Unit commitment solving method and system based on space-time deep learning and perceptual neighborhood search

Abstract

The invention discloses a unit commitment solving method and system based on spatio-temporal deep learning and perceptual neighborhood search, belongs to the technical field of power system dispatching, combines the advantages of a machine learning method and mathematical optimization, and excavates spatio-temporal correlation among decision variables from historical data to obtain a unit commitment solution. The problem that space-time correlation among variables is rarely considered in an existing unit commitment solving method is solved, and prediction precision can be effectively improved. And meanwhile, by utilizing a perceptual neighborhood search method, the situation that an infeasible or suboptimal solution possibly occurs due to the fact that part of decision variables are directly fixed after the initial solution is predicted and the fixation is not accurate enough is avoided, the solution speed is increased, the quality of understanding is ensured, and the dispatching accuracy of the power system is ensured.

Description

Technical Field

[0001] This invention belongs to the field of power system dispatching technology, and more specifically, relates to a method and system for solving unit combination based on spatiotemporal deep learning and perceptual neighborhood search. Background Technology

[0002] Security-constrained unit commitment (SCUC) is an optimization problem in power system operation. This problem pre-plans the start-up and shutdown states and power generation of generating units for each time period within a certain scheduling cycle (e.g., 24 hours) to meet system demand, aiming to minimize the total operating cost of the power system. Therefore, in the spot electricity market, solving the unit commitment problem within the clearing time window is crucial for obtaining market dispatch decisions. This problem is generally described as a mixed-integer linear programming (MILP) problem, where 0-1 variables include the start-up and shutdown states of generating units and their corresponding start-up actions, and continuous variables include the power generation and demand response power of the units. The problem includes generating unit constraints, demand response constraints, system supply and demand power balance, and grid security constraints. Currently, commercial optimization solvers based on the branch-and-cut method (such as Gurobi or COPT) are mainly used in practice. The development of electricity market rules, the addition of more new market participants, and the refined modeling of power systems make the efficient solution of power optimization problems even more challenging. Furthermore, although the SCUC problem is run repeatedly daily or hourly, commercial solvers treat each run of the unit combination problem as an independent MILP problem and solve it separately, resulting in a failure to accumulate useful experience. Existing research generally uses artificial intelligence methods to predict the 0-1 variables of the unit and obtain initial solutions by fixing the variables based on predicted probabilities or pre-defined rules, thereby transforming the SCUC problem into a smaller-scale MILP problem for solution.

[0003] In summary, the solution methods for the SCUC problem in power system dispatch have the following problems or areas for improvement: Existing solution prediction methods often predict the start-up and shutdown variables of each unit separately, ignoring the spatiotemporal correlation of the start-up and shutdown variables of each unit in the unit combination problem; at the same time, the method of directly fixing some decision variables after predicting the initial solution may lead to infeasible or suboptimal solutions due to insufficient accuracy of fixing, resulting in insufficient accuracy in solving the safety-constrained unit combination problem. Summary of the Invention

[0004] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a method and system for solving unit combination problems based on spatiotemporal deep learning and perceptual neighborhood search. This solves the problem that existing unit combination problem-solving methods rarely consider the spatiotemporal correlation between variables, and avoids the possibility that directly fixing some decision variables after predicting the initial solution may lead to infeasible or suboptimal solutions due to insufficient accuracy.

[0005] To achieve the above objectives, according to a first aspect of the present invention, a method for solving unit combination based on spatiotemporal deep learning and perceptual neighborhood search is provided, comprising: Obtain the power grid topology, node characteristic data of each node in the power grid topology, and edge characteristic data of the transmission lines between nodes in the safety-constrained unit combination problem to be solved; Based on the power grid topology, the node characteristic data of each node, and the edge characteristic data of the transmission line, graph data is constructed; the node characteristic data includes the historical net load data and static parameters of the corresponding node. Using a spatial correlation extraction model, the spatial correlation between nodes in the graph data is extracted to obtain the spatial correlation features of each node. Using a time correlation extraction model, the spatiotemporal correlation features of each node are extracted based on the spatial correlation features of each node in the graph data. Based on the spatiotemporal correlation characteristics of each node, the probability of unit activation in each node is predicted, the initial solution of the unit combination problem under security constraints is determined based on the probability of unit activation in each node, and the radius of the trust region is dynamically determined based on the probability of unit activation in each node. Within the trust region, the optimal solution to the security-constrained unit combination problem is obtained by solving the problem based on the initial solution of the problem.

[0006] According to the above-mentioned method for solving the unit combination problem based on spatiotemporal deep learning and perceptual neighborhood search, the determination of the initial solution to the safety-constrained unit combination problem based on the probability of unit operation in each node specifically includes: A variable index set is formed based on a pre-set first threshold and a second threshold, selecting units whose probability is less than the first threshold. and the variable index set consisting of units with a probability greater than the second threshold. ; Based on variable index set and variable index set Generate an initial solution to the safety-constrained unit combination problem to be solved. :

[0007] in, Representative unit g exist t Start and stop variables at specific times.

[0008] According to the above-mentioned method for solving unit combination based on spatiotemporal deep learning and perceptual neighborhood search, the step of dynamically determining the radius of the trust region based on the probability of unit activation in each node specifically includes: Set a threshold range, and select units whose probability falls within the threshold range from each unit as units to be evaluated; The radius of the trust region is determined based on the probability of the unit to be evaluated starting up and the preset adjustment coefficient.

[0009] According to the above-mentioned method for solving unit combination based on spatiotemporal deep learning and perceptual neighborhood search, the step of determining the radius of the trust region based on the probability of the unit to be evaluated being activated and the preset adjustment coefficient specifically includes: The radius of the trust region is determined based on the following formula. :

[0010] in, For preset adjustment coefficients, I The corresponding number of units to be evaluated. The corresponding threshold interval i The probability of each unit starting up.

[0011] Based on the above-mentioned method for solving the unit combination problem based on spatiotemporal deep learning and perceptual neighborhood search, the optimal solution to the safety-constrained unit combination problem is obtained by solving the problem within the trust region using the initial solution of the problem, specifically including: Based on the initial solution to the security-constrained unit combination problem to be solved and the radius of the trust region, the following neighborhood constraints are determined:

[0012]

[0013]

[0014] in, g Representative unit, t Representing a moment, and These correspond to the initial solution and the solution calculated during the solution process, respectively. This is the deviation value; Determine the objective function and basic constraints of the safety-constrained unit combination problem to be solved; the basic constraints include unit output constraints, unit reserve capacity constraints, system reserve capacity constraints, unit ramp-up constraints, unit minimum start-up time constraints and minimum stop time constraints, unit start-up constraints, power flow equations, transmission line constraints, node power balance constraints, and reference node voltage phase angle constraints. Based on the neighborhood constraints and the basic constraints, the objective function is solved using a branch-and-cut algorithm to obtain the optimal solution to the safety-constrained unit combination problem.

[0015] Based on the above-mentioned unit combination solution method based on spatiotemporal deep learning and perceptual neighborhood search, the step of using a spatial correlation extraction model to extract the spatial correlation between each node in the graph data and obtain the spatial correlation features of each node specifically includes: Calculate nodes based on the following formula i with neighboring nodes j Pre-activation value between :

[0016] Among them, nodes j For nodes i The neighboring nodes, and For nodes i and nodes j The node feature vectors, and For connecting nodes i and nodes j The edge feature vectors of the transmission line in different directions, This is the weight matrix. For bias terms, The function is called the activation function, and || represents the concatenation operation. For nodes i The aggregation value in the input direction is calculated according to the following formula. and aggregated values ​​in the output direction :

[0017]

[0018] in, For nodes i The set of neighboring nodes; Based on nodes i Aggregate value in the input direction and aggregated values ​​in the output direction Calculate the nodes according to the following formula iSpatial correlation characteristics :

[0019] in, This is the weight matrix. This is a bias term.

[0020] Based on the above-mentioned unit combination solution method based on spatiotemporal deep learning and perceptual neighborhood search, the step of using a time correlation extraction model to extract the spatiotemporal correlation features of each node based on the spatial correlation features of each node in the graph data specifically includes: The spatially relevant features of each node are transposed to obtain the attention input vector for each node. For any node, its query vector, key vector, and value vector are calculated based on the attention input vector of the node. Attention weights are then calculated based on the query vector, key vector, and value vector of the node. Finally, the value vector of the node is weighted based on the attention weights to obtain the spatiotemporal correlation features of the node.

[0021] According to a second aspect of the present invention, a unit combination solution system based on spatiotemporal deep learning and perceptual neighborhood search is provided, comprising: The data acquisition unit is used to acquire the power grid topology, node characteristic data of each node in the power grid topology, and edge characteristic data of the transmission lines between nodes in the safety-constrained unit combination problem to be solved. The graph data construction unit is used to construct graph data based on the power grid topology, the node characteristic data of each node, and the edge characteristic data of the transmission line; the node characteristic data includes the historical net load data and equipment static parameters of the corresponding node. The spatial correlation extraction unit is used to extract the spatial correlation between each node in the graph data using a spatial correlation extraction model, and obtain the spatial correlation features of each node. The temporal correlation extraction unit is used to extract the spatiotemporal correlation features of each node based on the spatial correlation features of each node in the graph data using a temporal correlation extraction model. The initial solution and radius determination unit is used to predict the probability of unit activation in each node based on the spatiotemporal correlation characteristics of each node, determine the initial solution of the safety constraint unit combination problem to be solved based on the probability of unit activation in each node, and dynamically determine the radius of the trust region based on the probability of unit activation in each node. The solution unit is used to solve the optimal solution of the security-constrained unit combination problem in the trust region based on the initial solution of the problem.

[0022] According to a third aspect of the present invention, an electronic device is provided, comprising: a computer-readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is configured to read executable instructions stored in the computer-readable storage medium and execute the method as described in the first aspect.

[0023] According to a fourth aspect of the invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to perform the method as described in the first aspect.

[0024] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: By combining the advantages of machine learning methods and mathematical optimization, this approach addresses the problem that existing unit combination solution methods rarely consider the spatiotemporal correlation between variables by mining the spatiotemporal correlation between decision variables from historical data, thus effectively improving prediction accuracy. At the same time, by utilizing the perceptual neighborhood search method, it avoids the situation where directly fixing some decision variables after predicting the initial solution may lead to infeasible or suboptimal solutions due to insufficient fixing. This approach accelerates the solution speed while ensuring the quality of the solution, thereby ensuring the accuracy of power system dispatch. Attached Figure Description

[0025] Figure 1 A flowchart illustrating a unit combination solution method based on spatiotemporal deep learning and perceptual neighborhood search provided in an embodiment of the present invention; Figure 2 A schematic diagram of the unit combination prediction-search framework provided in this embodiment of the invention; Figure 3 This is a schematic diagram illustrating the solution to the safety constraint unit combination problem provided in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0027] This invention provides a method for solving unit combination based on spatiotemporal deep learning and perceptual neighborhood search, such as... Figure 1 As shown, it includes: Step 110: Obtain the power grid topology in the safety-constrained unit combination problem to be solved, the node characteristic data of each node in the power grid topology, and the edge characteristic data of the transmission lines between the nodes; Step 120: Construct graph data based on the power grid topology, the node characteristic data of each node, and the edge characteristic data of the transmission line; the node characteristic data includes the historical net load data and static parameters of the corresponding node. Step 130: Using a spatial correlation extraction model, extract the spatial correlation between each node in the graph data to obtain the spatial correlation features of each node; Step 140: Using the time correlation extraction model, extract the spatiotemporal correlation features of each node based on the spatial correlation features of each node in the graph data. Step 150: Based on the spatiotemporal correlation characteristics of each node, predict the probability of unit activation in each node, determine the initial solution of the safety constraint unit combination problem to be solved based on the probability of unit activation in each node, and dynamically determine the radius of the trust region based on the probability of unit activation in each node. Step 160: In the trust region, solve the problem based on the initial solution of the security-constrained unit combination problem to be solved, and obtain the optimal solution of the security-constrained unit combination problem to be solved.

[0028] Here, Figure 2 A schematic diagram of the unit combination prediction-search framework provided in an embodiment of the present invention is given, as follows: Figure 2 As shown, the entire method for solving the unit combination problem based on spatiotemporal deep learning and perceptual neighborhood search is divided into two stages: spatiotemporal deep learning variable prediction and perceptual neighborhood search. The spatiotemporal deep learning variable prediction stage includes transforming the SCUC problem into graph data and extracting the spatial feature data of the variables, and then using the attention mechanism to consider the temporal correlation of the variables to obtain the probabilities corresponding to the variables. The perceptual neighborhood search stage dynamically adjusts the search radius by selecting predicted variable values ​​and conducting perceptual neighborhood search, and obtains a high-quality solution by combining the branching and cutting method.

[0029] Specifically, the first step is to obtain the power grid topology, node characteristic data of each node in the topology, and edge characteristic data of the transmission lines between nodes in the safety-constrained unit combination problem to be solved. In some embodiments, the nodes in the power grid topology consist of buses and generating units, and the node characteristic data includes the historical net load data and static parameters of the corresponding nodes, such as the lower and upper limits of the unit's output. and Start-up costs and the voltage phase angle of the bus Device static parameters, bus net load data at historical moments. In a power grid topology, edges correspond to transmission lines connecting adjacent nodes. The edge characteristic data of a transmission line includes its susceptance. Thermal limit of transmission lines and time t trend Based on the above power grid topology, node characteristic data of each node, and edge characteristic data of transmission lines, graph data is constructed.

[0030] Considering the influence of constraints such as minimum start / stop time and grid security, the start-up and shutdown variables (indicating whether a corresponding unit is started) of each unit in the safety-constrained unit combination problem have certain spatiotemporal correlations. Therefore, the spatiotemporal correlations between decision variables can be mined from historical data, and machine learning methods can be used to learn the mapping from input data to SCUC solutions to accelerate the solution speed. Specifically, applying machine learning to the SCUC problem can fix the start-up and shutdown variables of some units based on predicted probabilities, transforming the large-scale SCUC problem into a smaller-scale MILP problem, which can then be solved using the branch and split method.

[0031] Specifically, a spatial correlation extraction model can be used to extract the spatial correlations between nodes in the graph data, obtaining the spatial correlation features of each node. Then, a temporal correlation extraction model can be used to extract the spatiotemporal correlation features of each node based on its spatial correlation features. The spatial correlation extraction model can be built based on a graph neural network, which learns the complex relationships and patterns between nodes in the graph data to extract the spatial correlations between nodes in the SCUC problem. The temporal correlation extraction model can be built based on a Transformer model to further consider the temporal correlations of the data.

[0032] In some embodiments, when extracting the spatial correlation between nodes in graph data, the spatial correlation extraction model can calculate the node correlation based on the following formula. i with neighboring nodes j Pre-activation value between :

[0033] Among them, nodes j For nodes i The neighboring nodes, and For nodes i and nodes j The node feature vectors, and For connecting nodes i and nodes j The edge feature vectors of the transmission line in different directions, This is the weight matrix. For bias terms, The function is called the activation function, and || represents the concatenation operation. For nodes i The aggregation value in the input direction is calculated according to the following formula. and aggregated values ​​in the output direction :

[0034]

[0035] in, For nodes i The set of neighboring nodes; Based on nodes i Aggregate value in the input direction and aggregated values ​​in the output direction Calculate the nodes according to the following formula i Spatial correlation characteristics :

[0036] in, This is the weight matrix. This is a bias term.

[0037] In other embodiments, when the temporal correlation extraction model further extracts temporal correlation based on the spatial correlation features of each node, it can utilize an attention mechanism to achieve temporal correlation extraction. Specifically, the spatial correlation features of each node can be transposed to obtain the attention input vector for each node:

[0038] in, For nodes i The attention input vector.

[0039] For nodes i Based on nodes i The attention input vector is used to calculate its query vector using an attention mechanism. Key vector Sum value vector :

[0040] in, , , It is a learnable linear mapping matrix used to transform raw input features into query vectors, key vectors, and value vectors.

[0041] Subsequently based on nodes i query vector Key vector Sum value vector Calculate attention weights. This is done by obtaining the relevance score through the scaled dot product of the query vector and the key vector, followed by softmax normalization to generate attention weights. The value vectors are then weighted based on these attention weights to obtain the final attention output representation, i.e., the node. i The spatiotemporal correlation characteristics.

[0042]

[0043] In the formula, This is a scaling factor used to mitigate gradient instability caused by excessively large dot product values ​​in high-dimensional spaces.

[0044] It should be noted that a training set containing the sample power grid topology, sample node feature data of each node in the corresponding power grid topology, sample edge feature data of transmission lines, and ideal start-stop variables of each node can be obtained, and the spatial correlation extraction model and the temporal correlation extraction model can be jointly trained using this training set.

[0045] After obtaining the spatiotemporal correlation features of each node, the probability of unit startup (i.e., startup / shutdown variable equals 1) in each node can be predicted. For example, the probability of unit startup in any node can be predicted using a fully connected layer and a sigmoid activation function:

[0046] in, This represents the probability of a unit being turned on within a node. It is the sigmoid activation function. and These are the weight matrix and bias term in the fully connected layer, respectively.

[0047] Subsequently, based on the predicted probabilities of unit startup at each node, the predicted values ​​of the start-up and shutdown variables are selected to construct an initial solution to the safety-constrained unit combination problem, reducing the scale of the SCUC problem. Simultaneously, the radius of the trust region is dynamically set according to the predicted probabilities. Within this radius, using the initial solution as the center, a branch-and-cut method for mixed-integer programming is employed to solve the smaller-scale SCUC problem, yielding the optimal solution to the safety-constrained unit combination problem.

[0048] In some embodiments, such as Figure 3 As shown, in order to construct an initial solution to the safety-constrained unit combination problem, after obtaining the probability of unit operation at each node, variables can be selected based on a pre-set first threshold (e.g., ) and second threshold (e.g. The variable index set consisting of units with a screening probability less than the first threshold. And the variable index set consisting of units with a probability greater than the second threshold. Then, the selected variables are assigned predicted values, either 0 or 1, as initial solutions. That is, based on the variable index set... and variable index set Generate an initial solution to the safety-constrained unit combination problem to be solved. :

[0049] in, The variable represents the start-up or shutdown of unit g at time t, where 1 indicates that the corresponding unit is on and 0 indicates that the corresponding unit is off.

[0050] In other embodiments, a threshold range can be set to dynamically determine the radius of the trust region. Units whose probability falls within this threshold range are selected from all units as units to be evaluated. Specifically, units that meet the following conditions are selected as units to be evaluated:

[0051] in, This represents the probability of the unit starting up. i This is the index of the units within the threshold range.

[0052] Subsequently, the radius of the trust region is set based on the probability of the unit to be evaluated starting up and the preset adjustment coefficient.

[0053] The radius of the trust region is determined based on the following formula. :

[0054] in, For preset adjustment coefficients, I The corresponding number of units to be evaluated. The corresponding threshold interval i The probability of each unit starting up.

[0055] Once the initial solution and the radius of the trust region of the safety-constrained unit combination problem are determined, the branch-and-cut algorithm can be used to solve the problem within the trust region, centered on the initial solution, to obtain the optimal solution.

[0056] In some embodiments, the initial solution of the safety-constrained unit combination problem to be solved can be used as a basis. and the radius of the trust region Determine the following neighborhood constraints:

[0057]

[0058]

[0059] in, g Representative unit, t Representing a moment, and These correspond to the initial solution and the solution calculated during the solution process, respectively. This is the deviation value.

[0060] Simultaneously, the objective function and basic constraints of the safety-constrained unit combination problem to be solved are determined. These basic constraints include unit output constraints, unit reserve capacity constraints, system reserve capacity constraints, unit ramp-up constraints, minimum start-up time constraints, minimum stop-down time constraints, unit start-up constraints, power flow equations, transmission line constraints, node power balance constraints, and reference node voltage phase angle constraints.

[0061] Specifically, the objective of the safety-constrained unit combination problem to be solved is to minimize the generation cost, and its objective function is:

[0062] in, This represents the startup cost of unit g (representing the operating costs of the unit). This represents the no-load cost of unit g (representing the cost of keeping the unit running). This represents the linear power generation cost of unit g (representing the consumption caused by the power generation of this unit). Let g be the output of unit g at time t; This is a 0-1 integer variable representing the start-up and shutdown status of the unit. A value of 1 indicates that the unit is in the start-up state, and a value of 0 indicates that the unit is in the stop state. This is a 0-1 integer variable representing the unit's start-up action. A value of 1 indicates that the unit has started up, and a value of 0 indicates that the unit has not started up.

[0063] Furthermore, the basic constraints include: Unit output constraints:

[0064]

[0065] in, and These represent the lower and upper limits of the output of unit g, respectively. Let g be the reserve capacity of unit g at time t; Unit standby capacity constraints:

[0066] in, The ten-minute ramp rate limit for unit g; System backup capacity constraints:

[0067] in, G The number of generator sets, Let q be the reserve capacity of unit q at time t; Unit ramp-up constraints:

[0068]

[0069] in, The hourly average ramp rate limit for unit g. and These are the start-up and stop ramp-up rate limits for unit g, respectively; Minimum start-up time constraints and minimum stop time constraints for the generator unit:

[0070]

[0071] in, and These represent the minimum start-up and stop times that unit g must meet; Unit start-up constraints:

[0072] Power flow equation:

[0073] in, Let the power flow of transmission line k be at time t. Let k be the susceptance of the transmission line. Let i be the voltage phase angle of bus i at time t. The voltage phase angle of bus o, which is connected to bus i, at time t; Transmission line constraints:

[0074] in, The thermal limit of transmission line k; Node power balance constraints:

[0075] in, This represents the net load value of bus n at time t (i.e., load minus renewable energy output); For the set of units connected to bus n, It is a set of transmission lines with bus n as the receiving end (i.e., transmission lines into which power flows into bus n). It is a set of transmission lines with bus n as the sending end (i.e., transmission lines from which power flows out of bus n). Reference node voltage phase angle constraint:

[0076] In the formula, The voltage phase angle of the reference node at time t is set to 0 to fix the system's degrees of freedom.

[0077] Based on the aforementioned neighborhood constraints and basic constraints, the optimal solution to the safety-constrained unit combination problem can be obtained by using the branch-and-cut algorithm to solve the objective function.

[0078] In summary, the method provided by this invention combines the advantages of machine learning and mathematical optimization. By mining the spatiotemporal correlations between decision variables from historical data, it solves the problem that existing unit combination solution methods rarely consider the spatiotemporal correlations between variables, effectively improving the accuracy of prediction. At the same time, by using the perceptual neighborhood search method, it avoids the situation where directly fixing some decision variables after predicting the initial solution may lead to infeasible or suboptimal solutions due to insufficient accuracy. This accelerates the solution speed while ensuring the quality of the solution, thus ensuring the accuracy of power system dispatch.

[0079] The following describes the unit combination solution system based on spatiotemporal deep learning and perceptual neighborhood search provided by the present invention. The unit combination solution system based on spatiotemporal deep learning and perceptual neighborhood search described below can be referred to in correspondence with the unit combination solution method based on spatiotemporal deep learning and perceptual neighborhood search described above.

[0080] This invention provides a unit combination solution system based on spatiotemporal deep learning and perceptual neighborhood search, comprising: The data acquisition unit is used to acquire the power grid topology, node characteristic data of each node in the power grid topology, and edge characteristic data of the transmission lines between nodes in the safety-constrained unit combination problem to be solved. The graph data construction unit is used to construct graph data based on the power grid topology, the node characteristic data of each node, and the edge characteristic data of the transmission line; the node characteristic data includes the historical net load data and equipment static parameters of the corresponding node. The spatial correlation extraction unit is used to extract the spatial correlation between each node in the graph data using a spatial correlation extraction model, and obtain the spatial correlation features of each node. The temporal correlation extraction unit is used to extract the spatiotemporal correlation features of each node based on the spatial correlation features of each node in the graph data using a temporal correlation extraction model. The initial solution and radius determination unit is used to predict the probability of unit activation in each node based on the spatiotemporal correlation characteristics of each node, determine the initial solution of the safety constraint unit combination problem to be solved based on the probability of unit activation in each node, and dynamically determine the radius of the trust region based on the probability of unit activation in each node. The solution unit is used to solve the optimal solution of the security-constrained unit combination problem in the trust region based on the initial solution of the problem.

[0081] This invention provides an electronic device, including: a computer-readable storage medium and a processor; The computer-readable storage medium is used to store executable instructions; The processor is configured to read executable instructions stored in the computer-readable storage medium and execute the method as described in any of the above embodiments.

[0082] This invention provides a computer-readable storage medium storing computer instructions that cause a processor to perform the method described in any of the above embodiments.

[0083] This invention provides a computer program product, including a computer program or instructions, which, when executed by a processor, implement the method described in any of the above embodiments.

[0084] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for solving unit combination based on spatiotemporal deep learning and perceptual neighborhood search, characterized in that, include: Obtain the power grid topology, node characteristic data of each node in the power grid topology, and edge characteristic data of the transmission lines between nodes in the safety-constrained unit combination problem to be solved; Based on the power grid topology, the node characteristic data of each node, and the edge characteristic data of the transmission line, graph data is constructed; the node characteristic data includes the historical net load data and static parameters of the corresponding node. Using a spatial correlation extraction model, the spatial correlation between nodes in the graph data is extracted to obtain the spatial correlation features of each node. Using a time correlation extraction model, the spatiotemporal correlation features of each node are extracted based on the spatial correlation features of each node in the graph data. Based on the spatiotemporal correlation characteristics of each node, the probability of unit activation in each node is predicted, the initial solution of the unit combination problem under security constraints is determined based on the probability of unit activation in each node, and the radius of the trust region is dynamically determined based on the probability of unit activation in each node. Within the trust region, the optimal solution to the security-constrained unit combination problem is obtained by solving the problem based on the initial solution of the problem.

2. The unit combination solution method based on spatiotemporal deep learning and perceptual neighborhood search as described in claim 1, characterized in that, The process of determining the initial solution to the safety-constrained unit combination problem based on the probability of unit activation at each node specifically includes: A variable index set is formed based on a pre-set first threshold and a second threshold, selecting units whose probability is less than the first threshold. and the variable index set consisting of units with a probability greater than the second threshold. ; Based on variable index set and variable index set Generate an initial solution to the safety-constrained unit combination problem to be solved. : in, Representative unit g exist t Start and stop variables at specific times.

3. The unit combination solution method based on spatiotemporal deep learning and perceptual neighborhood search as described in claim 2, characterized in that, The method of dynamically determining the radius of the trust region based on the probability of unit activation in each node specifically includes: Set a threshold range, and select units whose probability falls within the threshold range from each unit as units to be evaluated; The radius of the trust region is determined based on the probability of the unit to be evaluated starting up and the preset adjustment coefficient.

4. The unit combination solution method based on spatiotemporal deep learning and perceptual neighborhood search as described in claim 3, characterized in that, The determination of the radius of the trust region based on the probability of the unit to be evaluated starting up and a preset adjustment coefficient specifically includes: The radius of the trust region is determined based on the following formula. : in, For preset adjustment coefficients, I The corresponding number of units to be evaluated. The corresponding threshold interval i The probability of each unit starting up.

5. The unit combination solution method based on spatiotemporal deep learning and perceptual neighborhood search as described in claim 4, characterized in that, Within the trust region, the optimal solution to the security-constrained unit combination problem is obtained by solving based on the initial solution of the problem, specifically including: Based on the initial solution to the security-constrained unit combination problem to be solved and the radius of the trust region, the following neighborhood constraints are determined: in, g Representative unit, t Representing a moment, and These correspond to the initial solution and the solution calculated during the solution process, respectively. This is the deviation value; Determine the objective function and basic constraints of the safety-constrained unit combination problem to be solved; the basic constraints include unit output constraints, unit reserve capacity constraints, system reserve capacity constraints, unit ramp-up constraints, unit minimum start-up time constraints and minimum stop time constraints, unit start-up constraints, power flow equations, transmission line constraints, node power balance constraints, and reference node voltage phase angle constraints. Based on the neighborhood constraints and the basic constraints, the objective function is solved using a branch-and-cut algorithm to obtain the optimal solution to the safety-constrained unit combination problem.

6. The unit combination solution method based on spatiotemporal deep learning and perceptual neighborhood search as described in claim 1, characterized in that, The spatial correlation extraction model is used to extract the spatial correlation between nodes in the graph data, obtaining the spatial correlation features of each node, specifically including: Calculate nodes based on the following formula i with neighboring nodes j Pre-activation value between : Among them, nodes j For nodes i The neighboring nodes, and For nodes i and nodes j The node feature vectors, and For connecting nodes i and nodes j The edge feature vectors of the transmission line in different directions, This is the weight matrix. For bias terms, The function is called the activation function, and || represents the concatenation operation. For nodes i The aggregation value in the input direction is calculated according to the following formula. and aggregated values ​​in the output direction : in, For nodes i The set of neighboring nodes; Based on nodes i Aggregate value in the input direction and aggregated values ​​in the output direction Calculate the nodes according to the following formula i Spatial correlation characteristics : in, This is the weight matrix. This is a bias term.

7. The unit combination solution method based on spatiotemporal deep learning and perceptual neighborhood search as described in claim 1, characterized in that, The method of using a time correlation extraction model to extract the spatiotemporal correlation features of each node based on the spatial correlation features of each node in the graph data specifically includes: The spatially relevant features of each node are transposed to obtain the attention input vector for each node. For any node, its query vector, key vector, and value vector are calculated based on the attention input vector of the node. Attention weights are then calculated based on the query vector, key vector, and value vector of the node. Finally, the value vector of the node is weighted based on the attention weights to obtain the spatiotemporal correlation features of the node.

8. A unit combination solution system based on spatiotemporal deep learning and perceptual neighborhood search, characterized in that, include: The data acquisition unit is used to acquire the power grid topology, node characteristic data of each node in the power grid topology, and edge characteristic data of the transmission lines between nodes in the safety-constrained unit combination problem to be solved. The graph data construction unit is used to construct graph data based on the power grid topology, the node characteristic data of each node, and the edge characteristic data of the transmission line; the node characteristic data includes the historical net load data and equipment static parameters of the corresponding node. The spatial correlation extraction unit is used to extract the spatial correlation between each node in the graph data using a spatial correlation extraction model, and obtain the spatial correlation features of each node. The temporal correlation extraction unit is used to extract the spatiotemporal correlation features of each node based on the spatial correlation features of each node in the graph data using a temporal correlation extraction model. The initial solution and radius determination unit is used to predict the probability of unit activation in each node based on the spatiotemporal correlation characteristics of each node, determine the initial solution of the safety constraint unit combination problem to be solved based on the probability of unit activation in each node, and dynamically determine the radius of the trust region based on the probability of unit activation in each node. The solution unit is used to solve the optimal solution of the security-constrained unit combination problem in the trust region based on the initial solution of the problem.

9. An electronic device, characterized in that, include: Computer-readable storage media and processors; The computer-readable storage medium is used to store executable instructions; The processor is configured to read executable instructions stored in the computer-readable storage medium and execute the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a processor to perform the method as described in any one of claims 1-7.

Images