source coordinated power system dispatch model include the operation constraints of wind, solar, hydro, thermal, nuclear, and energy storage power sources. The objective function is as follows:

[0179]

[0180] Where T is the total number of time periods in the scheduling cycle, T is 24 for the intraday planning model and 12 for the rolling planning model; These represent the costs of wind and solar power curtailment during time period t. Let i be the power generation cost and start-up cost of thermal power unit i in time period t; Let i be the cost of decommissioning nuclear power unit i during time period t;

[0181] The costs of curtailing wind, solar, and nuclear power are calculated as follows:

[0182]

[0183]

[0184]

[0185] Where, δ Wδ PV δ N These are the unit costs for wind curtailment, solar curtailment, and nuclear curtailment, respectively.

[0186] These represent the predicted available power for wind power and solar power in time period t, respectively.

[0187] P t W P t PV These represent the actual dispatched power of wind power and solar power during time period t, respectively.

[0188] P represents the actual dispatched power of nuclear power unit i during time period t; i N,max This is the maximum allowable output of nuclear power unit i.

[0189] Preferably, the power generation cost of a thermal power unit is calculated as follows:

[0190]

[0191] Among them, a i b i c i The correlation coefficient for the operating costs of thermal power units. This indicates the output of thermal power unit i during time period t;

[0192] right Perform a projection transformation to obtain

[0193]

[0194] Among them, P i T,max P i T,min The upper and lower limits of the output of thermal power unit i; u i,t Let i be the state variable of thermal power unit i during time period t, where "1" indicates operation and "0" indicates shutdown.

[0195] but Represented as:

[0196]

[0197]

[0198]

[0199]

[0200] Start-up cost of thermal power units

[0201] in, The hot start cost of thermal power unit i; The portion of the start-up cost of thermal power unit i that exceeds the hot start cost; v i,t This represents the startup status of thermal power unit i during time period t; 1 indicates unit startup and 0 indicates otherwise.

[0202] Daily planned power generation constraints for thermal power units:

[0203]

[0204] Daily planned power generation constraints for nuclear power units:

[0205]

[0206] Among them, T day This refers to the number of time slots planned for the day; These are the daily planned power generation limits for thermal power unit i and nuclear power unit i, respectively. These are the daily planned power limits for thermal power unit i and nuclear power unit i, respectively.

[0207] The operational constraints of wind, solar, hydro, thermal, nuclear, and energy storage power sources are as follows:

[0208] (1) Wind power operation constraints:

[0209]

[0210] (2) Photoelectric operation constraints:

[0211]

[0212] (3) Constraints on hydropower operation:

[0213] a. State constraints of pumped storage power stations:

[0214]

[0215] in, The variable is a binary variable representing the energy storage state of a pumped storage power station. It is 1 when the power station is in the energy storage state and 0 otherwise. The variable is a binary variable representing the power generation status of a pumped storage power station. It is 1 when the power station is generating electricity and 0 otherwise.

[0216] b. Power generation constraints and pumping power constraints of pumped storage power stations:

[0217]

[0218]

[0219] Among them, P t H,in P represents the electricity consumed by a pumped-storage power station during time period t for pumping water. Hin,max P Hin,min These represent the upper and lower limits of the electricity consumption of a pumped storage power station in a single time period; P t H,out P represents the output power of a pumped-storage hydroelectric power station during time period t. Hout,max P Hout,min These represent the upper and lower limits of the electricity output of a pumped storage power station during a single time period.

[0220] c. Reservoir capacity constraints of pumped storage power stations:

[0221] V min ≤V t ≤V max ;

[0222]

[0223] Among them, V t V represents the reservoir capacity of a pumped storage power station during time period t; max V min These represent the upper and lower limits of the reservoir capacity of a pumped storage power station; V t+1 η represents the reservoir capacity of the pumped storage power station at time t+1. in η out These are the pumping efficiency and power generation efficiency of the pumped storage power station's reservoir capacity, respectively.

[0224] (4) Constraints on thermal power plant operation:

[0225] a. Binary variable logical constraints for thermal power units:

[0226] v i,t -w i,t =u i,t -u i,t-1 ;

[0227] Among them, w i,t The shutdown status of thermal power unit i during time period t is 1 if the unit is shut down, and 0 otherwise.

[0228] b. Initial state constraints:

[0229] u i,t =u i,0 , t∈[1,…,U i +L i ];

[0230] U i =[min[T,u i,0 ( Ton,i -T i,0 )]] + ;

[0231] L i =[min[T,(1-u i,0 ()( T off,i +T i,0 )]] + ;

[0232] Among them, u i,0 This represents the state of thermal power unit i during the initial period, i.e., the period before the start of the dispatch cycle, where "1" indicates operation and "0" indicates shutdown; U i L represents the time that thermal power unit i still needs to run at the initial moment. i This indicates the time that thermal power unit i still needs to be shut down at the initial moment; T is the total number of scheduling periods, [·] + This represents max(0,·); T on,i This represents the minimum start-up time of thermal power unit i; T off,i T represents the minimum shutdown time of thermal power unit i; i,0 This represents the number of time periods during which thermal power unit i has been continuously started or shut down before the start of the dispatching cycle; T represents the number of time periods during which thermal power unit i has been continuously started before the start of the dispatching cycle. i,0 When T is positive, it remains positive during continuous shutdown. i,0 Negative;

[0233] c. Minimum start-up and shutdown constraints:

[0234]

[0235]

[0236] d. Startup cost constraints:

[0237]

[0238] in, The cold start cost of thermal power unit i; The hot start cost of thermal power unit i; T cold,i f is the cold start time of thermal power unit i; init,i,t It is a constant, if τ- T off,i - T cold,i ≤0 and [-T i,0 ] + <|τ- T off,i -T cold,i -1|+1, then f init,i,t =1, indicating that thermal power unit i does not belong to cold start at time t; otherwise, f init,i,t =0 indicates that thermal power unit i is in a cold start at time t;

[0239] e. Upper and lower limits of thermal power unit output:

[0240] Based on the fundamental data of the multi-source coordinated power system, the upper / lower bounds of unit output are projected to 0 to 1, constructing a continuous variable of projected unit output, and establishing upper and lower limit constraints for unit output:

[0241]

[0242]

[0243]

[0244]

[0245]

[0246] Among them, P i T,up This represents the upward ramp rate of thermal power unit i; P represents the upward climbing rate of thermal power unit i after projection transformation; i T,down This represents the downward ramp rate of thermal power unit i; P represents the downward ramp rate of thermal power unit i after projection transformation; i T,start This represents the minimum output value of thermal power unit i when it is started up; P represents the minimum output value of thermal power unit i at startup after projection transformation; i T,shut This indicates the maximum output value of thermal power unit i when it is shut down; This represents the maximum output value of thermal power unit i after projection transformation when it is shut down; This represents the maximum output value of thermal power unit i; This represents the minimum output value of thermal power unit i;

[0247] f. Gradient constraints for thermal power units:

[0248]

[0249]

[0250] (5) Nuclear power plant operation constraints:

[0251] Divide the nuclear power safety peak shaving depth range into n equal parts k For the k-th level of nuclear power unit i, the peak-shaving depth is:

[0252]

[0253] Among them, P i N,max The minimum output allowed for nuclear power unit i;

[0254] The nuclear power output of nuclear power unit i during the low-power phase at the k-th peak shaving depth is:

[0255]

[0256] u i,k,j,t Let be the power increase / decrease operating state variable of nuclear power unit i in the j-th state of the k-th peak shaving depth during time period t. There are 3 power increase / decrease states under each peak shaving depth, and the corresponding nuclear power output is:

[0257]

[0258] Nuclear power output can be linearly expressed as:

[0259]

[0260] Among them, h i,t Let l be the rated power operating state variable of nuclear power unit i during time period t; i,k,t For nuclear power unit i, the low-power operation state variable during time period t at peak shaving depth k;

[0261] b. Constraints on runtime state variables:

[0262]

[0263] c. Coupling constraints of operating state variables when the power rise / fall time is 2 hours:

[0264] h i,t+1 ≥l i,k,t-1 +u i,k,2,t -1;

[0265] l i,k,t+1 ≥h i,t-1 +u i,k,2,t -1;

[0266] d. Coupling constraints of operating state variables when the power rise / fall time is 3 hours:

[0267] h i,t+1 ≥u i,k,1,t-1 +u i,k,3,t -1;

[0268] li,k,t+1 ≥u i,k,3,t-1 +u i,k,1,t -1;

[0269] u i,k,1,t+1 ≥h i,t-1 +u i,k,3,t -1;

[0270] u i,k,3,t+1 ≥l i,k,t-1 +u i,k,1,t -1;

[0271] e. Constraints on rated power and low-power operation time of nuclear power units:

[0272]

[0273]

[0274] in, These represent the minimum continuous operating time at full power and the minimum continuous operating time at low power for nuclear power unit i, respectively; h i,x Let l be the rated power operating state variable of nuclear power unit i during time period x; i,k,t Let i be the low-power operation state variable of nuclear power unit i at the peak shaving depth of k during the x-period.

[0275] (6) Operational constraints of battery energy storage power stations:

[0276] a. Charge / discharge state constraints:

[0277] Battery energy storage actually has three states of charge and discharge: charging, discharging, and standby. In any given time period t, battery energy storage can only be in one state of charge and discharge, namely:

[0278] u c,t +u d,t ≤1

[0279] Among them, u c,t u d,t The variables representing the charge and discharge states of a battery energy storage power station are respectively: u during charging. c,t If u is 1, otherwise it is 0; when the battery discharges... d,t It is 1 if it is true, otherwise it is 0.

[0280] b. Charge / discharge power constraints:

[0281] 0≤P c,t ≤u c,t P c,max ;

[0282] 0≤P d,t ≤u d,t P d,max ;

[0283] Among them, P c,t P d,t P represents the charging power and discharging power of the battery energy storage power station during time period t, respectively. c,max P d,max These represent the maximum charging power and maximum discharging power of the battery energy storage power station, respectively.

[0284] c. Energy constraint:

[0285] Overcharging and over-discharging batteries will shorten their cycle life. Therefore, during any scheduling period, the energy of the battery energy storage station should be within its allowable energy range. To ensure the sustainability of the battery energy storage station's scheduling, the stored energy should be restored to its initial value at the end of the scheduling cycle.

[0286] E b,t+1 =E b,t +η c P c,t Δt-P d,t Δt / η d ;

[0287] E b,min ≤E b,t ≤E b,max ;

[0288] E b,T =E b,0 ;

[0289] Among them, E b,t η represents the remaining energy of the battery storage power station during time period t. c η d These represent the charging and discharging efficiency of the battery energy storage power station; Δt is the time length corresponding to the time period of the scheduling plan; E b,max E b,min These represent the upper and lower limits of the remaining energy of the battery energy storage power station; E b,T E represents the remaining energy of the battery storage power station at the end of the dispatch cycle. b,0 This indicates the initial remaining energy of the battery energy storage power station;

[0290] The specific rotational spare constraint is as follows:

[0291]

[0292]

[0293] in, Let t be the system's required value for positive rotational reserve during time period t; The system's negative spinning reserve requirement during time period t;

[0294] The power balance constraint is:

[0295]

[0296] Among them, P t D The system load power prediction value for time period t.

[0297] The scheduling problem of a multi-source coordinated power system can be represented as a mixed-integer programming problem. Let V be the set of variables in this mixed-integer programming problem, c be the coefficients in the objective function, l be the lower bound of the variables, and u be the upper bound of the variables. The nodes V in set V... i Using feature vectors Let 'i' be the index, representing the i-th element in the set. The first character is the variable v. i The coefficients in the objective function; the second is the variable v. i The lower bound; the third is the variable v. i The upper bound; the fourth bit represents the variable v. i The type is 0 for continuous variables and 1 for binary variables; the fifth character represents the variable v. i The average coefficient among all the constraints in which it appears; the sixth is the variable v. i The seventh most frequent occurrence of this variable is v in all constraints. i The eighth largest coefficient among all the constraints in which it appears is the variable v. i The smallest coefficient that exists among all the constraints in which it appears.

[0298] Given that C is the set of constraints for this problem, and b is the right-hand side of the constraint. Node C in set C... i Using feature vectors Let be the index, i, representing the i-th element in the set. The first character is the constraint C. i The right-hand term; if constraint C i The second digit of the equality constraint is 1, otherwise 0; the third digit is constraint C. i The average of all coefficients in the middle; the fourth digit is constraint C. i The number of variables present.

[0299] Edge E in set E i,j =A i,j , indicates V j Variables in C i The coefficients in the constraints are the values ​​in the i-th row and j-th column of the coefficient matrix. If constraint C... i There is no variable V in it j Then E i,j =0.

[0300] Concatenate the above feature vectors together in order, so that H E =A. H v It is a large matrix with n rows. Each row represents all the features of a variable. The i-th row uses... Let's represent this: there are n variables in total, so the subscripts range from 1 to n. For H... c Similarly, each row represents all the features of a constraint, and the i-th row uses... Let H represent the number of constraints as m, hence the subscripts range from 1 to m. E This represents all the features of an edge, and is an m*n matrix where m is the number of constraints and n is the number of variables. Here, an edge has only one feature: the coefficient of the variable within the constraints. The i-th row and j-th column represents V. j Variables in C i The coefficients in the constraints. The upper right letter 'v' represents a variable, 'c' represents a constraint, and 'e' represents an edge.

[0301] So H v It contains information about all variables, H c H contains information about all constraints E It contains information about the constraints and variables. These three parts represent a complete MIP problem losslessly, and thus a UC problem losslessly, whether it is an intraday problem or a rolling solution problem.

[0302] A bipartite graph G = (V∪C, E). V∪C forms the nodes of the graph and can be divided into two sets, V and C, which are disjoint. Let V represent the set of variable nodes and the set of constraint nodes, respectively. E is the set of edges in the graph. Each edge in E connects exactly one V node and one C node, and the weight of this edge is exactly the coefficient of the V variable in constraint C. There are no edges in E connecting V nodes to V nodes, nor are there any edges connecting C nodes to C nodes. These three parts represent a lossless multi-source cooperative power system problem, whether it is an intraday problem or a rolling solution problem.

[0303] Step S2: Establish an intraday planning forecasting model and a rolling planning forecasting model. Collect historical intraday planning load data and historical rolling planning load data, and generate datasets to train the intraday planning forecasting model and the rolling planning forecasting model, obtaining trained intraday planning forecasting model and trained rolling planning forecasting model, which are used to predict the scheduling plan of the intraday planning model and the scheduling plan of the rolling planning model, respectively. Both the intraday planning forecasting model and the rolling planning forecasting model adopt graph convolutional neural networks.

[0304] Specifically, historical load data is collected. The first type is historical data from an intraday planning model with a 1-hour timeframe and a 1-day cycle; the second type is historical data from a rolling planning model with a 15-minute timeframe and a 3-hour cycle. Two graph neural networks are used, with the same structure but different parameters. The intraday planning prediction model is trained using historical data from the intraday planning model to predict the scheduling plan for that model; the rolling planning prediction model is trained using historical data from the rolling planning model to predict the scheduling plan for that model.

[0305] The specific structure of a graph-based convolutional neural network is as follows:

[0306] Input layer:

[0307]

[0308] in It is a function containing learnable parameters. The subscript 1 indicates that this is the function of the first layer, which distinguishes it from the second and third layers. It is a multilayer perceptron (linear layers plus activation functions); n is the number of V nodes in the bipartite graph, and m is the number of C nodes in the bipartite graph. These represent the input data for the i-th variable and the j-th constraint, respectively. Let i and j represent the features of the i-th variable and the j-th constraint in the first layer of the graph convolutional neural network, respectively.

[0309] Second layer:

[0310]

[0311]

[0312] in It is a function that contains learnable parameters, where v indicates a function used on variables and c indicates a function used on constraints; This represents the characteristic of connecting the i-th variable and the j-th constraint. Let i and j represent the features of the i-th variable and the j-th constraint in the second layer of the graph convolutional neural network, respectively.

[0313] Third layer:

[0314]

[0315]

[0316] in It is a function that contains learnable parameters. The subscript 3 indicates that this function is the third-level function that contains learnable parameters.

[0317] Let i and j represent the features of the i-th variable and the j-th constraint in the third layer of the graph convolutional neural network, respectively.

[0318] Output layer:

[0319]

[0320] sigmoid(x)=(1+e -x ) -1 Used to map the real number space to the range 0-1. Includes learnable parameters for... Reduced to one dimension, as the final output; This represents the feature of the i-th variable in the output layer of the graph convolutional neural network.

[0321] The loss function used when training the intraday and rolling planning forecasting models is the binary cross-entropy function. Bipartite data and corresponding scheduling plans are generated based on real historical loads; the bipartite data serves as the input data, and the scheduling plans serve as the labels. The specific binary cross-entropy function is as follows:

[0322]

[0323] Where N is the total number of units, T is the total number of time periods; log is a logarithmic function with base e; the optimal scheduling plan is y, y i,t This represents the on / off state of the i-th generator in the optimal scheduling plan during the t-th time period; the predicted scheduling plan is x, x i,t This represents the on / off state of the i-th generator in the predicted scheduling plan during the t-th time period.

[0324] Step S3: The intraday scheduling plan is predicted using the trained intraday planning prediction model to obtain the power generation plan for each generator unit. Feasibility repair ensures the feasibility of the predicted solution. Units predicted to be operational are fixed to be operational during the corresponding time period in the rolling planning prediction model by adding constraints on the unit's operational time period. Units predicted to be shut down are not subject to any operation. The added constraints are as follows:

[0325] u i,t =1,u i,t It is the i-th unit that is determined to be turned on during time period t.

[0326] Step S4: The rolling plan model with added unit start-up time constraints is modeled into a bipartite graph. The scheduling plan of the rolling plan model is predicted by the trained rolling plan prediction model to obtain a more granular scheduling plan. The feasibility of the predicted solution is ensured by feasibility repair. The corresponding time period of the rolling plan model for the units predicted to start up is fixed as a constraint in the subsequent rolling plan model.

[0327] Step S5: Output the final scheduling plan.

[0328] Steps S3 / S4 also include:

[0329] Feasibility repair is used to ensure the quality of the scheduling plans from the intraday planning model and the dynamic planning model. Since the scheduling plans predicted by neural networks cannot guarantee feasibility, a feasibility repair method is introduced. The specific details of how feasibility repair ensures the feasibility of the predicted solution are as follows:

[0330] Assumption It is the predicted scheduling plan, and u is the on / off status of each unit in the multi-source collaborative power system scheduling model at each time period;

[0331] In the multi-source cooperative power system dispatch model, a new variable α is introduced and constraints are added. The number of variables α is the same as the number of variables u. The specific constraints added are as follows:

[0332]

[0333]

[0334]

[0335] Where k is the size of the neighborhood; N is the total number of units; T is the total number of time periods; α i Indicate u i The changing state, if α i Set the value of k to 1, otherwise set it to 0; initialize the value of k, you can set a small k, solve the scheduling plan, and if no feasible solution can be obtained, keep increasing the value of k until a feasible solution is obtained.

[0336] This invention greatly improves problem-solving efficiency with almost no loss of optimality, enabling shorter time intervals or larger-scale systems.

[0337] Example 2:

[0338] This embodiment, based on the same inventive concept as Embodiment 1, provides a multi-source cooperative power system rolling dispatching system based on a bipartite graph-graph convolutional neural network, applied to the method described, including:

[0339] The module for establishing a multi-source collaborative power system dispatch model is used to establish a multi-source collaborative power system dispatch model, including an intraday planning model with a time period of 1 hour and a cycle of 1 day, and a rolling planning model with a time period of 15 minutes and a cycle of 3 hours; and converts the intraday planning model and the rolling planning model into mixed integer programming models respectively; and models the intraday planning model that needs to be solved into a bipartite graph.

[0340] The prediction model building module is used to build an intraday planning prediction model and a rolling planning prediction model. It collects historical intraday planning load data and historical rolling planning load data, generates datasets to train the intraday planning prediction model and the rolling planning prediction model, and obtains trained intraday planning prediction model and trained rolling planning prediction model, which are used to predict the scheduling plan of the intraday planning model and the scheduling plan of the rolling planning model, respectively; both the intraday planning prediction model and the rolling planning prediction model adopt graph convolutional neural networks.

[0341] The intraday planning forecasting module is used to model the intraday planning model to be solved into a bipartite graph. It predicts the scheduling plan of the intraday planning model through the trained intraday planning forecasting model, and obtains the power generation plan of each generator unit. The units predicted to be started are fixed to be started in the corresponding time period of the rolling planning forecasting model by adding the start-up time period constraint of the units. No operation is performed on the units predicted to be shut down.

[0342] The rolling plan prediction module is used to model the rolling plan model with added unit start-up time constraints into a bipartite graph. It predicts the scheduling plan of the rolling plan model through the trained rolling plan prediction model and fixes the corresponding time period of the predicted start-up units as constraints in the subsequent rolling plan model.

[0343] The output module is used to output the final scheduling plan.

[0344] Example 3:

[0345] This embodiment, based on the same inventive concept as Embodiment 1, provides a multi-source cooperative power system rolling dispatching device based on a bipartite graph-graph convolutional neural network, the device comprising:

[0346] Memory containing executable program code;

[0347] A processor coupled to the memory;

[0348] The processor calls the executable program code stored in the memory to execute the multi-source cooperative power system rolling scheduling method based on a bipartite graph-graph convolutional neural network.

[0349] Example 4:

[0350] This embodiment, based on the same inventive concept as Embodiment 1, provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to execute the multi-source cooperative power system rolling dispatch method based on a bipartite graph-graph convolutional neural network.

[0351] Those skilled in the art will recognize that the modules of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0352] In the embodiments provided by this invention, it should be understood that the division of modules is only a logical functional division. In actual implementation, there may be other division methods, such as multiple modules can be combined into one module, one module can be split into multiple modules, or some features can be ignored.

[0353] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0354] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0355] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A multi-source collaborative power system rolling dispatch method based on a bipartite graph-graph convolutional neural network, characterized in that, Includes the following steps: Step S1: Establish a multi-source collaborative power system dispatch model, including an intraday planning model with a time period of 1 hour and a cycle of 1 day, and a rolling planning model with a time period of 15 minutes and a cycle of 3 hours; and convert the intraday planning model and the rolling planning model into mixed integer programming models respectively; and model the intraday planning model that needs to be solved into a bipartite graph. Step S2: Establish an intraday planning forecasting model and a rolling planning forecasting model. Collect historical intraday planning load data and historical rolling planning load data, and generate datasets to train the intraday planning forecasting model and the rolling planning forecasting model, obtaining trained intraday planning forecasting model and trained rolling planning forecasting model, which are used to predict the scheduling plan of the intraday planning model and the scheduling plan of the rolling planning model, respectively; both the intraday planning forecasting model and the rolling planning forecasting model adopt graph convolutional neural networks; Step S3: Predict the scheduling plan of the intraday planning model through the trained intraday planning prediction model to obtain the power generation plan of each generator unit. Fix the generator units predicted to be started in the corresponding time period of the rolling planning prediction model to be started, which is achieved by adding the start-up time period constraint of the generator units. No operation is performed on the generator units predicted to be shut down. Step S4: Model the rolling plan model with added unit start-up time constraints into a bipartite graph, predict the scheduling plan of the rolling plan model through the trained rolling plan prediction model, and fix the corresponding time period of the predicted start-up units in the subsequent rolling plan model as constraints. Step S5: Output the final scheduling plan. 2.The multi-source coordinated power system rolling dispatching method based on bipartite graph-graph convolutional neural network according to claim 1, wherein, The constraints of the multi-source coordinated power system dispatch model include the operation constraints of wind, solar, hydro, thermal, nuclear, and energy storage power sources, and the objective function is as follows: ； wherein, T is the total number of time periods in a dispatch cycle, T is 24 for the intra-day scheduling model and T is 12 for the rolling scheduling model; , respectively the wind curtailment and solar curtailment cost for time period t; , , the generation cost and start-up cost of the thermal power unit for time period t; , , the nuclear curtailment cost for time period t; and , , The costs of curtailing wind, solar, and nuclear power are calculated as follows: ； ； ； in, , , These are the unit costs for wind curtailment, solar curtailment, and nuclear curtailment, respectively. , Wind power and solar power, respectively, during the time period The predicted available power; , Wind power and solar power, respectively, during the time period The actual dispatched power volume; For nuclear power units During the period The actual dispatched power volume; For nuclear power units The maximum allowable output.

3. The rolling dispatch method for multi-source cooperative power systems based on a bipartite graph-graph convolutional neural network according to claim 2, characterized in that, The power generation cost of thermal power units is calculated as follows: ； in, , , The correlation coefficient for the operating costs of thermal power units. Indicates thermal power unit During the period The magnitude of the output force; right Perform a projection transformation to obtain : ； in, , For thermal power units Upper and lower limits of output; For thermal power units During the period The state variable, "1" indicates running, and "0" indicates stopped; but Represented as: ； ； ； Start-up cost of thermal power units ; in, For thermal power units Hot start cost; For thermal power units The portion of startup costs that exceeds the cost of a warm start; For thermal power units During the period The startup status is 1 if the unit is running, and 0 otherwise. Daily planned power generation constraints for thermal power units: ； Daily planned power generation constraints for nuclear power units: ； in, This refers to the number of time slots planned for the day; , thermal power units nuclear power units The daily planned power consumption limit; , thermal power units nuclear power units The lower limit of the planned daily electricity consumption.

4. The rolling dispatch method for multi-source cooperative power systems based on a bipartite graph-graph convolutional neural network according to claim 3, characterized in that, The operational constraints for the wind, solar, hydro, thermal, nuclear, and energy storage power sources are as follows: (1) Wind power operation constraints: ； (2) Photoelectric operation constraints: ； (3) Operational constraints of hydropower: a. State constraints of pumped storage power stations: ； in, The variable is a binary variable representing the energy storage state of a pumped storage power station. It is 1 when the power station is in the energy storage state and 0 otherwise. The variable is a binary variable representing the power generation status of a pumped storage power station. It is 1 when the power station is generating electricity and 0 otherwise. b. Power generation constraints and pumping power constraints of pumped storage power stations: ； ； in, For pumped storage power stations during the time period The electricity used for pumping water; , These are the upper and lower limits of the electricity consumption of a pumped storage power station in a single time period; For pumped storage power stations during the time period Output power; , These represent the upper and lower limits of the electricity output of a pumped storage power station during a single time period. c. Reservoir capacity constraints of pumped storage power stations: ； ； in, For pumped storage power stations during the time period Storage capacity; , These are the upper and lower limits of the reservoir capacity of pumped storage power stations; For pumped storage power stations during the time period Storage capacity; , These are the pumping efficiency and power generation efficiency of the pumped storage power station's reservoir capacity, respectively. (4) Constraints on thermal power plant operation: a. Binary variable logical constraints for thermal power units: ； in, thermal power units During the period The unit is in the shutdown state; 1 indicates the unit is off, and 0 indicates otherwise. b. Initial state constraints: ， ； ； ； in, For thermal power units In the initial period, which is the period before the start of the scheduling cycle, "1" indicates running and "0" indicates stopping. Indicates thermal power unit The time still needed to run at the initial moment Indicates thermal power unit There is still a downtime required at the initial stage; The total number of scheduling periods. express ; Indicates thermal power unit Minimum boot time; Indicates thermal power unit Minimum shutdown time; Indicates thermal power unit The number of periods during which the unit has been continuously started or shut down before the start of the scheduling cycle; thermal power units If continuous operation has been maintained before the start of the scheduling cycle When the value is positive, the device remains continuously powered off. Negative; c. Minimum start-up and shutdown constraints: ； ； ； ； d. Startup cost constraints: ； in, For thermal power units The cold start cost; For thermal power units Hot start cost; For thermal power units Cold start time; It is a constant, if and ,but This indicates thermal power units exist The start of the time period is not considered a cold start, otherwise This indicates thermal power units exist The start of the period is considered a cold start; e. Upper and lower limits of thermal power unit output: Based on the fundamental data of the multi-source coordinated power system, the upper / lower bounds of unit output are projected to 0~1, constructing a continuous variable of projected unit output, and establishing upper and lower limit constraints for unit output: ； ； ； ； ； in, Indicates thermal power unit The rate of upward climb; Indicates thermal power unit The upward climbing rate after projection transformation; Indicates thermal power unit The rate of downward climb; Indicates thermal power unit The downward climbing rate after projection transformation; Indicates thermal power unit Minimum output value at startup; Indicates thermal power unit The minimum output value at startup after projection transformation; Indicates thermal power unit Maximum output value when the machine is off; Indicates thermal power unit The maximum output value when the machine is off after projection transformation; Indicates thermal power unit Maximum output value; Indicates thermal power unit The minimum output value; f. Gradient constraints for thermal power units: ； ； (5) Nuclear power plant operation constraints: The depth range of nuclear power safety peak shaving is divided into... The nuclear power unit No. Peak shaving depth is: ； in, , nuclear power units The maximum and minimum allowable output; nuclear power units No. The nuclear power output during the low-power phase at the peak-shaving depth is: ； For nuclear power units exist Time period Peak shaving depth There are three power increase / decrease states for each peak shaving depth, and the corresponding nuclear power output is as follows: ； The linear expression for nuclear power output is: ； in, For nuclear power units exist Operating status variables at rated power during specific time periods; For nuclear power units In the Peak shaving depth at Low-power operation state variables during specific time periods; b. Constraints on runtime state variables: ； c. Coupling constraints of operating state variables when the power rise / fall time is 2 hours: ； ； d. Coupling constraints of operating state variables when the power rise / fall time is 3 hours: ； ； ； ； e. Constraints on rated power and low-power operation time of nuclear power units: ， ； ， ； in, , nuclear power units Minimum continuous operating time at full power and minimum continuous operating time at low power; For nuclear power units exist Operating status variables at rated power during specific time periods; For nuclear power units In the Peak shaving depth at Low-power operation state variables during specific time periods; (6) Operational constraints of battery energy storage power stations: a. Charge / discharge state constraints: Battery energy storage actually has three states of charge and discharge: charging state, discharging state, and standby state; during any scheduling period... Battery energy storage can only be in one state of charge and discharge, that is: in, , The two variables represent the charge and discharge states of a battery energy storage power station, respectively; during the charging process of the power station... The value is 1 if the battery is discharging, and 0 otherwise. It is 1 if it is true, otherwise it is 0. b. Charge / discharge power constraints: ； ； in, , These represent the time periods of the battery energy storage power station. The charging power and discharging power; , These represent the maximum charging power and maximum discharging power of the battery energy storage power station, respectively. c. Energy constraint: Overcharging and over-discharging batteries will shorten their cycle life. Therefore, during any scheduling period, the energy of the battery energy storage station should be within its allowable energy range. To ensure the sustainability of the battery energy storage station's scheduling, the stored energy should be restored to its initial value at the end of the scheduling cycle. ； ； ； in, For battery energy storage power stations during time periods The remaining energy; , These represent the charging and discharging efficiencies of the battery energy storage power station; The time length corresponding to the time period in the scheduling plan; , These represent the upper and lower limits of the remaining energy of the battery energy storage power station, respectively. This indicates the remaining energy of the battery storage power station at the end of the scheduling cycle; This indicates the initial remaining energy of the battery energy storage power station; The specific rotational spare constraint is as follows: ； ； in, For the system in the first The required value of rotating reserve in the correct time period; For the system in the first The demand for negative spinning reserve during the time period; The power balance constraint is: ； in, For time period The predicted system load power consumption.

5. The rolling dispatch method for multi-source cooperative power systems based on a bipartite graph-graph convolutional neural network according to claim 1, characterized in that, The specific structure of the graph convolutional neural network is as follows: Input layer: ； in It is a function containing learnable parameters; n is the number of nodes in the bipartite graph V, and m is the number of nodes in the bipartite graph C. , These represent the input data for the i-th variable and the j-th constraint, respectively. Let i and j represent the features of the i-th variable and the j-th constraint in the first layer of the graph convolutional neural network, respectively. Second layer: ； ； in It is a function that contains learnable parameters; This represents the characteristic of the edge connecting the i-th variable and the j-th constraint; Let i and j represent the features of the i-th variable and the j-th constraint in the second layer of the graph convolutional neural network, respectively. Third layer: ； ； in It is a function that contains learnable parameters; Let i and j represent the features of the i-th variable and the j-th constraint in the third layer of the graph convolutional neural network, respectively. Output layer: ； Used to map the real number space to the range 0-1. Includes learnable parameters for... Reduced to one dimension, as the final output; This represents the feature of the i-th variable in the output layer of the graph convolutional neural network.

6. The rolling dispatch method for multi-source cooperative power systems based on a bipartite graph-graph convolutional neural network according to claim 1, characterized in that, In step S2, the loss function used when training the intraday planning prediction model and the rolling planning prediction model is the binary cross-entropy function; specifically as follows: Where N is the total number of generating units and T is the total number of time periods; It is natural number The logarithmic function with base π; the optimal scheduling plan is , In the optimal scheduling plan, the first... The generator in the first The on / off status of each time period; the predicted scheduling plan is , This indicates the first element in the predicted scheduling plan. The generator in the first The power on / off status during each time period.

7. A rolling dispatching method for multi-source cooperative power systems based on a bipartite graph-graph convolutional neural network according to claim 1, characterized in that, Steps S3 / S4 also include: Feasibility repair is used to ensure the quality of the scheduling plans from the intraday planning model / rolling planning model obtained from the forecast. The specific feasibility repair is as follows: Assumption It is the predicted scheduling plan, and u is the on / off status of each unit in the multi-source collaborative power system scheduling model at each time period; Introducing new variables into the multi-source cooperative power system dispatch model And add constraints and variables. The number is the same as the number of u; the added constraints are as follows: ； ； ； Where k is the size of the neighborhood; N is the total number of units; and T is the total number of time periods. express The changing state, if Set the value of k to 1, otherwise set it to 0; initialize the value of k, solve the scheduling plan, and if no feasible solution can be obtained, continuously increase the value of k until a feasible solution is obtained.

8. A multi-source cooperative power system rolling dispatching system based on a bipartite graph-graph convolutional neural network, characterized in that, The method applied to any one of claims 1 to 7 includes: The module for establishing a multi-source collaborative power system dispatch model is used to establish a multi-source collaborative power system dispatch model, including an intraday planning model with a time period of 1 hour and a cycle of 1 day, and a rolling planning model with a time period of 15 minutes and a cycle of 3 hours; and converts the intraday planning model and the rolling planning model into mixed integer programming models respectively; and models the intraday planning model that needs to be solved into a bipartite graph. The prediction model building module is used to build an intraday planning prediction model and a rolling planning prediction model. It collects historical intraday planning load data and historical rolling planning load data, generates datasets to train the intraday planning prediction model and the rolling planning prediction model, and obtains trained intraday planning prediction model and trained rolling planning prediction model, which are used to predict the scheduling plan of the intraday planning model and the scheduling plan of the rolling planning model, respectively; both the intraday planning prediction model and the rolling planning prediction model adopt graph convolutional neural networks. The intraday planning forecasting module is used to model the intraday planning model to be solved into a bipartite graph. It predicts the scheduling plan of the intraday planning model through the trained intraday planning forecasting model, and obtains the power generation plan of each generator unit. The units predicted to be started are fixed to be started in the corresponding time period of the rolling planning forecasting model by adding the start-up time period constraint of the units. No operation is performed on the units predicted to be shut down. The rolling plan prediction module is used to model the rolling plan model with added unit start-up time constraints into a bipartite graph. It predicts the scheduling plan of the rolling plan model through the trained rolling plan prediction model and fixes the corresponding time period of the predicted start-up units as constraints in the subsequent rolling plan model. The output module is used to output the final scheduling plan.

9. A rolling dispatching device for multi-source cooperative power systems based on a bipartite graph-graph convolutional neural network, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the multi-source cooperative power system rolling dispatch method based on a bipartite graph-graph convolutional neural network as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the multi-source cooperative power system rolling dispatch method based on a bipartite graph-graph convolutional neural network as described in any one of claims 1 to 7.

Images