CN-122000989-A - Unit combination solving method and system based on space-time deep learning and perception neighborhood search

CN122000989ACN 122000989 ACN122000989 ACN 122000989ACN-122000989-A

Abstract

The invention discloses a unit combination solving method and a unit combination solving system based on space-time deep learning and perception neighborhood search, which belong to the technical field of power system dispatching, combine the advantages of a machine learning method and mathematical optimization, solve the problem that space-time correlation among variables is rarely considered in the traditional unit combination solving method by mining space-time correlation among decision variables from historical data, effectively improve the prediction precision, and simultaneously avoid the condition that the mode of directly fixing part of decision variables after the initial solution is predicted can be infeasible or suboptimal due to inaccurate fixation by utilizing the perception neighborhood searching method, ensure the knowing quality while accelerating the solving speed and ensure the dispatching accuracy of the power system.

Inventors

YU YAOWEN
ZHANG HAOXIANG
LI YUANZHENG
ZHAO YONG
Shi Chulun
LIU SHIWEI

Assignees

华中科技大学

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (10)

1. A unit combination solving method based on space-time deep learning and perception neighborhood search is characterized by comprising the following steps: Acquiring a power grid topological structure in a to-be-solved safety constraint unit combination problem, node characteristic data of each node in the power grid topological structure and edge characteristic data of a transmission line between the nodes; Constructing graph data based on the power grid topological structure, node characteristic data of each node and edge characteristic data of the transmission line, wherein the node characteristic data comprises historical payload data and equipment static parameters of the corresponding node; extracting the spatial correlation among all nodes in the graph data by using a spatial correlation extraction model to obtain spatial correlation characteristics of all nodes; Extracting space-time correlation characteristics of each node on the basis of space-time correlation characteristics of each node in the graph data by using a time correlation extraction model; predicting the probability of unit opening in each node based on the space-time related characteristics of each node, determining an initial solution of the safety constraint unit combination problem to be solved based on the probability of unit opening in each node, and dynamically determining the radius of a trust domain based on the probability of unit opening in each node; and in the trust domain, solving based on the initial solution of the to-be-solved safety constraint unit combination problem to obtain the optimal solution of the to-be-solved safety constraint unit combination problem.
2. The method for solving the set combination problem based on space-time deep learning and perceptual neighborhood search according to claim 1, wherein the determining the initial solution of the set combination problem of the security constraint to be solved based on the probability of set opening in each node specifically comprises: Variable index set composed of units with probability smaller than first threshold value based on preset first threshold value and second threshold value And a variable index set formed by units with probability larger than the second threshold value ; Based on variable index set Variable index set Generating an initial solution of the combined problem of the safety constraint unit to be solved : Wherein, the Representing the start-stop variable of the unit g at the time t.
3. The method for solving the unit combination based on space-time deep learning and sensing neighborhood search according to claim 2, wherein the method for dynamically determining the radius of the trust domain based on the probability of unit turn-on in each node comprises the following steps: setting a threshold interval, and screening units with probability within the threshold interval from each unit as units to be evaluated; And determining the radius of the trust zone based on the probability of the unit to be evaluated being started and a preset adjustment coefficient.
4. The method for solving the unit combination based on space-time deep learning and sensing neighborhood search according to claim 3, wherein the determining the radius of the trust zone based on the probability of the unit to be evaluated being started and a preset adjustment coefficient specifically comprises: Determining radius of trust domain based on the following formula : Wherein, the For presetting the adjustment coefficient, I corresponds to the number of units to be evaluated, And the probability of the ith unit opening in the corresponding threshold interval.
5. The method for solving the set combination based on space-time deep learning and perceptual neighborhood search according to claim 4, wherein in the trust domain, the solution is performed based on the initial solution of the set combination problem of the safety constraint to be solved, so as to obtain the optimal solution of the set combination problem of the safety constraint to be solved, and the method specifically comprises: based on the initial solution of the safety constraint unit combination problem to be solved and the radius of the trust zone, determining the following neighborhood constraint: wherein g represents a unit, t represents a time, And Respectively corresponding to the initial solution and the solution calculated in the solving process, Is a deviation value; Determining an objective function of the combination problem of the safety constraint unit to be solved and a basic constraint condition, wherein the basic constraint condition comprises unit output constraint, unit standby capacity constraint, system standby capacity constraint, unit climbing constraint, unit shortest starting time constraint and shortest stopping time constraint, unit starting constraint, tide equation, transmission line constraint, node power balance constraint and reference node voltage phase angle constraint; And solving the objective function by utilizing a branch cutting algorithm based on the neighborhood constraint and the basic constraint condition to obtain an optimal solution of the safety constraint unit combination problem to be solved.
6. The method for solving the unit combination based on space-time deep learning and sensing neighborhood search according to claim 1, wherein the extracting spatial correlation among nodes in the graph data by using a spatial correlation extraction model to obtain spatial correlation characteristics of each node comprises the following steps: Calculating a pre-activation value between node i and neighbor node j based on : Wherein node j is a neighbor node of node i, And (3) with The node feature vectors for node i and node j, And (3) with To connect the edge feature vectors of the transmission line of node i and node j in different directions, As a matrix of weights, the weight matrix, As a result of the bias term, In order to activate the function, the I represents a splicing operation; For node i, the aggregate value in its input direction is calculated according to And aggregate value in output direction : Wherein, the A neighbor node set of the node i; Aggregation value based on node i input direction And aggregate value in output direction Calculating the spatial correlation characteristic of node i according to the following formula : Wherein, the As a matrix of weights, the weight matrix, Is a bias term.
7. The method for solving the unit combination based on space-time deep learning and sensing neighborhood search according to claim 1, wherein the extracting the space-time correlation feature of each node based on the space correlation feature of each node in the graph data by using a time correlation extraction model specifically comprises: Transpose the spatial correlation characteristics of each node to obtain the attention input vector of each node; For any node, respectively calculating a query vector, a key vector and a value vector based on the attention input vector of the any node, calculating attention weights based on the query vector, the key vector and the value vector of the any node, and weighting the value vector of the any node based on the attention weights to obtain the space-time related characteristics of the any node.
8. A unit combination solving system based on space-time deep learning and perception neighborhood search is characterized by comprising: The data acquisition unit is used for acquiring a power grid topological structure in a to-be-solved safety constraint unit combination problem, node characteristic data of each node in the power grid topological structure and edge characteristic data of a transmission line between the nodes; the system comprises a graph data construction unit, a graph data generation unit and a data storage unit, wherein the graph data construction unit is used for constructing graph data based on the power grid topological structure, node characteristic data of each node and edge characteristic data of the transmission line, and the node characteristic data comprises historical payload data and equipment static parameters of the corresponding node; The spatial correlation extraction unit is used for extracting the spatial correlation among all nodes in the graph data by using a spatial correlation extraction model to obtain the spatial correlation characteristics of all nodes; A time correlation extraction unit, configured to extract space-time correlation features of each node on the basis of space-time correlation features of each node in the graph data by using a time correlation extraction model; The initial solution and radius determining unit is used for predicting the probability of unit opening in each node based on the space-time related characteristics of each node, determining the initial solution of the safety constraint unit combination problem to be solved based on the probability of unit opening in each node, and dynamically determining the radius of the trust domain based on the probability of unit opening in each node; and the solving unit is used for solving the safety constraint unit combination problem to be solved based on the initial solution of the safety constraint unit combination problem to be solved in the trust domain to obtain the optimal solution of the safety constraint unit combination problem to be solved.
9. An electronic device comprises a computer readable storage medium and a processor; The computer-readable storage medium is for storing executable instructions; The processor is configured to read executable instructions stored in the computer readable storage medium and perform the method of any one of claims 1-7.
10. A computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, the computer instructions for causing a processor to perform the method of any one of claims 1-7.

Description

Unit combination solving method and system based on space-time deep learning and perception neighborhood search Technical Field The invention belongs to the technical field of power system dispatching, and particularly relates to a unit combination solving method and system based on space-time deep learning and perception neighborhood searching. Background Safety restraint unit combination (security constrained unit commitment, SCUC) is an optimization problem in power system operation. The problem is to arrange the start-stop state of the unit and the generated power of each period in a certain scheduling period (such as 24 h) in advance to meet the system requirement, and the aim is to minimize the total running cost of the electric power system. Therefore, in the spot power market, how to solve the unit combination problem within the clearing time window is important to obtain the market scheduling decision. The problem is generally described as a form of Mixed-integer linear Programming (MILP), where the 0-1 variables include the start-stop state and corresponding start-up actions of the genset, and the continuous variables include the genset generated power and demand response power. The problems include generator set constraints, demand response constraints, system supply and demand power balance, grid safety constraints, and the like. Currently practice is to use mainly branch-and-cut based commercial optimization solvers (such as Gurobi or fir COPT) for solving. The development of electric power market rules, the addition of more novel market bodies and the refined modeling of an electric power system make efficient solution of electric power optimization problems more challenging. Furthermore, while the SCUC problem is repeatedly run every day or hour, the business solver treats the crew combination problem of each run as a separate MILP problem and solves it separately, resulting in a failure to accumulate useful experience. The existing research generally predicts the 0-1 variable of the unit by an artificial intelligence method, and obtains an initial solution based on the prediction probability or a preset rule fixed variable, so that the SCUC problem is converted into a MILP problem with a smaller scale to solve. In summary, the method for solving the SCUC problem in power system dispatching has the following problems or is worth improving that the existing solution prediction method is used for predicting the start-stop variables of all units respectively, so that the start-stop variables of all units in the unit combination problem are ignored to have certain time-space correlation, and meanwhile, the condition of infeasibility or suboptimal solution can be caused due to the fact that the mode of directly fixing part of decision variables after the initial solution is predicted is not sufficiently accurately fixed, so that the accuracy of solving the safety constraint unit combination problem is insufficient. Disclosure of Invention Aiming at the defects or improvement demands of the prior art, the invention provides a unit combination solving method and a system based on space-time deep learning and perception neighborhood search, so that the problem that space-time correlation among variables is rarely considered in the existing unit combination solving method is solved, and meanwhile, the condition that infeasibility or suboptimal solutions occur due to the fact that partial decision variables are fixed directly after initial solution prediction is not accurate enough is avoided. To achieve the above object, according to a first aspect of the present invention, there is provided a unit combination solving method based on space-time deep learning and perceptual neighborhood search, including: Acquiring a power grid topological structure in a to-be-solved safety constraint unit combination problem, node characteristic data of each node in the power grid topological structure and edge characteristic data of a transmission line between the nodes; Constructing graph data based on the power grid topological structure, node characteristic data of each node and edge characteristic data of the transmission line, wherein the node characteristic data comprises historical payload data and equipment static parameters of the corresponding node; extracting the spatial correlation among all nodes in the graph data by using a spatial correlation extraction model to obtain spatial correlation characteristics of all nodes; Extracting space-time correlation characteristics of each node on the basis of space-time correlation characteristics of each node in the graph data by using a time correlation extraction model; predicting the probability of unit opening in each node based on the space-time related characteristics of each node, determining an initial solution of the safety constraint unit combination problem to be solved based on the probability of unit opening in each node, and dynamically determining th