Data mechanism joint driving method and system for direct current optimal power flow solution

Abstract

The application provides a data mechanism joint driving method and system for direct current optimal power flow solution, and is applied to the technical field of electric power data processing. Electric power demand is input into a DNN to obtain an initial solution. The initial solution does not satisfy problem constraints, and a variable elimination method is used to process the problem to be solved and the initial solution, so that the problem is converted into an optimization problem without equation constraints and a corresponding initial solution. Based on a constraint set including only inequality constraints and a normal vector violating the inequality constraints, the position of the initial solution is iteratively adjusted until the infeasible solution of the initial solution falls into a feasible region; an intermediate feasible solution satisfying the problem constraints is obtained by restoring the initial solution. The intermediate feasible solution is input into a FONET neural network to obtain adjustment information, and the intermediate feasible solution is adjusted into a target feasible solution based on the adjustment information. The application is suitable for a DCOPF problem with a feasible solution under any electric power topology, can simultaneously guarantee the solving efficiency and the feasibility of the solving result, and the solving result has higher optimality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power data processing, in particular to a data mechanism joint driving method and system for direct current optimal power flow solution. BACKGROUND

[0002] Direct current optimal power flow (DCOPF) is a simplified model in power system optimization, which is used to solve the most economic unit output scheme without violating the problem constraints at a lower computational cost. The solution of the DCOPF problem must satisfy the equality constraints and inequality constraints.

[0003] The neural network method is one of the general solution methods for the direct current optimal power flow problem, but for the neural network for solving the direct current optimal power flow problem, it lacks the power grid topology generalization ability, that is, the neural network can output the DCOPF solution with high optimality based on the given power grid topology, but after the change of the power grid topology corresponding to the target DCOPF problem, the optimality and feasibility probability of the solution given by the neural network will gradually decrease with the deepening of the topology change degree.

[0004] Therefore, in the prior art, there is a problem that the solving efficiency of the DCOPF problem of the power system is low and the feasibility of the solving result is low when the power grid topology changes. SUMMARY

[0005] Therefore, the present application provides a data mechanism joint driving method and system for direct current optimal power flow solution, which is suitable for the DCOPF problem with feasible solution existing under any power topology, can guarantee the solving efficiency and the feasibility of the solving result, and the obtained solving result has higher optimality.

[0006] To solve the above problems, the technical scheme provided by the present application is as follows:

[0007] In a first aspect, the present application provides a data mechanism joint driving method for direct current optimal power flow solution, which comprises:

[0008] inputting the power demand into a deep neural network DNN to obtain a unit output scheme output by the deep neural network DNN, any one of the unit output schemes in the unit processing scheme being an initial solution;

[0009] if the initial solution satisfies the problem constraints, taking the initial solution as an intermediate feasible solution, wherein the problem constraints include equality constraints and inequality constraints;

[0010] If the initial solution does not satisfy the problem constraints, the constraints and initial solution of the problem to be solved are processed based on the equality constraints by the variable elimination method to obtain the transformed constraint set and initial solution. The constraint set consists only of inequality constraints.

[0011] Based on the constraint set and the normal vector that violates the inequality constraint, the position of the initial solution is iteratively adjusted until the infeasible solution of the initial solution falls into the feasible region;

[0012] The initial solution is restored based on the equality constraints and the infeasible solution of the initial solution to obtain an intermediate feasible solution that satisfies the equality constraints and the inequality constraints.

[0013] The intermediate feasible solution is input into the pre-constructed neural network FONET, which guarantees feasibility and is embedded in optimality, to determine the adjustment information of the intermediate feasible solution;

[0014] The intermediate feasible solution is adjusted based on the adjustment information to obtain the target feasible solution.

[0015] In one possible implementation, iteratively adjusting the position of the initial solution based on the constraint set and the normal vector violating the inequality constraints until an infeasible solution of the initial solution falls into the feasible region includes:

[0016] Calculate the inequality constraints violated by the infeasible solutions of the initial solution based on the constraint set;

[0017] Calculate the normal vector that violates the inequality constraint, and use the sum of the normal vectors as the direction of movement for the infeasible solution of the initial solution;

[0018] The distance between the infeasible solution of the initial solution and the initially violated inequality constraint is taken as the single iteration movement distance of the infeasible solution of the initial solution, and the initially violated inequality constraint is the first violated inequality constraint encountered in the movement direction.

[0019] Calculate the intersection plane of the hyperplane where the infeasible solutions of the initial solution are located after adjustment according to the moving direction and moving distance;

[0020] The movement direction is projected onto the intersection plane to generate a new movement direction;

[0021] Until the intersection plane or the new direction of movement satisfies a preset condition, it is determined that the infeasible solution of the initial solution falls into the feasible region. The preset condition is that the intersection plane is a point or the new direction of movement is a vector of all zeros.

[0022] In one possible implementation, the adjustment information includes adjustment direction and adjustment distance. The step of inputting the intermediate feasible solution into a pre-constructed neural network FONET, which guarantees feasibility and embeds optimality, to determine the adjustment information of the intermediate feasible solution includes:

[0023] The intermediate feasible solution is input into the FONET to obtain a first activation function, which is used to determine the adjustment direction.

[0024] The backup adjustment distance is determined based on the adjustment direction and the first preset formula;

[0025] Calculate the main adjustment distance according to the second preset formula;

[0026] The adjustment distance is determined from the primary adjustment distance and the backup adjustment distance.

[0027] In one possible implementation, the FONET includes an optimal neural network OAERN that guarantees the satisfaction of equality constraints and an optimal neural network OAERN that guarantees the satisfaction of inequality constraints; the OAERN is used to determine the adjustment direction and the OAERN is used to determine the adjustment distance.

[0028] Secondly, embodiments of this application provide a data mechanism-driven joint system for solving DC optimal power flow problems, the system comprising:

[0029] The acquisition module is used to input the power demand into the deep neural network (DNN) to obtain the unit output scheme output by the deep neural network (DNN), and any one of the unit output schemes in the unit processing scheme is the initial solution;

[0030] The judgment module is used to determine if the initial solution satisfies the problem constraints, and then use the initial solution as an intermediate feasible solution, wherein the problem constraints include equality constraints and inequality constraints.

[0031] The judgment module is also used to process the constraints and initial solution of the problem to be solved by the variable elimination method based on the equality constraints if the initial solution does not satisfy the problem constraints, so as to obtain the transformed constraint set and initial solution. The constraint set consists only of inequality constraints.

[0032] An adjustment module is used to iteratively adjust the position of the initial solution based on the constraint set and the normal vector that violates the inequality constraint, until the infeasible solution of the initial solution falls into the feasible region;

[0033] The restoration module is used to restore the initial solution based on the equality constraints and the infeasible solutions of the initial solution to obtain an intermediate feasible solution that satisfies the equality constraints and the inequality constraints.

[0034] The determination module is used to input the intermediate feasible solution into the pre-constructed neural network FONET, which guarantees feasibility and is embedded in optimality, and determine the adjustment information of the intermediate feasible solution;

[0035] The adjustment module is further configured to adjust the intermediate feasible solution based on the adjustment information to obtain the target feasible solution.

[0036] In one possible implementation, the adjustment module is used to iteratively adjust the position of the initial solution based on the constraint set and the normal vectors that violate the inequality constraints, until an infeasible solution of the initial solution falls into the feasible region, including:

[0037] The calculation submodule is used to calculate the inequality constraints violated by the infeasible solution of the initial solution based on the constraint set; calculate the normal vectors that violate the inequality constraints, and use the sum of the normal vectors as the direction of movement of the infeasible solution of the initial solution;

[0038] The determination submodule is used to take the distance between the infeasible solution of the initial solution and the initially violated inequality constraint as the single iteration movement distance of the infeasible solution of the initial solution, wherein the initially violated inequality constraint is the first violated inequality constraint encountered in the movement direction.

[0039] The calculation submodule is also used to calculate the intersection plane of the hyperplane where the infeasible solution of the initial solution is located after being adjusted according to the moving direction and moving distance;

[0040] A generation submodule is used to project the movement direction onto the intersection plane to generate a new movement direction;

[0041] The determining submodule is used to determine that the infeasible solution of the initial solution falls into the feasible region until the intersection plane or the new movement direction meets a preset condition. The preset condition is that the intersection plane is a point or the new movement direction is a vector with all zeros.

[0042] In one possible implementation, the adjustment information includes the adjustment direction and the adjustment distance, and the determining module is specifically used for:

[0043] The intermediate feasible solution is input into the FONET to obtain a first activation function, which is used to determine the adjustment direction.

[0044] The backup adjustment distance is determined based on the adjustment direction and the first preset formula;

[0045] Calculate the main adjustment distance according to the second preset formula;

[0046] The adjustment distance is determined from the primary adjustment distance and the backup adjustment distance.

[0047] In one possible implementation, the FONET includes an optimal neural network OAERN that guarantees the satisfaction of equality constraints and an optimal neural network OAERN that guarantees the satisfaction of inequality constraints; the OAERN is used to determine the adjustment direction and the OAERN is used to determine the adjustment distance.

[0048] Thirdly, embodiments of this application provide a data mechanism joint driving device for solving DC optimal power flow problems, the device including: a processor, a memory, and a system bus;

[0049] The processor and the memory are connected via the system bus;

[0050] The memory is used to store one or more programs, the one or more programs including instructions, which, when executed by the processor, cause the processor to perform the data mechanism joint driving method for solving DC optimal power flow as described in the first aspect above.

[0051] Fourthly, this application provides a storage medium storing instructions that, when executed by a computer, cause the computer to implement the data mechanism joint driving method for solving DC optimal power flow as described in the first aspect above.

[0052] Therefore, this application has the following beneficial effects:

[0053] This application provides a data mechanism-driven method for solving DC optimal power flow problems, applicable to DCOPF problems with feasible solutions under any power topology. The power demand is input into a deep neural network (DNN) to obtain unit output schemes output by the DNN. Any one of these schemes serves as the initial solution. If the initial solution satisfies the problem constraints, it is used as an intermediate feasible solution. These constraints include equality and inequality constraints. If the initial solution does not satisfy the constraints, a variable elimination method is used to process the constraints and initial solution based on equality constraints, resulting in a transformed constraint set and initial solution. This transformed constraint set contains only inequality constraints. Based on the constraint set and the normal vectors violating inequality constraints, the position of the initial solution is iteratively adjusted until an infeasible solution falls into the feasible region. Finally, the initial solution is restored based on the equality constraints and the infeasible solution, yielding an intermediate feasible solution that satisfies both equality and inequality constraints. The intermediate feasible solution is input into the pre-constructed FONET neural network, which guarantees feasibility and embeds optimality, to determine the adjustment information of the intermediate feasible solution. Based on the adjustment information, the intermediate feasible solution is adjusted to obtain the target feasible solution. In this way, determining the adjustment information and adjusting the initial solution with the adjustment information can simultaneously guarantee the solution efficiency and the feasibility of the solution result, and the obtained solution result has higher optimality.

[0054] This application also provides a system corresponding to the above method, which has the same beneficial effects as the above method. Attached Figure Description

[0055] Figure 1 A flowchart illustrating a data mechanism-driven method for solving optimal DC power flow, provided in an embodiment of this application;

[0056] Figure 2 A flowchart illustrating the overall process of solving DCOPF using UDAP neural network after preprocessing DNN in an embodiment of this application.

[0057] Figure 3 A schematic diagram of the data flow of the OAERN neural network, which has optimality and can guarantee the satisfaction of equality constraints, provided for an embodiment of this application.

[0058] Figure 4 A schematic diagram of the data flow of the OAIRN neural network, which has optimality and can guarantee the satisfaction of inequality constraints, provided for an embodiment of this application.

[0059] Figure 5 This is a schematic diagram illustrating the movement effect of an infeasible solution according to an embodiment of this application;

[0060] Figure 6 A schematic diagram of a data mechanism-driven system for solving optimal DC power flow is provided in this application embodiment.

[0061] Figure 7 This is a schematic diagram of a data mechanism-driven device for solving optimal DC power flow, provided in an embodiment of this application. Detailed Implementation

[0062] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0063] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0064] In power system optimization, the DCOPF problem is used to find the most economical unit output scheme that does not violate problem constraints with low computational cost. The required data involves energy demand and power system parameters. The specific expression is:

[0065] (1)

[0066] (2)

[0067] (3)

[0068] (4)

[0069] (5)

[0070] in, For unit combination, This is the generator set output matrix. The output of the generator set is The cost of electricity generation at that time It is the first The production cost of a generator These are the model parameters for power generation costs. for The first in One value, and These are the minimum / maximum output matrices of the generator sets. It is a load demand matrix. It is a matrix describing the power flow limits of a line. It is the nodal input power-line power flow transfer function of the power system under the target topology. Given the power grid topology under the target state, It is the sum of all elements in the matrix. This indicates that the absolute values ​​of each element in the matrix are taken.

[0071] The above-mentioned constraints are the equality and inequality relationships that must be satisfied to unlock the problem, including equality constraints and inequality constraints. The objective function is an indicator for evaluating the quality of the solution. For the DCOPF problem, the objective function is generally to minimize the economic cost of the unit output scheme. Feasibility of the result indicates that the solution satisfies all the constraints of the objective problem.

[0072] Currently, in scenarios where the parameters of deep neural networks (DNNs) used to solve DCOPF problems do not update during the solution process, DNN applications aimed at achieving feasible output data mainly focus on four technical directions: learning constraints of the target problem type from historical data (Learn From History, LFH), learning constraints of the target problem type from the problem model (Learn From Model, LFM), using DNN results as a warm start for optimization to achieve speedup (Predict As Warm Start, PAWS), and using DNN to obtain preliminary results and obtaining corresponding feasible solutions through mapping (Predict And Projection, PAP).

[0073] in,

[0074] LFH drives the DNN to fully learn the mapping relationship between energy demand and feasible region boundary by controlling the internal construction of the DNN model or the preparation process of historical data. This is achieved during the process of learning the mapping law between energy demand and the optimal solution of the problem, so that the final output data satisfies the feasible region boundary, which is jointly restricted by equality constraints and inequality constraints.

[0075] LFM adds the Lagrangian function of a class of problems to the loss function, and improves the probability that the output data is feasible by directly learning the mapping relationship between the input data and the feasible region boundary.

[0076] PAWS uses the output data of DNN as the initial solution or parameters of the optimization model, and uses traditional methods to solve the simplified optimization problem, thereby ensuring the feasibility of the results.

[0077] The PAP framework uses the infeasible data output by the DNN as a starting point and guides that starting point into the feasible region through one or more constraint-based mappings.

[0078] Currently, the commonly used algorithms under the PAP framework are Gauge Mapping (GM) and DC3 (DeepConstraint Completion and Correction).

[0079] The GM algorithm can only be used for problems with a constant feasible region. It restricts the output data of the DNN to a fixed, complex, high-dimensional space. The specific steps include:

[0080] 1. Construct a sample set using simulation experiments or historical data, with electricity demand as the sample input and power system dispatch schemes as the sample labels.

[0081] 2. Construct a DNN using the tanh activation function, so that it outputs a vector with all elements in the range [-1, 1] based on the input power demand.

[0082] 3. Connect the differentiable canonical mapping GM function to the DNN. When the input is a vector with all elements in the range [-1,1], the GM function outputs a vector with the same dimension that is within the spatial range of the GM function that passes through the origin.

[0083] 4. Combine DNN and GM function to obtain a canonical mapping DNN, and train the DNN using historical data.

[0084] 5. In practical applications, the power demand is input into the DNN, which then outputs a power dispatching scheme.

[0085] The GM algorithm can only map two convex sets that pass through the same point and whose position is definite. However, the feasible region of the DCOPF problem changes with the power system topology and energy demand, making it impossible to determine a fixed point that is always within the DCOPF feasible region. Therefore, the GM algorithm cannot be directly applied to solving DCOPF problems where the feasible region changes with energy demand.

[0086] The DC3 algorithm consists of two parts: preparation and application. The preparation part includes the following specific steps:

[0087] 1. Divide the variables of the unit output scheme into a set of independent variables. and dependent variable Construct a mapping function between the independent and dependent variables using the equality constraints in the target problem. .

[0088] 2. Construct a model using electricity demand as the input sample, based on simulation experiments or historical data. The sample set is the sample label.

[0089] 3. The structure can obtain power by inputting electricity demand. DNN.

[0090] 4. Using the gradient backpropagation algorithm to enable DNN to learn electricity demand and The mapping relationship between them.

[0091] The specific steps in the application section include:

[0092] 1. By inputting electricity demand into a trained DNN, obtain .

[0093] 2. Calculation Construct a set of variables that satisfy the equality constraints. .

[0094] 3. Calculate the variable set Calculate the degree of dissatisfaction for each unsatisfied inequality constraint. and calculate right The gradient is used to update the value. until It satisfies all inequality constraints.

[0095] The DC3 algorithm cannot guarantee convergence, thus failing to guarantee its ability to obtain feasible solutions. Furthermore, step 3 in the application section involves an indefinite number of adjustments based solely on inequality constraints. Since the number of mappings and the mapping direction are only affected by the constraints and not guided by the DCOPF objective function, the DC3 algorithm struggles to ensure the optimality of the output data when the DCOPF problem varies significantly.

[0096] Furthermore, apart from the two related technologies mentioned above, other major technical solutions also fail to quickly obtain feasible solutions for DCOPF in variable power topologies. Specifically, LFH and LFM suffer from a lack of output data feasibility because the DNN used for the final output data has a black-box nature, which means they cannot guarantee that the directly output scheduling scheme satisfies all problem constraints. PAWS, on the other hand, inevitably uses optimization methods, and its solution time increases exponentially with the problem size, thus lacking efficiency.

[0097] In summary, when faced with changes in power grid topology, the initial solution obtained by using DNN has low feasibility, while the post-processing method using mapping cannot guarantee both the acquisition of feasible solutions and the computational efficiency of the acquisition process. This results in low solution efficiency and low feasibility of the DCOPF problem.

[0098] In view of this, embodiments of this application provide a data mechanism-driven method and system for solving DC optimal power flow problems, applicable to DCOPF problems with feasible solutions under any power topology. The power demand is input into a deep neural network (DNN) to obtain the unit output schemes output by the DNN. Any one of the unit output schemes is the initial solution. If the initial solution satisfies the problem constraints, it is used as an intermediate feasible solution. These constraints include equality constraints and inequality constraints. If the initial solution does not satisfy the problem constraints, the constraints and initial solution of the problem to be solved are processed based on equality constraints using a variable elimination method, resulting in a transformed constraint set and initial solution. The transformed constraint set only contains inequality constraints. Based on the constraint set and the normal vectors violating inequality constraints, the position of the initial solution is iteratively adjusted until an infeasible solution of the initial solution falls into the feasible region. The initial solution is then restored based on the equality constraints and the infeasible solution of the initial solution to obtain an intermediate feasible solution that satisfies both equality and inequality constraints. The intermediate feasible solution is input into the pre-constructed FONET neural network, which guarantees feasibility and embeds optimality, to determine the adjustment information of the intermediate feasible solution. Based on the adjustment information, the intermediate feasible solution is adjusted to obtain the target feasible solution. In this way, determining the adjustment information and adjusting the initial solution with the adjustment information can simultaneously guarantee the solution efficiency and the feasibility of the solution result, and the obtained solution result has higher optimality.

[0099] To facilitate understanding of the technical solutions provided in the embodiments of this application, the following description, in conjunction with the accompanying drawings, illustrates a data mechanism-driven method and system for solving optimal DC power flow.

[0100] It should be noted that the embodiments of this application can be applied to single-moment unit scheduling of large-scale power systems to quickly obtain scheduling schemes that meet power balance, unit output constraints, and line power flow constraints.

[0101] In this embodiment, the power demand is input into an optimal solution prediction neural network, such as a deep neural network (DNN), to obtain an initial solution for the optimal power flow output by the neural network. If the initial solution does not meet the problem constraints, an infeasibility correction algorithm is used to process the initial solution and output an intermediate feasible solution. If the initial solution meets the problem constraints, it is used as an intermediate feasible solution. Finally, the intermediate feasible solution is input into FONET in the third step to output adjustment information for the intermediate feasible solution, guiding the adjustment of the intermediate feasible solution.

[0102] The infeasibility correction algorithm first uses the variable elimination method to process the constraints and initial solution of the problem to be solved based on the equality constraints, and obtains the transformed constraint set and initial solution. The constraint set consists only of inequality constraints. Secondly, based on the constraint set and the normal vector that violates the inequality constraints, the position of the transformed initial solution is iteratively adjusted until the infeasible solution of the initial solution falls into the feasible region. Finally, based on the equality constraints and the infeasible solution of the initial solution, the initial solution is restored to obtain an intermediate feasible solution that satisfies the equality constraints and inequality constraints.

[0103] FONET includes a feasible solution adjustment direction activation function and a feasible solution adjustment distance activation function. The former adjusts its own parameters based on intermediate feasible solutions, equality constraints of the optimal DC power flow problem, and the objective function, ensuring that a better feasible solution exists in the adjustment direction output by the feasible solution adjustment neural network. The latter adjusts its own parameters based on the adjustment direction, intermediate feasible solutions, inequality constraints of the optimal DC power flow problem, and the objective function, ensuring that the adjustment displacement output by the feasible solution adjustment neural network is a non-suboptimal feasible solution. The feasible solution adjustment direction activation function is described in the OAIRN section below; the feasible solution adjustment distance activation function is described in the OAIRN section below.

[0104] See Figure 1 , Figure 1 This is a flowchart illustrating a data mechanism-driven method for solving optimal DC power flow, provided in an embodiment of this application. This method can be applied to servers or server clusters. Specifically, the method includes steps S101-S107.

[0105] The embodiments of this application can be applied to a DNN application framework that uses a DNN after preprocessing (UDAP), ensuring the feasibility and optimality of the DNN output data. The method and DNN application framework described in the embodiments of this application can achieve a balance between the computational efficiency of the DNN method, the feasibility of the output data, and the optimality of the output data.

[0106] See Figure 2 , Figure 2This application provides an overall flowchart for solving the DCOPF problem using UDAP (Unified Activated Neural Network Approach) based on preprocessing a DNN. The "solving the DNN" refers to any neural network capable of providing an initial solution to the DCOPF problem; UDAP does not require the neural network to be feasible or optimal. Furthermore, all computational steps in UDAP, including the constraint-guided feasibility modification algorithm (FEMA II) and the neural network with feasibility-ensured and optimality-embedded (FONET), are crucial steps ensuring the effectiveness of the proposed framework. The application of the FEMA II algorithm includes steps S103-S105, and the application of the FONET neural network includes steps S106-S107.

[0107] S101: Input the power demand into a deep neural network (DNN) to obtain the unit output schemes output by the deep neural network (DNN). Any one of the unit output schemes in the unit processing schemes is the initial solution.

[0108] Step S101 is used to "solve the DNN", which inputs the power demand into multiple pre-trained deep neural network models (DNNs) to obtain different unit output schemes from the multiple DNNs. Any one of the multiple unit output schemes can be the initial solution. In one possible implementation, the unit output scheme with the least degree of inequality constraint violation is the initial solution, but this application does not limit this.

[0109] After determining the initial solution, the variable elimination method can be used to process the problem to be solved and the initial solution, transforming it into an optimization problem without equality constraints and the corresponding initial solution.

[0110] Determine whether the initial solution satisfies the problem constraints. If the initial solution satisfies the problem constraints, proceed to step S102 and subsequent steps; if the initial solution does not satisfy the problem constraints, proceed to step S103 and subsequent steps. The problem constraints include equality constraints and inequality constraints.

[0111] S102: Use the initial solution as an intermediate feasible solution. Jump to steps S106-S107.

[0112] S103: The constraints and initial solutions of the problem to be solved are processed based on equality constraints by the variable elimination method to obtain the transformed constraint set and initial solution. The constraint set consists only of inequality constraints.

[0113] The FEMA II algorithm consists of three steps. Step A is the transformation of the constraint set and infeasible solutions based on equality constraints. As a linear programming problem, all inequality constraints (2) and (4) and equality constraints (3) in DCOPF can be expressed in the form of matrix multiplication as follows: and Due to the special nature of the DCOPF problem, For a matrix with only one row, Let be a 1x1 matrix. The Middle Divide the value in the first column of the row by The value in the first column is By and The first in row subtraction and Make the inequality constraints and Decouple the variables corresponding to the first column in the constraints, and then... and The first column is removed, transforming the constraint set of the DCOPF problem into a constraint set consisting only of inequality constraints. .

[0114] For infeasible solutions ,Will Remove the first variable from the list and arrange the remaining variables in their original order. . This makes conform to and , equivalent to conform to .

[0115] S104: Based on the constraint set and the normal vector that violates the inequality constraint, iteratively adjust the position of the initial solution until the infeasible solution of the initial solution falls into the feasible region.

[0116] Step B is an iterative algorithm for making infeasible solutions feasible based on the relationship between infeasible solutions and inequality constraint spaces. The transformed infeasible solutions... Iterative movement, making The final set of constraints is satisfied after transformation.

[0117] In one possible implementation, the iterative adjustment of the position of the initial solution based on the constraint set and the normal vector violating the inequality constraints until an infeasible solution of the initial solution falls into the feasible region includes the following steps:

[0118] 1. Calculate the infeasible solutions to the initial solution based on the constraint set. The inequality constraint that was violated.

[0119] 2. Calculate the normal vectors that violate the inequality constraints, and use the sum of the normal vectors as an infeasible solution to the initial solution. direction of movement .

[0120] 3. Calculate the infeasible solutions of the initial solution. In the direction of movement The first inequality constraint it encounters that it violates, as distance of movement Infeasible solutions to the initial solution The distance between the initial violated inequality constraint and the infeasible solution is the single-iteration move distance of the initial solution. The initially violated inequality constraint is in the direction of movement. The first inequality constraint violated.

[0121] 4. Infeasible solutions from the initial solution According to the above direction of movement and distance traveled After making the move, calculate the infeasible solutions of the initial solution. Find all hyperplanes in which it is located, and calculate the intersection plane of the hyperplanes.

[0122] 5. Change the current direction of movement Projected onto the intersection plane, a new direction of movement is generated.

[0123] 6. Repeat steps 3-5 above until the intersection plane or the new direction of movement meets the preset conditions. The preset conditions are that the intersection plane is a point or the new direction of movement is a vector with all zeros.

[0124] 7. Repeat steps 1-6 above until an infeasible solution to the initial solution is determined. It falls into the feasible region.

[0125] S105: Based on the equality constraints and the infeasible solutions of the initial solution, the initial solution is restored to obtain an intermediate feasible solution that satisfies the equality constraints and the inequality constraints.

[0126] Step C involves restoring the feasible solution based on equality constraints. This is done by considering the infeasible solutions from the initial solution. Calculation of DCOPF equality constraints The first variable in the equation, by combining this variable with... By concatenating the solutions, we can obtain solutions that simultaneously satisfy the equality and inequality constraints of the DCOPF problem. This should satisfy both equality and inequality constraints. This is an intermediate feasible solution.

[0127] S106: Input the intermediate feasible solution into the pre-constructed neural network FONET, which guarantees feasibility and is embedded in optimality, and determine the adjustment information of the intermediate feasible solution.

[0128] FONET can obtain adjustment directions based on existing feasible solutions, ensuring that the adjustment result is an equally or better feasible solution. In this embodiment, FONET is used for fast feasible solution iteration, thereby efficiently obtaining feasible solutions to the DCOPF problem considering optimality. The adjustment information includes adjustment direction and adjustment distance. FONET includes an optimality-aware-equality-constraint-remain neural network (OAERN) that guarantees the satisfaction of equality constraints and an optimality-aware-inequality-constraint-remain neural network (OAIRN) that guarantees the satisfaction of inequality constraints. OAERN is used to determine the adjustment direction, and OAIRN is used to determine the adjustment distance.

[0129] In one possible implementation, the step of inputting the intermediate feasible solution into a pre-constructed neural network FONET, which guarantees feasibility and is embedded in optimality, to determine the adjustment information of the intermediate feasible solution includes: inputting the intermediate feasible solution into FONET to obtain a first activation function OAERN, which is used to determine the adjustment direction; determining a backup adjustment distance based on the adjustment direction and a first preset formula; calculating a primary adjustment distance according to a second preset formula; and determining the adjustment distance from the primary adjustment distance and the backup adjustment distance.

[0130] In this embodiment, an activation function is designed that ensures the movement direction of the DNN output does not change the satisfaction of the equality constraints of the existing feasible solution, and that a better solution than the known feasible solution exists in this movement direction. The neural network that uses the above activation function as the activation function of the last layer is called OAERN. The internal framework of OAERN is as follows: Figure 3 As shown.

[0131] The main steps in determining the activation function include:

[0132] 1. Calculate the average marginal cost matrix of all units currently under solution. .

[0133] 2. Mark units with marginal costs higher than average costs as negative labels, and units with marginal costs lower than average costs as positive labels.

[0134] 3. Incorporate positive and negative labels into the activation function. middle:

[0135] (6)

[0136] (7)

[0137] (8)

[0138] (9)

[0139] in, It is a description of the characteristics of each unit output by the neural layer. 3D matrix It refers to the number of generating units and has no direct physical properties. and It is a matrix used to extract features from positive and negative labeled units, with dimensions of respectively. and .exist and In this context, each row represents a unit labeled positive or negative, and each column represents a unit in a target problem. If the first... The unit is the first For each unit marked as positive, then The row and number The column is 1, while all other elements in the row are 0. This indicates the transpose operation. This represents the diagonal matrix representation of a vector. This indicates taking the absolute value. This represents the summation of all elements in the matrix.

[0140] Once the direction is determined, the maximum distance that can be adjusted is... In Substituting into formulas (2) and (4) We obtain the following two equations:

[0141] (10)

[0142] (11)

[0143] By unifying these formulas into The adjustment direction of the current solution in the OAERN output can be calculated. The first preset formula includes the following formulas (12) and (13), and the maximum adjustable distance. (That is, the backup adjustment distance) is:

[0144] (12)

[0145] (13)

[0146] To ensure that a better solution exists in the final direction of movement, the minimum adjustment distance cannot be negative; therefore, the minimum adjustment distance... The value is 0. Based on the boundary values ​​of the above distance interval, a function with the form of SoftMax can be used. This restricts the output data of the neural network to between the minimum and maximum adjustment distance. The specific function is:

[0147] (14)

[0148] by A neural network with the last activation function can guarantee the feasibility of the adjustment result, but cannot guarantee optimality. Similar to the maximum adjustment distance, the optimal adjustment distance is also determined as the adjustment direction is determined. The second preset formula is the following formula (15), and the main adjustment distance can be expressed as:

[0149] (15)

[0150] in, It consists of the marginal cost of each unit. 3D matrix and yes The unit output cost coefficient matrix is ​​a dimensioned matrix, where the value of each element in each row is the secondary or primary cost coefficient of the unit in that row. This is an element-wise multiplication.

[0151] comprehensive and This can simultaneously ensure the feasibility of the adjustment results is maintained and their optimality is improved. The specific principle is: when When DCOPF optimality is increased, utilize As the final adjustment distance, otherwise utilize As the final adjustment distance The neural network described above, which incorporates inequality constraints and an objective function, is called OAIRN. The flowchart for obtaining the adjusted displacement in collaboration with OAIRN is shown below. Figure 4 As shown in the figure. EC, IC, and OF represent the equality constraint, inequality constraint, and objective function, respectively.

[0152] Using a DNN after the preprocessing (UDAP) is a DNN application for the DCOPF problem. It employs a DNN preprocessing method centered on the initial feasible solution, the FEMA II algorithm capable of obtaining feasible solutions in a finite number of steps, and a DNN activation function that guarantees the adjusted result is an equally or better feasible solution. and .

[0153] S107: Adjust the intermediate feasible solution based on the target distance to obtain the target feasible solution.

[0154] The advantage of the FEMA II algorithm lies in its pioneering approach of providing a feasible solution within a finite number of computations without requiring gradient calculations. Its satisfaction of equality constraints is guaranteed in step C above, while its satisfaction of inequality constraints is guaranteed by the following explanation:

[0155] See Figure 5 , Figure 5 This is a schematic diagram illustrating the movement effect of an infeasible solution according to an embodiment of this application. The feasibility-to-feasibility iteration is explained as follows: I: For an infeasible solution and a constraint it violates, as the solution moves along the normal to the hyperplane corresponding to the constraint towards that hyperplane, and before reaching the hyperplane, the infeasible solution will continuously approach the feasible region.

[0156] See Figure 5 As shown in part (a), once the infeasible solution and the hyperplane are given, the point in the feasible region can be determined. and Value. Clearly in The distance between infeasible solutions and any feasible point as the solution approaches zero. It is bound to keep shrinking.

[0157] II: For an infeasible solution and any number of constraints violated by that infeasible solution, the hyperplane corresponding to the constraints is denoted as . And the normal vector of the hyperplane that is far from the infeasible solution is denoted as Then the following conclusion must hold: when the infeasible solution is along the direction obtained by arbitrarily combining these normal vectors... ,in For any number, when making a move, the infeasible solution is always close to the feasible region before moving to any of the above hyperplanes.

[0158] Assuming the current point is along the direction Above and The minimum distance between each hyperplane is Then, by displacement Decomposed into Second-rate The sub-displacements are then further decomposed into sub-directions under each sub-direction. The displacement. At this level of subdivision, the starting point of each sub-displacement violates all constraints violated by the starting point of the entire displacement. Therefore, regardless of which hyperplane corresponds to the violated constraint, the displacement will move one step closer to the feasible region, as described in section I above. For further understanding, please refer to... Figure 5 Parts (b) and (c) shown illustrate... Two specific situations at that time.

[0159] III: After projecting the movement direction onto the hyperplane corresponding to the inequality constraint of the infeasible solution or the intersection of hyperplanes, the infeasible solution continues to approach the feasible region before encountering the new inequality constraint hyperplane in the new movement direction.

[0160] The distance relationship between any three points in a high-dimensional space will not change after dimensionality reduction. Therefore, if a direction in the original multidimensional space can make an infeasible solution approach the feasible region, then after dimensionality reduction, the infeasible solution after dimensionality reduction will also move closer to the feasible region after dimensionality reduction if it moves in the direction after dimensionality reduction.

[0161] For easier understanding, please refer to Figure 5 Part (d) is shown, where the two movements correspond to... and The first move will resolve the infeasible solution from... Move to At this time, it is under constraint On the corresponding hyperplane, the direction of the second movement is normal vector Under constraints The projection of the hyperplane it lies on is therefore an infeasible solution. will along Move towards the feasible region Thus, a feasible solution can be obtained.

[0162] IV: The number of correction steps required to make the infeasible solution satisfy any inequality constraint is no greater than the number of inequality constraints violated during the correction process. The total number of correction steps must be less than .

[0163] During the transition, when an infeasible solution is moved to any inequality constraint... After moving to the corresponding hyperplane or hyperplane intersection, only when moving to a vertex of a high-dimensional polyhedron composed of multiple inequality constraints will the next move lead to an infeasible solution for the inequality constraints. The infeasible solution will not repeatedly move to this vertex because it is always moving towards the feasible region. The number of times an infeasible solution moves to the feasible region is less than the number of vertices. Since a vertex is composed of a complex number of inequality constraints that are not satisfied by infeasible solutions, the number of moves is less than the number of infeasible solutions that were violated during the move.

[0164] Each iteration in OAIRN moves infeasible solutions closer to the feasible region. Therefore, as the number of iterations increases, infeasible solutions will eventually enter the feasible region.

[0165] OAERN guarantees that no point in the moving direction changes the current solution's satisfaction of the equality constraints, and that there must exist a better solution in the moving direction than the current solution.

[0166] According to the distributive and associative laws of matrix multiplication, the right side of formula (5) can be divided into two parts, namely:

[0167] (16)

[0168] According to the matrix and From the construction process, we can deduce that: and Each column is a vector consisting of one element equal to 1 and multiple elements equal to 0. Therefore, the sum of the left multiplication or the front and back multiplication of a matrix remains unchanged. This property is expressed by the following equations (17) and (18):

[0169] (17)

[0170] (18)

[0171] It can also be inferred that:

[0172] If the diagonal matrix If the diagonal element of any column is +1, then There must be one and only one element equal to 1 in the same row, otherwise The entire row is zero. Conversely, if If the diagonal element of any column is -1, then If there is one and only one element equal to 1 in the same row, otherwise... The entire row is zero. It can be deduced that... as well as Therefore, equation (16) can be expressed as:

[0173] (19)

[0174] Summing equation (19):

[0175] (20)

[0176] According to the definition of formula (7), and The sum of all elements in the matrix is ​​1, therefore, the result of equation (20) must be equal to 0. When adjusting the direction equal At that time, it must exist: That is, moving the solution does not change the satisfaction of the equality constraints of the current solution.

[0177] Secondly, according to equation (5), the objective function (1) can be rewritten as:

[0178] (twenty one)

[0179] Therefore, it exists:

[0180] (twenty two)

[0181] Wherein, the satisfaction of equation (22) is equivalent to Compare Even better. According to the distributive property of multiplication and equation (20), when adjusting the distance... When, the left side of inequality (22) can be equivalent to:

[0182] (twenty three)

[0183] in, It consists of the marginal cost of each unit. A 3D matrix. Based on the criteria for labeling positive and negative tags and... The process of building and using it, once The following phenomenon always exists: in the marginal cost matrix, In a marginal cost matrix, the value at each negative element must be greater than the average marginal cost, and vice versa. Therefore, if we denote the minimum value greater than the average marginal cost and the maximum value less than the average marginal cost in the marginal cost matrix as... and Then it must be true that:

[0184] (twenty four)

[0185] That is when When the condition is met, equation (24) is satisfied, therefore there is always a better solution than the known feasible solution in the adjustment direction of the OAERN output.

[0186] It should be noted that the first neural layer of FONET receives intermediate feasible solutions, while the other neural layers receive relevant calculated data such as electricity demand, generator cost model parameters, equality constraints, and inequality constraints. Running FONET to process the above data can yield a better objective feasible solution.

[0187] The foregoing embodiments of this application provide a data mechanism-based joint driving method for solving optimal DC power flow. Next, we will describe a data mechanism-based joint driving system for solving optimal DC power flow, also provided in this application, which is used to execute the aforementioned... Figure 1 The method is shown. The function of the data mechanism joint drive system for solving DC optimal power flow is explained, and a schematic diagram of the structure of this data mechanism joint drive system for solving DC optimal power flow is shown below. Figure 6 As shown, it includes an acquisition module 601, a judgment module 602, an adjustment module 603, a restoration module 604, and a determination module 605.

[0188] The acquisition module 601 is used to input the power demand into the deep neural network DNN to obtain the unit output scheme output by the deep neural network DNN, and any one of the unit output schemes in the unit processing scheme is the initial solution;

[0189] The judgment module is used to determine if the initial solution satisfies the problem constraints, and then use the initial solution as an intermediate feasible solution, wherein the problem constraints include equality constraints and inequality constraints.

[0190] The judgment module 602 is also used to process the constraints and initial solution of the problem to be solved based on the equality constraints by using the variable elimination method if the initial solution does not satisfy the problem constraints, so as to obtain the transformed constraint set and initial solution, wherein the constraint set consists only of inequality constraints.

[0191] The adjustment module 603 is used to iteratively adjust the position of the initial solution based on the constraint set and the normal vector that violates the inequality constraint, until the infeasible solution of the initial solution falls into the feasible region;

[0192] The restoration module 604 is used to restore the initial solution based on the equality constraints and the infeasible solutions of the initial solution to obtain an intermediate feasible solution that satisfies the equality constraints and the inequality constraints.

[0193] The determination module 605 is used to input the intermediate feasible solution into the pre-constructed neural network FONET that guarantees feasibility and embeds optimality, and determine the adjustment information of the intermediate feasible solution;

[0194] The adjustment module 603 is further configured to adjust the intermediate feasible solution based on the adjustment information to obtain the target feasible solution.

[0195] In one possible implementation, the adjustment module 603 is used to iteratively adjust the position of the initial solution based on the constraint set and the normal vectors that violate the inequality constraints, until the infeasible solution of the initial solution falls into the feasible region, including:

[0196] The calculation submodule is used to calculate the inequality constraints violated by the infeasible solution of the initial solution based on the constraint set; calculate the normal vectors that violate the inequality constraints, and use the sum of the normal vectors as the direction of movement of the infeasible solution of the initial solution;

[0197] The determination submodule is used to take the distance between the infeasible solution of the initial solution and the initially violated inequality constraint as the single iteration movement distance of the infeasible solution of the initial solution, wherein the initially violated inequality constraint is the first violated inequality constraint encountered in the movement direction.

[0198] The calculation submodule is also used to calculate the intersection plane of the hyperplane where the infeasible solution of the initial solution is located after being adjusted according to the moving direction and moving distance;

[0199] A generation submodule is used to project the movement direction onto the intersection plane to generate a new movement direction;

[0200] The determining submodule is used to determine that the infeasible solution of the initial solution falls into the feasible region until the intersection plane or the new movement direction meets a preset condition. The preset condition is that the intersection plane is a point or the new movement direction is a vector with all zeros.

[0201] In one possible implementation, the adjustment information includes the adjustment direction and the adjustment distance, and the determining module 605 is specifically used for:

[0202] The intermediate feasible solution is input into the FONET to obtain a first activation function, which is used to determine the adjustment direction.

[0203] The backup adjustment distance is determined based on the adjustment direction and the first preset formula;

[0204] Calculate the main adjustment distance according to the second preset formula;

[0205] The adjustment distance is determined from the primary adjustment distance and the backup adjustment distance.

[0206] In one possible implementation, the FONET includes an optimal neural network OAERN that guarantees the satisfaction of equality constraints and an optimal neural network OAERN that guarantees the satisfaction of inequality constraints; the OAERN is used to determine the adjustment direction and the OAERN is used to determine the adjustment distance.

[0207] This application provides a data mechanism-driven system for solving DC optimal power flow problems, including an acquisition module, a judgment module, an adjustment module, a restoration module, and a determination module. This system is applicable to DCOPF problems with feasible solutions under any power topology. The acquisition module inputs power demand into a deep neural network (DNN) to obtain the unit output schemes output by the DNN. Any one of the unit output schemes is the initial solution. The judgment module, if the initial solution satisfies the problem constraints (including equality and inequality constraints), uses the variable elimination method to process the constraints and initial solution based on equality constraints, obtaining a transformed constraint set and initial solution. The transformed constraint set only contains inequality constraints. The adjustment module iteratively adjusts the position of the initial solution based on the constraint set and the normal vectors violating inequality constraints until an infeasible solution of the initial solution falls into the feasible region. The restoration module restores the initial solution based on equality constraints and infeasible solutions of the initial solution, obtaining an intermediate feasible solution that satisfies both equality and inequality constraints. The determination module is used to input the intermediate feasible solution into a pre-constructed neural network FONET, which guarantees feasibility and embeds optimality, to determine the adjustment information of the intermediate feasible solution. The adjustment module is also used to adjust the intermediate feasible solution based on the adjustment information to obtain the target feasible solution. In this way, determining the adjustment information and adjusting the initial solution through the adjustment information can simultaneously ensure the solution efficiency and the feasibility of the solution result, and the obtained solution result has higher optimality.

[0208] Based on the data mechanism joint driving method for solving DC optimal power flow provided in the above-described method embodiments, this application embodiment provides a data mechanism joint driving device for solving DC optimal power flow. See [link to relevant documentation]. Figure 7 The device includes: a processor, a memory, and a system bus;

[0209] The processor and the memory are connected via the system bus;

[0210] The memory is used to store one or more programs, the one or more programs including instructions, which, when executed by the processor, cause the processor to perform the data mechanism joint driving method for solving DC optimal power flow as described in any of the above embodiments.

[0211] Based on the above-described method embodiments, this application provides a data mechanism joint driving method for solving DC optimal power flow. The storage medium stores instructions that, when executed by a computer, cause the computer to implement the data mechanism joint driving method for solving DC optimal power flow as described in any of the above embodiments.

[0212] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0213] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0214] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A data mechanism-driven method for solving optimal DC power flow, characterized in that, The method includes: The power demand is input into a deep neural network (DNN) to obtain the unit output schemes output by the DNN. Any one of the unit output schemes is the initial solution. If the initial solution satisfies the problem constraints, then the initial solution is taken as an intermediate feasible solution, wherein the problem constraints include equality constraints and inequality constraints; If the initial solution does not satisfy the problem constraints, the constraints and initial solution of the problem to be solved are processed based on the equality constraints by the variable elimination method to obtain the transformed constraint set and initial solution. The constraint set consists only of inequality constraints. Based on the constraint set and the normal vector that violates the inequality constraint, the position of the initial solution is iteratively adjusted until the infeasible solution of the initial solution falls into the feasible region; The initial solution is restored based on the equality constraints and the infeasible solution of the initial solution to obtain an intermediate feasible solution that satisfies the equality constraints and the inequality constraints. The intermediate feasible solution is input into the pre-constructed neural network FONET, which guarantees feasibility and is embedded in optimality, to determine the adjustment information of the intermediate feasible solution; The intermediate feasible solution is adjusted based on the adjustment information to obtain the target feasible solution.

2. The method according to claim 1, characterized in that, The iterative adjustment of the position of the initial solution based on the constraint set and the normal vector violating the inequality constraints, until the infeasible solution of the initial solution falls into the feasible region, includes: Calculate the inequality constraints violated by the infeasible solutions of the initial solution based on the constraint set; Calculate the normal vector that violates the inequality constraint, and use the sum of the normal vectors as the direction of movement for the infeasible solution of the initial solution; The distance between the infeasible solution of the initial solution and the initially violated inequality constraint is taken as the single iteration movement distance of the infeasible solution of the initial solution, and the initially violated inequality constraint is the first violated inequality constraint encountered in the movement direction. Calculate the intersection plane of the hyperplane where the infeasible solutions of the initial solution are located after adjustment according to the moving direction and moving distance; The movement direction is projected onto the intersection plane to generate a new movement direction; Until the intersection plane or the new direction of movement satisfies a preset condition, it is determined that the infeasible solution of the initial solution falls into the feasible region. The preset condition is that the intersection plane is a point or the new direction of movement is a vector of all zeros.

3. The method according to claim 1, characterized in that, The adjustment information includes adjustment direction and adjustment distance. The step of inputting the intermediate feasible solution into a pre-constructed neural network FONET, which guarantees feasibility and embeds optimality, to determine the adjustment information of the intermediate feasible solution includes: The intermediate feasible solution is input into the FONET to obtain a first activation function, which is used to determine the adjustment direction. The backup adjustment distance is determined based on the adjustment direction and the first preset formula; Calculate the main adjustment distance according to the second preset formula; The adjustment distance is determined from the primary adjustment distance and the backup adjustment distance.

4. The method according to claim 3, characterized in that, The FONET includes an optimal neural network OAERN that guarantees the satisfaction of equality constraints and an optimal neural network OAIRN that guarantees the satisfaction of inequality constraints; the OAERN is used to determine the adjustment direction and the OAIRN is used to determine the adjustment distance.

5. A data mechanism-based joint drive system for solving optimal DC power flow, characterized in that, The system includes: The acquisition module is used to input the power demand into the deep neural network (DNN) to obtain the unit output schemes output by the deep neural network (DNN), where any one of the unit output schemes is the initial solution. The judgment module is used to determine if the initial solution satisfies the problem constraints, and then use the initial solution as an intermediate feasible solution, wherein the problem constraints include equality constraints and inequality constraints. The judgment module is also used to process the constraints and initial solution of the problem to be solved by the variable elimination method based on the equality constraints if the initial solution does not satisfy the problem constraints, so as to obtain the transformed constraint set and initial solution. The constraint set consists only of inequality constraints. An adjustment module is used to iteratively adjust the position of the initial solution based on the constraint set and the normal vector that violates the inequality constraint, until the infeasible solution of the initial solution falls into the feasible region; The restoration module is used to restore the initial solution based on the equality constraints and the infeasible solutions of the initial solution to obtain an intermediate feasible solution that satisfies the equality constraints and the inequality constraints. The determination module is used to input the intermediate feasible solution into the pre-constructed neural network FONET, which guarantees feasibility and is embedded in optimality, and determine the adjustment information of the intermediate feasible solution; The adjustment module is further configured to adjust the intermediate feasible solution based on the adjustment information to obtain the target feasible solution.

6. The system according to claim 5, characterized in that, The adjustment module is used to iteratively adjust the position of the initial solution based on the constraint set and the normal vector that violates the inequality constraint, until the infeasible solution of the initial solution falls into the feasible region, including: The calculation submodule is used to calculate the inequality constraints violated by the infeasible solution of the initial solution based on the constraint set; calculate the normal vectors that violate the inequality constraints, and use the sum of the normal vectors as the direction of movement of the infeasible solution of the initial solution; The determination submodule is used to take the distance between the infeasible solution of the initial solution and the initially violated inequality constraint as the single iteration movement distance of the infeasible solution of the initial solution, wherein the initially violated inequality constraint is the first violated inequality constraint encountered in the movement direction. The calculation submodule is also used to calculate the intersection plane of the hyperplane where the infeasible solution of the initial solution is located after being adjusted according to the moving direction and moving distance; A generation submodule is used to project the movement direction onto the intersection plane to generate a new movement direction; The determining submodule is used to determine that the infeasible solution of the initial solution falls into the feasible region until the intersection plane or the new movement direction meets a preset condition. The preset condition is that the intersection plane is a point or the new movement direction is a vector with all zeros.

7. The system according to claim 5, characterized in that, The adjustment information includes the adjustment direction and the adjustment distance, and the determining module is specifically used for: The intermediate feasible solution is input into the FONET to obtain a first activation function, which is used to determine the adjustment direction. The backup adjustment distance is determined based on the adjustment direction and the first preset formula; Calculate the main adjustment distance according to the second preset formula; The adjustment distance is determined from the primary adjustment distance and the backup adjustment distance.

8. The system according to claim 7, characterized in that, The FONET includes an optimal neural network OAERN that guarantees the satisfaction of equality constraints and an optimal neural network OAERN that guarantees the satisfaction of inequality constraints; the OAERN is used to determine the adjustment direction and the OAERN is used to determine the adjustment distance.

9. A data mechanism-based joint drive device for solving optimal DC power flow, characterized in that, The device includes: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform the data mechanism joint driving method for solving DC optimal power flow as described in any one of claims 1 to 4.

10. A storage medium, characterized in that, The storage medium stores instructions that, when executed by a computer, cause the computer to implement the data mechanism joint driving method for solving DC optimal power flow as described in any one of claims 1 to 4.

Images