Machine learning-based redundancy constraint screening method and system

Abstract

The invention discloses a redundant constraint screening method and system based on machine learning, and the method comprises the steps: obtaining constraint and historical boundary data of a power system, constructing an original model of a security constraint unit combination and a training database of the original model, carrying out association and dense constraint transformation on line power flow and unit start and stop in the constraints, and carrying out training through a machine learning model, so as to identify and eliminate redundant constraints and invalid constraints which have no influence on a result, and obtain an optimization model of the security constraint unit combination; and iteratively solving the optimization model until the optimization model has a solution, thereby obtaining an optimization result of the security constraint unit commitment. Therefore, according to the embodiment of the invention, redundant constraints and invalid constraints which have no influence on the result can be identified and eliminated, and the calculation efficiency and accuracy of the security constraint unit commitment model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and system for screening redundant constraints based on machine learning. Background Art

[0002] As the core link of power generation plan preparation, safety constrained unit commitment (SCUC) is of great significance in the optimal dispatch of power system. The problem of safety constrained unit commitment of power system needs to consider a variety of safety constraints, such as power balance, branch power flow, spinning reserve, etc. Due to the high complexity of the problem, involving many units, constraints and operation scenarios, and usually taking conservative design strategies into account system safety, it is necessary to add additional constraints to reduce the scale of the problem or simplify the problem structure, resulting in a large number of redundant constraints in the safety constrained unit commitment, which brings great difficulties to the problem solving and seriously affects the solution efficiency. At the same time, the solution of the MILP (mixed integer linear programming) model established by the SCUC problem requires repeated solution of the relaxed LP (linear programming) problem. The constraint scale will affect its solution efficiency. Redundant constraints in the model may lead to a decrease in computational efficiency, which in turn affects the solution process of the original MILP model. Therefore, it is of great significance to identify and identify redundant constraints before performing model solution calculations to simplify the problem statement, which is of great significance for saving computing resources and improving computing efficiency.

[0003] Since line flow constraints account for the highest proportion of power system constraints, the research object is usually mainly line flow constraints. At present, the iterative method based on mathematical models and the redundant constraint identification method of safety constraint unit combination based on auxiliary optimization need to construct a series of MILP sub-problems and solve the sub-problems. Although such methods can accurately identify redundant constraints, it still takes a lot of time to solve a large number of small-scale MILP problems. And the efficiency of the iterative method will be further affected as the scale of the subsequent new effective constraints increases. For large-scale systems, it may be slow to solve the optimization problem in each iteration, and multiple iterations are required to satisfy all constraints, resulting in the overall solution process taking longer than the time required to solve the original SCUC problem; and the redundant constraint identification method of safety constraint unit combination based on boundary judgment and historical experience can save computing time, but because such methods rely too much on the experience of dispatchers, they may delete the effective constraints due to manual misestimation. In addition, after the redundant constraint identification is completed by such methods, a large part of the remaining constraints are still ineffective constraints, so the constraint set is still huge and needs to be further simplified.

[0004] Therefore, there is an urgent need for a method that can improve the efficiency and accuracy of redundant constraint identification for safety-constrained unit combinations. Summary of the invention

[0005] The present invention provides a redundant constraint screening method and system based on machine learning, which associates line flow constraints with existing unit start-stop constraints and performs dense constraint transformation, identifies and eliminates redundant constraints and invalid constraints that have no effect on the results, and improves the calculation efficiency and accuracy of the safety constraint unit combination model.

[0006] In order to achieve the above objectives, in a first aspect, an embodiment of the present invention provides a method for screening redundant constraints based on machine learning, comprising: Obtaining power system constraints and historical boundary data of time T, and constructing an original model of safety-constrained unit commitment; and constructing a training database of the original model based on the historical boundary data; Associating the line flow and the start and stop of the unit in the constraint and transforming them into dense constraints, so as to obtain the sub-problems of the original model; solving the sub-problems, so as to obtain the solution set of the sub-problems; Inputting the solution set and the training database into a machine learning model for training to obtain new boundary data and a relationship between load demand and offset of the power system; According to the relationship between the load demand and the offset of the power system, the new boundary data is input into the trained machine learning model to obtain the unit start-stop prediction and the line flow constraint set that works under the new boundary data, and the line flow constraints that do not work are deleted to obtain the optimization model of the safety constraint unit combination; Solving the optimization model, if there is no solution, calculating the power flow of all lines, adding the line power flow constraint corresponding to the power flow exceeding the limit to the optimization model, and continuing to iteratively solve the optimization model until the optimization model has a solution, thereby obtaining the optimization result of the safety constrained unit commitment; The constraints include power balance, system reserve, start and stop of the unit, unit ramp, unit output, and line flow; The historical boundary data include tie line load, upper and lower limits of unit standby output, unit ramp rate, unit planned output plan, minimum start and stop time of unit, upper and lower limits of unit output, and upper and lower limits of line flow; The offset refers to the degree of offset between the start-stop result of each unit in the solution set and the optimal solution of the original model.

[0007] In a second aspect, an embodiment of the present invention provides a redundant constraint screening system based on machine learning, comprising: A data acquisition and construction module is used to acquire the constraints of the power system and the historical boundary data of time T, and to construct an original model of the safety constraint unit commitment; and to construct a training database of the original model based on the historical boundary data; A constraint association solving module is used to associate the line flow and the start and stop of the unit in the constraint and to transform the dense constraint to obtain the sub-problem of the original model; solve the sub-problem to obtain the solution set of the sub-problem; A model learning training module, used for inputting the solution set and the training database into a machine learning model for training, and obtaining new boundary data and the relationship between the load demand and the offset of the power system; A redundant constraint deletion module is used to input the new boundary data into the trained machine learning model according to the relationship between the load demand and the offset of the power system, obtain the unit start-stop prediction and the line flow constraint set that works under the new boundary data, delete the ineffective line flow constraints, and obtain the optimization model of the safety constraint unit commitment; A model iterative solution module is used to solve the optimization model. If there is no solution, the power flow of all lines is calculated, and after the line power flow constraint corresponding to the power flow exceeding the limit is added to the optimization model, the optimization model is continued to be iteratively solved until the optimization model has a solution, thereby obtaining the optimization result of the safety constrained unit commitment; The constraints include power balance, system reserve, start and stop of the unit, unit ramp, unit output, and line flow; The historical boundary data include tie line load, upper and lower limits of unit standby output, unit ramp rate, unit planned output plan, minimum start and stop time of unit, upper and lower limits of unit output, and upper and lower limits of line flow; The offset refers to the degree of offset between the start-stop result of each unit in the solution set and the optimal solution of the original model.

[0008] Compared with the prior art, the present invention discloses a method and system for screening redundant constraints based on machine learning, which constructs an original model of safe constraint unit combination by acquiring the constraints of the power system and the historical boundary data of time T; constructs a training database of the original model according to the historical boundary data; associates the line flow in the constraints with the start and stop of the units and performs dense constraint transformation to obtain the sub-problems of the original model; solves the sub-problems to obtain the solution set of the sub-problems; inputs the solution set and the training database into the machine learning model for training, and obtains the new boundary data and the load demand of the power system. The relationship between the load demand and the offset is obtained; according to the relationship between the load demand and the offset of the power system, the new boundary data is input into the trained machine learning model to obtain the prediction of the start and stop of the unit and the set of line flow constraints that work under the new boundary data, and the line flow constraints that do not work are deleted to obtain the optimization model of the safety constraint unit combination; the optimization model is solved, and if there is no solution, the flow of all lines is calculated, and the line flow constraints corresponding to the flow exceeding the limit are added to the optimization model, and the optimization model is continued to be iteratively solved until the optimization model has a solution, thereby obtaining the optimization result of the safety constraint unit combination. Therefore, the embodiment of the present invention associates the line flow with the start and stop variables of the existing units for dense constraint transformation, identifies and eliminates redundant constraints and invalid constraints that have no effect on the results, and improves the calculation efficiency and accuracy of the safety constraint unit combination model, and introduces machine learning technology, through redundant variables and constraint identification based on matrix operations, the system characteristics are learned, so as to quickly identify the legality of constraints and decision variables and the redundancy of dense constraints, thereby improving the solution efficiency of the mathematical programming model of the safety constraint unit combination. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 It is a flowchart of a method for screening redundant constraints based on machine learning provided by an embodiment of the present invention; Figure 2 It is a structural diagram of a redundant constraint screening system based on machine learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0011] It should be noted that the terms "comprises" and "specifically" and any variations of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0012] Existing redundant constraint identification and screening methods often only focus on line flow constraints, and design and analyze redundant constraint methods based on flow constraint characteristics, which are not reusable for other types of constraints. The data-driven redundant constraint screening designed in this application focuses on historical data characteristics rather than modeling and power grid topology characteristics, so it has a wider range of applications and can be extended to all redundant constraint screening with dense mathematical matrix expressions except flow constraints.

[0013] Different from the traditional redundant constraint screening based on historical experience, this application introduces a machine learning model to judge the redundancy of constraints based on the characteristics of the constraint matrix itself. Compared with the boundary data judgment method from the perspective of business understanding, it has stronger confidence and comprehensiveness.

[0014] By introducing machine learning technology, on the one hand, the information value of historical data is retained and the extensiveness of the judgment method based on experience and boundary data is avoided. On the other hand, by repeatedly iteratively solving alternative sub-problems, the efficiency of redundant constraint screening is improved compared to the iterative method, providing a higher guarantee for the efficiency and quality improvement of the subsequent optimization solution of unit combination problems.

[0015] See also Figure 1 , Figure 1 1 is a flow chart of a method for screening redundant constraints based on machine learning provided by an embodiment of the present invention. The method for screening redundant constraints based on machine learning comprises steps S1 to S5: S1, obtaining the constraints of the power system and the historical boundary data of time T, and constructing an original model of the safety constraint unit combination; and constructing a training database of the original model based on the historical boundary data; S2, associating the line flow and the unit start and stop in the constraint and transforming them into dense constraints to obtain the sub-problems of the original model; solving the sub-problems to obtain the solution set of the sub-problems; S3, inputting the solution set and the training database into a machine learning model for training, and obtaining new boundary data and a relationship between the load demand and the offset of the power system; S4, according to the relationship between the load demand and the offset of the power system, the new boundary data is input into the trained machine learning model to obtain the unit start-stop prediction and the line flow constraint set that works under the new boundary data, delete the ineffective line flow constraints, and obtain the optimization model of the safety constraint unit combination; S5, solving the optimization model. If there is no solution, the power flow of all lines is calculated, and the line power flow constraints corresponding to the power flow exceeding the limit are added to the optimization model. Then, the optimization model is solved iteratively until the optimization model has a solution, thereby obtaining the optimization result of the safety constrained unit combination.

[0016] The constraints include power balance, system reserve, start and stop of the unit, unit ramp, unit output, and line flow; The historical boundary data include tie line load, upper and lower limits of unit standby output, unit ramp rate, unit planned output plan, minimum start and stop time of unit, upper and lower limits of unit output, and upper and lower limits of line flow; The offset refers to the degree of offset between the start-stop result of each unit in the solution set and the optimal solution of the original model.

[0017] For example, the constraints of the power system and the historical boundary data of time T are obtained, and according to the original model of the safety constraint unit combination, data is collected from the historical boundary data. For the test data of the same day, the topology structure, the fixed output plan of the unit (such as the upper and lower limits of the unit standby output, the unit climbing rate, the unit planned output plan, etc.), the minimum start and stop time of the unit, the upper and lower limits of the output, the upper and lower limits of the line flow, and other historical boundary data are clarified, and the optimal unit start and stop plan of the original model is accurately calculated by adjusting the load, thereby generating multiple sets of training data to form a training database of the original model. The line flow in the constraint is associated with the start and stop of the unit, and the start and stop of the unit is used to determine whether the line flow exceeds the limit. The start and stop of the unit is represented by a 0-1 variable, so as to control whether it is effective by using a 0-1 variable, and the 0-1 variable is used to predict the 0-1 variable, and the effective line flow constraint set in the original model is obtained, and the ineffective line flow constraints are deleted to obtain the optimization model of the safety constraint unit combination, so as to identify and eliminate redundant constraints and invalid constraints that have no effect on the results.

[0018] It should be noted that the original model of safety-constrained unit commitment takes the minimization of operating cost as the objective function, and clarifies the network topology structure, number of units and capacity of the power system. The T time can be set as needed, for example, , wait.

[0019] In a preferred embodiment, the step S2 specifically includes: S21, associating the line power flow in the constraint with the start and stop of the unit, and converting the line power flow constraint into an upper semi-continuous constraint and a lower semi-continuous constraint of the line power flow by using the start and stop variables of the unit; S22, if the line power flow constraint is established, the upper and lower limits of the line power flow are close to the upper and lower limits of the semi-continuous constraint, and a sub-problem of the original model is obtained; S23, adopting a 0-1 variable control method to control the 0-1 variables so that the line power flow constraint takes effect, and obtaining a solution set of the sub-problem by solving the value of the 0-1 variable for starting and stopping the unit.

[0020] In a preferred embodiment, the step S21 specifically includes: S211, obtaining the upper and lower limit constraints of the unit output and the line flow constraints in the original model, and associating the line flow in the constraints with the start and stop of the unit; S212, by using the upper and lower limit constraints of the unit output and the unit start and stop variables, the line power flow constraint is converted into an upper semi-continuous constraint and a lower semi-continuous constraint of the line power flow.

[0021] For example, the expression of the upper and lower limit constraints of the unit output is: , The expression of the line power flow constraint is: , The expressions of the upper semi-continuous constraint and the lower semi-continuous constraint of the line power flow are:

[0022]

[0023] In the formula, and For the crew The upper and lower limits of output; For the unit exist Decision variables for time period output; For the unit exist The decision variables of the on / off state of the time period are When exist The unit is in shutdown state during the period. exist The output of the time period is 0. When exist The unit is in the on state during the period. exist The output during the period is greater than 0; For line The transmission limit of the line power flow; and The units are , load node To line The generator output power transfer distribution factor; For the load node exist Bus load during the time period; For the unit The total number of For the load node The total quantity.

[0024] Specifically, the objective function of the sub-problem is to minimize the sum of the distances between the upper and lower limits of the unit output constraint and the upper semi-continuous constraint and the lower semi-continuous constraint of the line power flow.

[0025] It should be noted that the constraints of the sub-problem only include the constraints related to the 0-1 variables in the original model of the safety constraint unit combination; at the same time, the sub-problem objective is added to the original model as a penalty item to ensure that the solution of the sub-problem can be as close as possible to the solution of the original model.

[0026] For example, the objective function of the sub-problem is expressed as:

[0027] In the formula, and For the crew The upper and lower limits of output; For the unit exist Decision variables for time period output; For the unit exist The decision variables of the on / off state of the time period are When exist The unit is in shutdown state during the period. exist The output of the time period is 0. When exist The unit is in the on state during the period. exist The output during the period is greater than 0; For line The transmission limit of the line power flow; and The units are , load node To line The generator output power transfer distribution factor; For the load node exist Bus load during the time period; For the load node The total quantity.

[0028] In a preferred embodiment, the step S3 specifically includes: S31, inputting the solution set and the training database into a machine learning model for training, and comparing the offset with the maximum offset of the right-hand term in the minimum continuous start-stop time constraint; S32: If the offset is less than the maximum offset of the right-hand term, the minimum start / stop time constraint is a redundant constraint. Based on the association between the start / stop of the unit and the line flow, new boundary data and the relationship between the load demand and the offset of the power system are obtained.

[0029] It should be noted that the relationship between load demand and offset is learned by machine learning methods using methods such as neural network, KNN (k nearest neighbor method), SVM (support vector machine), and logistic regression, with the solution set of the above sub-problems and training data as input. Since the start and stop of the unit is a 0-1 variable, it can be understood that if the optimization result of the sub-problem of a single unit is 1 and the optimization result in the original model is 0, the offset is 1, and the offset of the entire model is the sum of the offsets of all units. With the help of this offset and the maximum offset of the right-hand term in the minimum continuous start and stop time constraint, it is judged whether the minimum continuous start and stop time constraint is redundant. If the predicted offset is less than the maximum offset of the right-hand term, the minimum start and stop time constraint is a redundant constraint. At the same time, synchronous control is achieved by associating the start and stop of the unit with the line flow. Based on the training data, the hyperparameter tuning of the machine learning model and the historical data learning are performed to obtain the line blocking state, the flow over-limit and reaching the limit, the correlation between the sub-problem offset and the power generation and load power, and the start and stop of the unit. Taking KNN as an example, the line load, minimum start and stop time and other data are input as features, the distance between each sample point in the training sample and the test sample is calculated, all distance values ​​are sorted, and the first k samples with the smallest distance are selected. Voting and averaging are performed based on their sample coordinates to determine the offset of the test sample.

[0030] For example, the minimum continuous start-stop time constraint is: , , , , In the formula, and For each unit exist Time period and Decision variables for the on / off status of the time period; and They are the minimum continuous shutdown and startup time of the unit respectively; and For each unit exist The shutdown and startup time of the time period; For the crew exist The decision variables for the on / off state of the time period.

[0031] In a preferred embodiment, the step S4 specifically includes: S41, inputting the new boundary data into a trained machine learning model according to the relationship between the load demand and the offset of the power system to obtain a 0-1 variable prediction of the start and stop of the unit; S42, based on the association between the start and stop of the unit and the line flow, obtaining a set of line flow constraints that are effective under the new boundary data; and marking the line flow constraints that are ineffective under the new boundary data as redundant constraints; S43, according to the identification, the original model is reduced to remove the redundant constraints to obtain the optimization model of the safety-constrained unit commitment.

[0032] In a preferred embodiment, the step S5 specifically includes: S51, solving the optimization model, if the optimization model has no solution, calculating the power flow of all lines of the power system according to the minimum infeasible conflict set of the original model; S52, if the power flow exceeds the limit, adding the line power flow constraint corresponding to the power flow exceeding the limit to the optimization model, S53, returning to repeatedly execute S51-S52 until the optimization model has a solution, thereby obtaining the optimization result of the safety constrained unit commitment.

[0033] Figure 2 1 is a schematic diagram of a structure of a redundant constraint screening system based on machine learning provided by an embodiment of the present invention. The redundant constraint screening system based on machine learning includes: The data acquisition and construction module 21 is used to acquire the constraints of the power system and the historical boundary data of time T, and construct the original model of the safety constraint unit commitment; and construct the training database of the original model according to the historical boundary data; The constraint association solving module 22 is used to associate the line flow and the unit start and stop in the constraint and perform dense constraint transformation to obtain the sub-problem of the original model; solve the sub-problem to obtain the solution set of the sub-problem; A model learning training module 23, used for inputting the solution set and the training database into a machine learning model for training, and obtaining new boundary data and the relationship between the load demand and the offset of the power system; A redundant constraint deletion module 24 is used to input the new boundary data into the trained machine learning model according to the relationship between the load demand and the offset of the power system, obtain the unit start-stop prediction and the line flow constraint set that works under the new boundary data, delete the ineffective line flow constraints, and obtain the optimization model of the safety constraint unit commitment; The model iteration solution module 25 is used to solve the optimization model. If there is no solution, the power flow of all lines is calculated, and the line power flow constraint corresponding to the power flow exceeding the limit is added to the optimization model, and the optimization model is continued to be iteratively solved until the optimization model has a solution, thereby obtaining the optimization result of the safety constrained unit commitment; The constraints include power balance, system reserve, start and stop of the unit, unit ramp, unit output, and line flow; The historical boundary data include tie line load, upper and lower limits of unit standby output, unit ramp rate, unit planned output plan, minimum start and stop time of unit, upper and lower limits of unit output, and upper and lower limits of line flow; The offset refers to the degree of offset between the start-stop result of each unit in the solution set and the optimal solution of the original model.

[0034] A redundant constraint screening system based on machine learning provided in an embodiment of the present invention can implement all processes of the redundant constraint screening method based on machine learning in the above-mentioned embodiment. The functions of each module in the system and the technical effects achieved are respectively the same as the functions and technical effects achieved by the redundant constraint screening method based on machine learning in the above-mentioned embodiment, and will not be repeated here.

[0035] In summary, the present invention discloses a method and system for screening redundant constraints based on machine learning, which constructs an original model of a safety-constrained unit combination by acquiring the constraints of the power system and the historical boundary data of time T; constructs a training database of the original model according to the historical boundary data; associates the line flow in the constraints with the start and stop of the unit and transforms them into dense constraints to obtain a sub-problem of the original model; solves the sub-problem to obtain a solution set of the sub-problem; inputs the solution set and the training database into the machine learning model for training to obtain new boundary data and the relationship between the load demand and the offset of the power system; according to the relationship between the load demand and the offset of the power system, inputs the new boundary data into the trained machine learning model to obtain the unit start and stop prediction and the line flow constraint set that works under the new boundary data, deletes the line flow constraints that do not work, and obtains the optimization model of the safety-constrained unit combination; solves the optimization model, and if there is no solution, calculates the flow of all lines, adds the line flow constraint corresponding to the flow exceeding the limit to the optimization model, and continues to iteratively solve the optimization model until the optimization model has a solution, thereby obtaining the optimization result of the safety-constrained unit combination. Therefore, the embodiment of the present invention associates the line flow with the start-stop variables of the existing units to perform dense constraint transformation, uses 0-1 variables to control whether they are effective, and uses machine learning methods to predict the 0-1 variables, thereby identifying and eliminating redundant constraints and invalid constraints that have no effect on the results, thereby improving the calculation efficiency and accuracy of the safety constraint unit combination model, and by introducing machine learning technology, through redundant variables and constraint identification based on matrix operations, the system characteristics are learned, thereby quickly identifying the legitimacy of constraints and decision variables, and the redundancy of dense constraints, thereby improving the solution efficiency of the safety constraint unit combination mathematical programming model.

[0036] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for screening redundant constraints based on machine learning, characterized in that: include: Obtaining power system constraints and historical boundary data of time T, and constructing an original model of safety-constrained unit commitment; and constructing a training database of the original model based on the historical boundary data; The line power flow in the constraint is associated with the start and stop of the unit and transformed into a dense constraint to obtain a sub-problem of the original model; Solving the sub-problems to obtain a solution set of the sub-problems; Inputting the solution set and the training database into a machine learning model for training to obtain new boundary data and a relationship between load demand and offset of the power system; According to the relationship between the load demand and the offset of the power system, the new boundary data is input into the trained machine learning model to obtain the unit start-stop prediction and the line flow constraint set that works under the new boundary data, and the line flow constraints that do not work are deleted to obtain the optimization model of the safety constraint unit combination; Solving the optimization model, if there is no solution, calculating the power flow of all lines, adding the line power flow constraint corresponding to the power flow exceeding the limit to the optimization model, and continuing to iteratively solve the optimization model until the optimization model has a solution, thereby obtaining the optimization result of the safety constrained unit commitment; The constraints include power balance, system reserve, start and stop of the unit, unit ramp, unit output, and line flow; The historical boundary data include tie line load, upper and lower limits of unit standby output, unit ramp rate, unit planned output plan, minimum unit start and stop time, upper and lower limits of unit output, and upper and lower limits of line flow; The offset refers to the degree of offset between the start-stop result of each unit in the solution set and the optimal solution of the original model.

2. The method for screening redundant constraints based on machine learning as claimed in claim 1, characterized in that: The line flow in the constraint is associated with the start and stop of the unit and transformed into a dense constraint to obtain a sub-problem of the original model; Solve the sub-problems to obtain a solution set of the sub-problems, which specifically includes: Associating the line power flow in the constraint with the start and stop of the unit, and converting the line power flow constraint into an upper semi-continuous constraint and a lower semi-continuous constraint of the line power flow by using the start and stop variables of the unit; If the line power flow constraint is established, the upper and lower limits of the line power flow are close to the upper and lower limits of the semi-continuous constraint, and a sub-problem of the original model is obtained; A 0-1 variable control method is adopted to control the 0-1 variables so that the line power flow constraint takes effect, and a solution set of the sub-problem is obtained by solving the values ​​of the 0-1 variables for starting and stopping the unit.

3. The method for screening redundant constraints based on machine learning as claimed in claim 2, characterized in that: The step of associating the line flow in the constraint with the start and stop of the unit, and converting the line flow constraint into an upper semi-continuous constraint and a lower semi-continuous constraint of the line flow by using the start and stop variables of the unit, specifically includes: Obtaining the upper and lower limit constraints of unit output and the line flow constraints in the original model, and associating the line flow in the constraints with the start and stop of the unit; By means of the upper and lower limit constraints of the unit output and using the unit start and stop variables, the line power flow constraint is converted into an upper semi-continuous constraint and a lower semi-continuous constraint of the line power flow.

4. The method for screening redundant constraints based on machine learning as claimed in claim 1, characterized in that: The objective function of the sub-problem is to minimize the sum of the distances between the upper and lower limits of the unit output constraint and the upper semi-continuous constraint and the lower semi-continuous constraint of the line power flow.

5. The method for screening redundant constraints based on machine learning as claimed in claim 1, characterized in that: The step of inputting the solution set and the training database into a machine learning model for training to obtain new boundary data and a relationship between the load demand and the offset of the power system specifically includes: Inputting the solution set and the training database into a machine learning model for training, and comparing the offset with the maximum offset of the right-hand term in the minimum continuous start-stop time constraint; If the offset is less than the maximum offset of the right-hand term, the minimum start / stop time constraint is a redundant constraint, and new boundary data and the relationship between the load demand and the offset of the power system are obtained based on the association between the start / stop of the unit and the line flow.

6. The method for screening redundant constraints based on machine learning as claimed in claim 1, characterized in that: According to the relationship between the load demand and the offset of the power system, the new boundary data is input into the trained machine learning model to obtain the unit start-stop prediction and the line flow constraint set that works under the new boundary data, delete the line flow constraints that do not work, and obtain the optimization model of the safety constraint unit combination, which specifically includes: According to the relationship between the load demand and the offset of the power system, the new boundary data is input into the trained machine learning model to obtain a 0-1 variable prediction of the start and stop of the unit; Based on the association between the start and stop of the unit and the line flow, a set of line flow constraints that are effective under the new boundary data is obtained; and the line flow constraints that are ineffective under the new boundary data are marked as redundant constraints; According to the identification, the original model is pared down to eliminate the redundant constraints, thereby obtaining an optimized model of the safety-constrained unit combination.

7. The method for screening redundant constraints based on machine learning as claimed in claim 1, characterized in that: The step of solving the optimization model, if there is no solution, calculating the power flow of all lines, adding the line power flow constraint corresponding to the power flow exceeding the limit to the optimization model, and continuing to iteratively solve the optimization model until the optimization model has a solution, thereby obtaining the optimization result of the safety constrained unit commitment, specifically includes: S51, solving the optimization model, if the optimization model has no solution, calculating the power flow of all lines of the power system according to the minimum infeasible conflict set of the original model; S52, if the power flow exceeds the limit, adding the line power flow constraint corresponding to the power flow exceeding the limit to the optimization model, S53, returning to repeatedly execute S51-S52 until the optimization model has a solution, thereby obtaining the optimization result of the safety constrained unit commitment.

8. A redundant constraint screening system based on machine learning, characterized in that: include: A data acquisition and construction module is used to acquire the constraints of the power system and the historical boundary data of time T, and to construct an original model of the safety constraint unit commitment; and to construct a training database of the original model based on the historical boundary data; A constraint association solving module is used to associate the line power flow and the unit start and stop in the constraint and perform dense constraint transformation to obtain the sub-problem of the original model; Solving the sub-problems to obtain a solution set of the sub-problems; A model learning training module, used for inputting the solution set and the training database into a machine learning model for training, and obtaining new boundary data and the relationship between the load demand and the offset of the power system; A redundant constraint deletion module is used to input the new boundary data into the trained machine learning model according to the relationship between the load demand and the offset of the power system, obtain the unit start-stop prediction and the line flow constraint set that works under the new boundary data, delete the ineffective line flow constraints, and obtain the optimization model of the safety constraint unit commitment; A model iterative solution module is used to solve the optimization model. If there is no solution, the power flow of all lines is calculated, and after the line power flow constraint corresponding to the power flow exceeding the limit is added to the optimization model, the optimization model is continued to be iteratively solved until the optimization model has a solution, thereby obtaining the optimization result of the safety constrained unit commitment; The constraints include power balance, system reserve, start and stop of the unit, unit ramp, unit output, and line flow; The historical boundary data include tie line load, upper and lower limits of unit standby output, unit ramp rate, unit planned output plan, minimum start and stop time of unit, upper and lower limits of unit output, and upper and lower limits of line flow; The offset refers to the degree of offset between the start-stop result of each unit in the solution set and the optimal solution of the original model.

Images