Security constraint unit commitment model optimization method and device of power system

Abstract

The invention discloses a method and a device for optimizing a security constraint unit commitment model of a power system, and belongs to the technical field of power system dispatching. The method comprises the following steps: training an initial graph neural network by taking a bigraph corresponding to a model parameter feature in a historical power flow constraint as input and power flow out-of-limit data corresponding to the model parameter feature as output, thereby learning a mapping relation between the model parameter feature and a power flow constraint out-of-limit condition, and finally obtaining a target graph neural network; the target graph neural network is utilized to predict the current invalid power flow constraint, and the current invalid power flow constraint is eliminated from all constraints of the current security constraint unit commitment model to obtain the target security constraint unit commitment model, so that the consumption of the power flow constraint on computing resources can be reduced, the solving efficiency of the SCUC problem can be improved, and the optimal solution can be found; and the output dispatching efficiency of the power system can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system dispatching, and more specifically, relates to a method and device for optimizing a security-constrained unit commitment model of a power system. Background Art

[0002] With the continuous growth of the scale of the power system, the need for independent system operators to efficiently solve large-scale security-constrained unit commitment (SCUC) problems has become particularly urgent. The solution goal of the SCUC problem is to reasonably optimize the allocation of power generation resources, determine the start-stop states and output levels of generator sets under the conditions of meeting system balance, reserve, and network operation constraints, so as to minimize the total operating cost of the entire system.

[0003] The SCUC problem needs to consider both the economy and security of system operation, and the introduction of power flow constraints is used to ensure system power balance and line thermal stability. However, dense power flow constraints usually significantly increase the scale and complexity of the SCUC problem, resulting in too long a solution time or difficulty in finding the optimal solution. Summary of the Invention

[0004] In view of the above defects or improvement requirements of the prior art, the present invention provides a method and device for optimizing a security-constrained unit commitment model of a power system, aiming to solve the technical problem that the existing SCUC model has too long a solution time or difficulty in finding the optimal solution.

[0005] To achieve the above object, according to one aspect of the present invention, a method for optimizing a security-constrained unit commitment model of a power system is provided, including: S1: Determine a training sample set from the historical power flow constraint state data corresponding to multiple security-constrained unit commitment models of the power system, and each training sample includes model parameter features and their corresponding power flow violation data; wherein, the security-constrained unit commitment model is used to formulate a multi-period unit start-stop plan with the lowest total operating cost as the optimization goal under the condition of meeting the security constraints of the power system; S2: Use the bipartite graph corresponding to the model parameter features as the input and the power flow violation data corresponding to the model parameter features as the output to train an initial graph neural network to obtain a target graph neural network; S3: Input the model parameter features of the current security-constrained unit commitment model to be optimized corresponding to the power system into the target graph neural network to predict the current invalid power flow constraint; S4: Remove the current invalid power flow constraint from all the constraints of the current security-constrained unit commitment model to obtain a target security-constrained unit commitment model.

[0006] Furthermore, the training sample is expressed as: ;in, i =2, ..., N-1; j =1, ..., M; S j For the j The model parameter characteristics of training samples, For the j Of the training samples i The limit state of the power flow constraint; N is the j The total number of flow crossing data in the training samples; M is the total number of training samples.

[0007] Furthermore, if j The limit state corresponding to each power flow constraint in the training samples X j is 1, then the historical mid-tidal flow constraint state data g ( u g,t , s g,t , d g,t , p g,t ) = 0; if j The limit state corresponding to each power flow constraint in the training samples X j is 0, then the historical mid-tidal flow constraint state data g ( u g,t , s g,t , d g,t , p g,t ) < 0; in, u g,t , s g,t , d g,t , p g,t The units are defined separately g At the moment t state variables, startup variables, shutdown variables and output levels, g ( . ) is the line power flow constraint of the UC problem.

[0008] Further, the S3 includes: Input the model parameter characteristics of the current security-constrained unit commitment model to be optimized for the power system into the target graph neural network, and output the effective probabilities of each current power flow constraint; Regard the current power flow constraints corresponding to the effective probabilities exceeding the preset threshold as the current invalid power flow constraints.

[0009] Furthermore, the loss function for training is: ; where S is the training set, X is its index, and P j is the effective probability of the j th current power flow constraint, J 0 , J 1 are respectively the sets of current power flow constraints with the reaching limit state taking values of 0 and 1, and a0 and a1 are unbalanced learning parameters.

[0010] According to another aspect of the present invention, there is provided an optimization device for a security-constrained unit commitment model of a power system, including: A determination module, configured to determine a training sample set from the historical power flow constraint state data corresponding to multiple security-constrained unit commitment models of the power system, and each training sample includes model parameter characteristics and their corresponding power flow over-limit data; wherein, the security-constrained unit commitment model is used to formulate a multi-period unit on-off schedule with the lowest total operating cost as the optimization goal under the condition of satisfying the security constraints of the power system; A training module, configured to train an initial graph neural network with the bipartite graph corresponding to the model parameter characteristics as the input and the power flow over-limit data corresponding to the model parameter characteristics as the output to obtain a target graph neural network; A prediction module, configured to input the model parameter characteristics of the current security-constrained unit commitment model to be optimized for the power system into the target graph neural network to predict the current invalid power flow constraints; An elimination module, configured to eliminate the current invalid power flow constraints from all the constraints of the current security-constrained unit commitment model to obtain a target security-constrained unit commitment model.

[0011] According to another aspect of the present invention, there is provided a solution method for a security-constrained unit commitment model of a power system, including: executing the optimization method for the security-constrained unit commitment model of the power system; solving the target security-constrained unit commitment model to obtain an optimal scheduling solution.

[0012] According to another aspect of the present invention, there is provided a scheduling method for a power system, including: executing the solution method for the security-constrained unit commitment model of the power system; using the optimal scheduling solution to perform power output scheduling for the power system.

[0013] According to another aspect of the present invention, there is provided a power system including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the above method are implemented.

[0014] According to another aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0015] Generally speaking, compared with the prior art by the above technical solution conceived by the present invention, the following beneficial effects can be achieved: (1) The present invention provides an optimization method for a security-constrained unit commitment model of a power system. By using the bipartite graph corresponding to the model parameter characteristics in the historical power flow constraint as the input and the power flow violation data corresponding to the model parameter characteristics as the output, the initial graph neural network is trained to learn the mapping relationship between the model parameter characteristics and the power flow constraint violation situation, and finally the target graph neural network is obtained; the target graph neural network is used to predict the current invalid power flow constraint, and the current invalid power flow constraint is removed from all the constraints of the current security-constrained unit commitment model to obtain the target security-constrained unit commitment model, which can reduce the consumption of computing resources by the power flow constraint, improve the solution efficiency of the SCUC problem, find the optimal solution, and further improve the output scheduling efficiency of the power system.

[0016] (2) The training sample in this solution is expressed as: , designed in this way, mainly collects the historical states of the power flow constraints and can realize the identification of the power flow constraints.

[0017] (3) In this solution, if the reaching limit state j corresponding to each power flow constraint in the X j -th training sample is 1, then the historical power flow constraint state data g ( u g,t , s g,t , d g,t , p g,t ) = 0; if the reaching limit state j corresponding to each power flow constraint in the X j -th training sample is 0, then the historical power flow constraint state data g ( u g,t , s g,t , dg,t , p g,t ) < 0; Designed in this way, considering xxx, xxx can be achieved.

[0018] (4) In this solution, the current power flow constraint corresponding to the effective probability exceeding the preset threshold is regarded as the current ineffective power flow constraint. By setting the threshold to judge whether the current power flow constraint is effective, the operation is simple and the calculation complexity is low.

[0019] (5) The loss function trained in this solution is designed in this way, considering the unbalanced characteristics of the power flow constraint, and can achieve efficient learning for small samples and more important samples that reach the limit. Brief Description of the Drawings

[0020] Figure 1 is a flowchart of an optimization method for a security-constrained unit commitment model of a power system provided by an embodiment of the present invention.

[0021] Figure 2 is a schematic structural diagram of a security-constrained unit commitment model provided by an embodiment of the present invention.

[0022] Figure 3 is a simulation comparison diagram before and after the execution of an optimization method for a security-constrained unit commitment model of a power system provided by an embodiment of the present invention. Detailed Embodiment

[0023] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present 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 only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0024] Such as Figure 1As shown in the figure, this embodiment provides an optimization method for a security-constrained unit commitment model of a power system, including: S1: Determine a training sample set from the historical power flow constraint state data corresponding to multiple security-constrained unit commitment models of the power system. Each training sample contains model parameter features and their corresponding power flow violation data. Among them, the security-constrained unit commitment model is used to formulate a multi-period unit on / off schedule with the lowest total operating cost as the optimization goal under the condition of satisfying the security constraints of the power system. S2: Use the bipartite graph corresponding to the model parameter features as the input and the power flow violation data corresponding to the model parameter features as the output to train the initial graph neural network to obtain the target graph neural network. S3: Input the model parameter features of the current security-constrained unit commitment model to be optimized corresponding to the power system into the target graph neural network to predict the current invalid power flow constraint. Among them, the structure of the current security-constrained unit commitment model is as Figure 2 shown. S4: Eliminate the current invalid power flow constraint from all the constraints of the current security-constrained unit commitment model to obtain the target security-constrained unit commitment model. The comparison of the calculation efficiency before and after reduction is as Figure 3 shown.

[0025] It should be noted that the current security-constrained unit commitment model can be expressed as: ; In the formula u g,t , s g,t , d g,t respectively define the state variable, start-up variable, and shutdown variable of unit g at time t . p g,t defines the output level of unit g at time t . g ( . ) is the line power flow constraint of the UC problem, z ( . ) is the other system-level constraint of the UC problem, h ( . ) is the single-unit constraint of the UC problem, f ( . ) is this objective function, indicating the minimization of the operating cost. Then the target security-constrained unit commitment model is expressed as: ; g'( . ) is the power flow constraint after reduction. The reduced SCUC model can greatly reduce the problem complexity, so the calculation efficiency can be improved.

[0026] As an alternative implementation, the training sample is represented as: ; where i = 2,..., N - 1; j = 1,..., M; S j is the model parameter feature of the j th training sample, is the violation state of the j th power flow constraint in the i th training sample; N is the total number of power flow violation data in the j th training sample; M is the total number of training samples.

[0027] As an alternative implementation, if the violation state j corresponding to each power flow constraint in the X j th training sample is 1, then the historical power flow constraint state data g ( u g,t , s g,t , d g,t , p g,t ) = 0; if the violation state j corresponding to each power flow constraint in the X j th training sample is 0, then the historical power flow constraint state data g ( u g,t , s g,t , d g,t , p g,t ) < 0; where u g,t , s g,t , d g,t , p g,t respectively define the state variable, start-up variable, shutdown variable, and output level of the unit g at time t , g ( . ) is the line power flow constraint of the UC problem.

[0028] As an alternative implementation, S3 includes: inputting the model parameter characteristics of the current security-constrained unit commitment model corresponding to the power system into the target graph neural network to output the effective probabilities of each current power flow constraint; regarding the current power flow constraints with effective probabilities exceeding the preset threshold as the current invalid power flow constraints. Specifically, based on the following formula, the results of the graph neural network are converted into the power flow constraint identification results of the SCUC problem: ; according to the preset threshold decide J 1 , J 0 , for the original model, J 1 , J 0 the power flow constraints are respectively retained and reduced.

[0029] As an alternative implementation, the loss function for training is: ; where S is the training set, X is its index, P j is the effective probability of the j th current power flow constraint, J 0 , J 1 are respectively the sets of current power flow constraints with the reaching limit state taking values of 0 and 1, and a0, a1 are the imbalance learning parameters.

[0030] It should be noted that P j is the probability that the j th power flow constraint fitted by the graph neural network reaches the limit. After passing through the sigmoid activation function, its value must be in [0,1]. J 0 , J 1 are respectively the sets of power flow constraints with the reaching limit state taking values of 0 and 1. Since in the process of power flow constraint identification, the number of active power flow constraints, that is, the number of power flow constraints reaching the limit, is extremely small; and misjudging an un-reached limit power flow constraint as reaching the limit has a much smaller impact on the subsequent calculation results than misjudging a reached limit power flow constraint as not reaching the limit. Therefore, we introduce the imbalance learning parameters a0, a1 to make the learning process pay more attention to the learning of small samples and more important reached limit samples. In practice, usually set a 0 << a 1 .

[0031] In one of the embodiments, an optimization device for the security-constrained unit commitment model of a power system is provided, including: A determination module, configured to determine a training sample set from historical power flow constraint status data corresponding to multiple security-constrained unit commitment models of a power system, where each training sample includes model parameter features and corresponding power flow violation data; wherein, the security-constrained unit commitment model is used to formulate a multi-period unit on-off schedule with the lowest total operating cost as the optimization goal under the condition of satisfying the security constraints of the power system. A training module, configured to use the bipartite graph corresponding to the model parameter features as the input and the power flow violation data corresponding to the model parameter features as the output to train an initial graph neural network to obtain a target graph neural network. A prediction module, configured to input the model parameter features of the current security-constrained unit commitment model to be optimized corresponding to the power system into the target graph neural network to predict the current invalid power flow constraint. An elimination module, configured to eliminate the current invalid power flow constraint from all the constraints of the current security-constrained unit commitment model to obtain a target security-constrained unit commitment model.

[0032] In one embodiment, a method for solving a security-constrained unit commitment model of a power system is provided, including: executing an optimization method for the security-constrained unit commitment model of the power system; solving the target security-constrained unit commitment model to obtain an optimal scheduling solution.

[0033] In one embodiment, a method for scheduling a power system is provided, including: executing a method for solving a security-constrained unit commitment model of the power system; using the optimal scheduling solution to perform power output scheduling of the power system.

[0034] In one embodiment, a power system is provided, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the above method are implemented.

[0035] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0036] Those skilled in the art can easily understand that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for optimizing a security-constrained unit commitment model of a power system, characterized in that, Comprising: S1: Determine a training sample set from the historical power flow constraint status data corresponding to multiple security-constrained unit commitment models of a power system. Each training sample includes model parameter features and their corresponding power flow violation data. Wherein, the security-constrained unit commitment model is used to formulate a multi-period unit on / off schedule with the lowest total operating cost as the optimization goal under the condition of satisfying the security constraints of the power system; S2: Use the bipartite graph corresponding to the model parameter features as the input and the power flow violation data corresponding to the model parameter features as the output to train an initial graph neural network to obtain a target graph neural network; S3: Input the model parameter features of the current security-constrained unit commitment model to be optimized corresponding to the power system into the target graph neural network to predict the current invalid power flow constraint; S4: Remove the current invalid power flow constraint from all the constraints of the current security-constrained unit commitment model to obtain a target security-constrained unit commitment model.

2. The safety-constrained unit commitment model optimization method for the power system according to claim 1, wherein, The training samples are represented as: ; wherein, i = 2, ..., N - 1; j = 1, ..., M; S j is the model parameter feature of the j th training sample, is the reaching limit state of the j th power flow constraint in the i th training sample; N is the total number of power flow over - limit data in the j th training sample; M is the total number of training samples.

3. The method for optimizing a security-constrained unit commitment model of a power system according to claim 2, wherein If for each violation state corresponding to each power flow constraint in the j th training sample is 1, then the historical power flow constraint state data X j is 0; g ( u g,t , s g,t , d g,t , p g,t ) If for each violation state corresponding to each power flow constraint in the j th training sample is 0, then the historical power flow constraint state data X j is 0; g ( u g,t , s g,t , d g,t , p g,t ) = 0; Among them, u g,t , s g,t , d g,t , p g,t respectively define the state variables, start-up variables, shutdown variables and output levels of the unit g at time t . g ( . ) is the line power flow constraint of the UC problem.

4. The safety-constrained unit commitment model optimization method for a power system according to claim 1, wherein, S3 includes: Input the model parameter features of the current security-constrained unit commitment model to be optimized corresponding to the power system into the target graph neural network to output the effective probability of each current power flow constraint; Regard the current power flow constraint whose effective probability exceeds a preset threshold as the current invalid power flow constraint.

5. The safety-constrained unit commitment model optimization method for the power system according to claim 4, characterized in that The loss function for training is: ； Where S is the training set, X is its index, and P j is the effective probability of the j th current power flow constraint, J 0 , J 1 are the sets of current power flow constraints with the out-of-limit state values of 0 and 1 respectively, and a0 and a1 are the imbalance learning parameters.

6. A method for solving a security-constrained unit commitment model of a power system, characterized in that Comprising: Execute the method for optimizing a security-constrained unit commitment model of a power system according to any one of claims 1-5; Solve the target security-constrained unit commitment model to obtain an optimal scheduling solution.

7. A dispatching method for a power system, characterized in that Comprising: Execute the method for solving a security-constrained unit commitment model of a power system according to claim 6; Use the optimal scheduling solution to perform power output scheduling of the power system.

8. A safety-constrained unit commitment model optimization device for a power system, characterized in that Comprising: A determination module for determining a training sample set from the historical power flow constraint status data corresponding to multiple security-constrained unit commitment models of a power system. Each training sample includes model parameter features and their corresponding power flow violation data. Wherein, the security-constrained unit commitment model is used to formulate a multi-period unit on / off schedule with the lowest total operating cost as the optimization goal under the condition of satisfying the security constraints of the power system; A training module for using the bipartite graph corresponding to the model parameter features as the input and the power flow violation data corresponding to the model parameter features as the output to train an initial graph neural network to obtain a target graph neural network; A prediction module for inputting the model parameter features of the current security-constrained unit commitment model to be optimized corresponding to the power system into the target graph neural network to predict the current invalid power flow constraint; A removal module for removing the current invalid power flow constraint from all the constraints of the current security-constrained unit commitment model to obtain a target security-constrained unit commitment model.

9. A power system, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images