A reliability-constrained machine learning robust planning method for ac-dc hybrid distribution networks

Abstract

The application discloses a reliability-constrained AC / DC hybrid power distribution network machine learning robust planning method, which comprises the following steps: acquiring and initializing basic operation parameters of an AC / DC hybrid power distribution network, constructing a target planning model and adding basic operation constraints of the power distribution network; deriving core reliability indexes of the power distribution network from line and node fault double-dimension analysis and converting them into a mixed integer linear constraint embedded model to obtain a reliability coupled planning model; constructing a budget-constrained uncertainty set and converting it into a two-stage robust planning model; and adopting a machine learning enhanced column constraint generation algorithm to decompose and iteratively solve, and calling a solver to output an optimal planning result after convergence. The application realizes the collaborative optimization of the economy and reliability of a planning scheme, greatly improves the solving efficiency, and adapts to the engineering planning requirements of the AC / DC hybrid power distribution network.

Description

Technical Field

[0001] This invention relates to the field of power system distribution network planning technology, specifically to a robust machine learning planning method for reliability-constrained AC / DC hybrid distribution networks. Background Technology

[0002] With the rapid development of the new energy industry, the limitations of traditional AC distribution networks in terms of power conversion efficiency and new energy absorption capacity are becoming increasingly apparent, making them unsuitable for the development needs of new power systems. While DC distribution networks possess technological advantages in new energy absorption and loss control, large-scale deployment is unlikely in the short term. Against this backdrop, hybrid AC / DC distribution networks have become the optimal transitional solution connecting AC and DC power supply elements, balancing economic efficiency and reliability, and represent an important direction for the development of new distribution networks.

[0003] Currently, research on AC / DC hybrid distribution network planning has been conducted from different dimensions. Some studies use deterministic or heuristic algorithms to handle source-load uncertainty, while others focus on collaborative planning involving multiple devices and stakeholders. However, existing research generally suffers from two major shortcomings: First, most studies treat reliability as a post-event evaluation indicator, failing to deeply couple it with decision variables in the planning model. This makes it difficult to coordinate the optimization of the economy and reliability of the planning scheme, and it is impossible to accurately quantify and control the power supply reliability level during the planning stage. Second, the strong randomness of distributed power output does not fully consider the impact on the robustness of the planning scheme. Even if some studies introduce robust optimization methods, the traditional column constraint generation (C&CG) algorithm suffers from excessive iterations, long solution time, and exponential growth of variable size with the number of scenarios, making it difficult to adapt to the planning scenarios of AC / DC hybrid distribution networks with multiple nodes and branches.

[0004] Therefore, developing a hybrid AC / DC distribution network planning method that can explicitly integrate reliability and planning model while taking into account the robustness of the solution and the efficiency of the solution has become an urgent technical problem to be solved in the industry. Summary of the Invention

[0005] The purpose of this invention is to provide a robust planning method for machine learning in a reliability-constrained AC / DC hybrid distribution network, so as to achieve synergistic optimization of the economic efficiency and power supply reliability of the planning scheme, while taking into account the application requirements of scheme robustness and rapid engineering planning.

[0006] To address the aforementioned technical problems, this invention provides a robust machine learning planning method for reliability-constrained AC / DC hybrid distribution networks, comprising: S1. Obtain and initialize the basic operating parameters of the AC / DC hybrid distribution network; the basic operating parameters serve as the input basis for subsequent model construction, constraint derivation and solution calculation; S2. Based on the aforementioned basic operating parameters, construct a target planning model for the AC / DC hybrid distribution network, and add basic operating constraints for the distribution network to the target planning model; S3. The preset core reliability indicators of the distribution network are analytically derived from the dual dimensions of line faults and node faults, and then transformed into mixed integer linear constraints and embedded into the target planning model to obtain the reliability coupled planning model. S4. To address the uncertainty of distributed power output, a budget-constrained uncertainty set is constructed. This budget-constrained uncertainty set is used as the fluctuation constraint boundary of distributed power output, and the reliability coupling planning model is transformed into a two-stage robust planning model. S5. Employ a machine learning-enhanced column constraint generation algorithm to decompose the two-stage robust programming model into a main problem and sub-problems. The model is then solved and optimized through iterative interaction between the main problem and sub-problems until it converges. S6. Call the commercial solver to calculate the two-stage robust programming model after iterative convergence and output the optimal planning result of the AC / DC hybrid distribution network.

[0007] According to the above scheme, the basic operating parameters of the AC / DC hybrid distribution network include distribution network line parameters, distributed power source access capacity and node information, and load baseline values.

[0008] According to the above scheme, the objective function of the target programming model is to minimize the comprehensive cost, which includes investment and construction costs, operation and maintenance costs, power production costs, and reliability costs.

[0009] According to the above scheme, the basic operating constraints of the distribution network include power flow constraints, node voltage constraints, branch current constraints, and radial operating constraints.

[0010] According to the above scheme, the preset core reliability indicators of the distribution network include the expected value of insufficient power supply, the average power outage frequency index, and the average power outage duration index.

[0011] According to the above scheme, the two-stage robust planning model includes a first-stage planning model and a second-stage operation model. The first-stage planning model is used to determine the distribution network node type and branch construction scheme before the uncertainty is realized, and the second-stage operation model is used to adjust the distribution network operation parameters to cope with the output fluctuation of distributed power sources after the uncertainty is realized.

[0012] According to the above scheme, the machine learning-enhanced column constraint generation algorithm uses a pre-trained neural network, which is used to predict the optimal value of the running model in the second stage based on the decision variables and uncertainty parameters in the first stage.

[0013] According to the above scheme, the decomposed main problem is used to make distribution network planning and optimization decisions based on the adverse scenarios output by the sub-problems. The decomposed sub-problems are used to solve the worst-case scenario of distributed power generation output and feed it back to the main problem to complete iterative optimization.

[0014] According to the above scheme, the commercial solver used is the Gurobi solver.

[0015] This invention also provides a robust machine learning planning system for reliability-constrained AC / DC hybrid distribution networks, comprising: The data initialization module is used to acquire and initialize the basic operating parameters of the AC / DC hybrid distribution network; these basic operating parameters serve as the input basis for subsequent model construction, constraint derivation, and solution calculation. The basic model building module is used to build a target planning model for the AC / DC hybrid distribution network based on the basic operating parameters, and to add basic operating constraints of the distribution network to the target planning model. The reliability coupling module is used to analyze and derive the preset core reliability indicators of the distribution network from the dual dimensions of line faults and node faults, transform them into mixed integer linear constraints, and then embed them into the target planning model to obtain the reliability coupling planning model. The robust model construction module is used to construct a budget-constrained uncertainty set for the output uncertainty of distributed power sources. The budget-constrained uncertainty set is used as the fluctuation constraint boundary of the output of distributed power sources to transform the reliability coupled planning model into a two-stage robust planning model. The model iterative solution module is used to decompose the two-stage robust programming model into a main problem and sub-problems using a machine learning-enhanced column constraint generation algorithm. The solution optimization is completed through iterative interaction between the main problem and sub-problems until the model converges. The results output module is used to call a commercial solver to calculate the two-stage robust programming model after iterative convergence and output the optimal planning results for the AC / DC hybrid distribution network.

[0016] Beneficial effects This invention achieves deep coupling between power supply reliability and distribution network planning decisions by analytically deriving core reliability indicators of the distribution network from both line faults and node faults, transforming them into mixed-integer linear constraints, and embedding them into the target planning model. This allows for direct quantification and control of power supply reliability levels during the planning stage. Furthermore, it constructs a budget-constrained uncertainty set for distributed power generation output uncertainty and transforms it into a two-stage robust planning model, effectively addressing fluctuations in new energy output and ensuring the robustness of the planning scheme. Finally, it employs a machine learning-enhanced column constraint generation algorithm to decompose the model into main and sub-problems for iterative solution, significantly improving model solving efficiency and ultimately outputting the optimal planning result for an AC / DC hybrid distribution network that balances reliability, robustness, and economy. Attached Figure Description

[0017] Figure 1 This is a flowchart of a reliability-constrained AC / DC hybrid distribution network machine learning robust planning method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a machine learning-enhanced C&CG algorithm framework according to an embodiment of the present invention; Figure 3 This is a schematic diagram of an improved 54-node distribution network topology according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the planning result according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0019] To address the technical problems in existing AC / DC hybrid distribution network planning, such as the inability to balance economy and reliability, the separation of reliability assessment and planning models, the difficulty in handling uncertainties of distributed generation sources, and the low efficiency of model solving, this embodiment provides a robust machine learning planning method for reliability-constrained AC / DC hybrid distribution networks. First, a target planning model for the AC / DC hybrid distribution network, encompassing investment and construction, operation and maintenance, power production, and reliability costs, is constructed, incorporating basic operational constraints such as power flow, node voltage, branch current, and radial operation. Second, the three core reliability indices—EENS (Expected Energy Not Supplied), SAIFI (System Average Interruption Frequency Index), and SAIDI (System Average Interruption Duration Index)—are analytically derived from both line and node fault dimensions, and transformed into mixed-integer linear constraints embedded in the planning model, achieving explicit fusion and deep coupling between reliability and the planning model. Then, a two-stage robust planning model is established to address the uncertainty of distributed generation output. Finally, a machine learning-enhanced C&CG algorithm is used to decompose the robust model into a main problem and sub-problems for iterative solving.

[0020] Step 1: Initialization, input the distribution network line parameters, distributed power source access capacity and nodes, and load baseline values.

[0021] Step 2: Construct a target planning model for the AC / DC hybrid distribution network that covers investment and construction, operation and maintenance, power production and reliability costs, and incorporates basic operational constraints such as power flow, node voltage, branch current, and radial operation. (1) Objective function ① Investment and construction costs

[0022] In the formula, This represents the total investment and construction cost, covering the investment and construction expenses for nodes and lines, etc. This indicates the investment cost of the communication node; Indicates the investment variables at the communication nodes; This represents the investment cost of a DC node; Indicates the DC node investment variable; This represents the investment cost of type A branch road; This represents the investment variable for type A branch roads, and is a binary variable. This indicates the investment cost of type B branch; Indicates the investment variables for type B branches; This indicates the investment cost of the C-type branch. Indicates the investment variable for type C branches; This represents the set of all branches in a power distribution network; It represents the set of all nodes in the distribution network.

[0023] ②Operating and maintenance costs

[0024] In the formula, This represents the total operating and maintenance cost; and These represent the maintenance costs of AC nodes and DC nodes, respectively. , and These represent the maintenance costs of branches A, B, and C, respectively. This represents the set of substation nodes.

[0025] ③ Electricity production costs

[0026] In the formula, This represents the total cost of electricity production; This represents the average power delivered by the substation to the system; This represents the average annual electricity price purchased from the public power grid; Represents a node The unit power generation cost of AC distributed power sources; Represents a node The unit power generation subsidy obtained by the distributed AC power generation; Represents a node The average power generation of distributed power sources; Represents a node The unit generation cost of a DC distributed power source; Represents a node The unit power generation subsidy obtained by the DC distributed power source; This represents the substation node in the entire node set; Represents the set of all nodes in an AC distributed power source; This represents the set of all nodes in a DC distributed power source.

[0027] ④ Reliability-related costs

[0028] In the formula, Indicates the value of EENS; Indicates the relevant electricity purchase price; and These represent the excitation functions corresponding to the reliability indices SAIFI and SAIDI, respectively. and These represent the SAIFI and SAIDI indices for the planning scheme, respectively.

[0029] (2) Constraints ① Current constraints

[0030]

[0031]

[0032] In the formula, This represents the impedance of a type A branch; Indicates a branch Current in; Represents a node The voltage amplitude; Indicates the impedance of a type B branch; , These are the modulation index and converter constant of the voltage source converter, respectively; This represents the impedance of a type C branch. Represents the node in the node association matrix Line number The elements corresponding to the column; M is a sufficiently large positive integer.

[0033] Because the connection between AC and DC nodes includes an converter, the balance of the converter needs to be considered. Therefore, the KCL equations between AC and DC nodes in the load node are different, and the expression is:

[0034]

[0035] In the formula, Represents a node AC load demand; Represents a node DC load demand; Represents a node The installed AC DG capacity; Represents a node The capacity of the installed DC DG; This indicates the commutation efficiency of the inverter; This indicates the commutation efficiency of the rectifier; Indicates the rated voltage of the AC node; Indicates the rated voltage of the DC node; This indicates the converter efficiency; This represents the current coefficient of branch b, which is taken as a value for type B lines. Otherwise, it is 1; Indicates the DC node investment variable; This represents the commutation constant of the voltage source converter; This represents the modulation coefficient of the voltage source converter.

[0036] ② Node voltage and branch current constraints

[0037]

[0038] In the formula, Represents a node Minimum permissible voltage value; Represents a node Maximum permissible voltage value; Represents a node The actual voltage value; Indicates a branch The absolute value of the current in the medium; Indicates a branch The maximum allowable current value.

[0039] ③Radial operational constraints

[0040] In the formula, Indicates the number of branch roads constructed; This represents the total number of nodes in the system. This indicates the total number of substation nodes in the system.

[0041] The third step is to analyze and derive the three core reliability indicators, EENS, SAIFI, and SAIDI, from the dual dimensions of line and node failures, and transform them into a mixed integer linear constraint embedded planning model to achieve explicit integration and deep coupling between reliability and planning models. ①EENS

[0042]

[0043]

[0044]

[0045] In the formula, This indicates that the battery level is below the expected value. Indicates the type of branch; Indicates by Type branch Energy loss caused by the fault; Represents nodes Energy loss caused by connected equipment failure; express Type branch The equivalent branch interruption duration; Indicates a branch Reduced power; When =1, it means that the branch has been selected; When =0, it means that the branch was not selected; This indicates the failure rate of the AC lines in the branch circuit; Indicates the repair time for the AC lines in the branch circuit; This indicates the failure rate of the AC circuit breakers in the branch circuit; Indicates the repair time of the AC circuit breaker in the branch circuit; This indicates the failure rate of the DC lines in the branch. Indicates the repair time of the DC line in the branch; This indicates the failure rate of the voltage source converter in the branch circuit; Indicates the repair time of the voltage source converter in the branch circuit; This indicates the failure rate of the DC circuit breaker in the branch circuit; Indicates the repair time of the DC circuit breaker in the branch circuit; Indicates a branch The power demand at the downstream load point, i.e., the branch The power required for the connected load to operate normally; This indicates the capacity of the distributed power source downstream of the branch, that is, the power that the distributed power source downstream of the branch can provide.

[0046]

[0047] In the formula, Represents a node The expected power supply was not provided due to a connection device malfunction. This indicates the inverter's failure rate; This indicates the average repair time of the inverter; Indicates the failure rate of the AC circuit breaker; This indicates the average repair time for AC circuit breakers. This indicates the failure rate of the DC circuit breaker; This indicates the average repair time of a DC circuit breaker; This indicates the failure rate of the rectifier; This indicates the average repair time of the rectifier; Represents a node The AC load demand at the location; Represents a node DC load demand at the location.

[0048] ②SAIFI

[0049]

[0050]

[0051]

[0052] In the formula, Indicates due to branch road Number of power outages per year caused by faults; Pointer node Number of power outages per year caused by faults; Represents a node Total number of users at the location; express Type branch The failure rate; Indicates a branch The number of users affected during the failure; Indicates a branch Total number of users at downstream load points; Indicates a branch b The total number of users affected by the power outage during a fault; It is a binary auxiliary variable.

[0053]

[0054] In the formula, Indicates due to node Number of power outages per year caused by faults; Represents a node The number of users connected to the communication interface; Represents a node The number of DC users connected at this location.

[0055] ③SAIDI

[0056]

[0057]

[0058] In the formula, SAIDI; Indicates due to branch road Total duration of power outages for users due to faults; Indicates due to node Total duration of power outages for users due to faults; Represents a node Total number of users at the location; express Type branch Equivalent fault repair time; Indicates a branch The number of users affected during the failure; Represents a node i The total number of DC users.

[0059] Step 4: Considering the uncertainty of distributed power output, the planning model established in the above steps is converted into a two-stage robust planning model. The output of distributed generation (DG) exhibits strong randomness due to the influence of natural environmental factors such as weather, necessitating the construction of a reasonable uncertainty set to describe its fluctuation range. This invention employs a budget-constrained uncertainty set to characterize the output fluctuations of ACDG and DCG, expressed as follows:

[0060] In the formula, For nodes The actual output of DG; , These represent the minimum and maximum output of the DG, respectively. The nominal value of the output of DG; , These are the maximum fluctuation ranges of ACDG and DCG, respectively. , These are the sets of installation nodes for ACDG and DCG, respectively. This is an uncertainty budget parameter used to adjust the conservatism of the model.

[0061] Considering the uncertainty of DG output, the ADHDS two-stage robust planning model is divided into a first stage (planning stage) and a second stage (operation stage). The first stage, before the uncertainty materializes, determines the node types and branch road construction schemes; the second stage, after the uncertainty materializes, adjusts operating parameters to cope with fluctuations and ensure reliable system operation. The mathematical form of the model is:

[0062] In the formula, These are the decision variables for the first stage. [ , , , , ], for The feasible domain; For the second stage decision variables, [ , , , , , , , , , , , , ], for The feasible domain; For variables with uncertain parameters, This is a budget-constrained uncertainty set (a convex bounded uncertainty set). Indicates the first phase of decision-making x The target coefficient vector; Indicates the second phase of decision-making y The target coefficient vector; Indicates the first stage of decision-making in the constraints. x The coefficient matrix; Indicates the second-stage decision in the constraints y The coefficient matrix; This represents the vector of terms on the right-hand side of the constraint.

[0063] Step 5: Use the machine learning-enhanced C&CG algorithm to decompose the robust model into a main problem and sub-problems for iterative solution; A neural network-enhanced C&CG algorithm is employed to transform the problem into a main problem and sub-problems for solution. The fitting and predictive capabilities of the neural network replace the difficult inner-layer optimization and blind scene search stages in the classic C&CG algorithm. The trained neural network... Based on the given first-stage decision and the uncertain outcome, it can accurately predict the optimal value of the second-stage problem:

[0064] The decomposed main problem makes optimization decisions based on the adverse scenarios of the subproblems, within a finite subset of the given scenario set. Inside, the main problem is reconstructed using the arg max operator, which selects a scenario such that after replacing the second-stage objective value with a neural network expression, the objective function takes the optimal value under the worst-case scenario.

[0065]

[0066] In the formula, This represents the worst possible scenario out of all scenarios.

[0067] Step 6: Use the commercial solver Gurobi to solve the mixed-integer linear programming problem established and transformed in the first five steps, and output the programming results; In summary, compared with existing methods, the present invention can achieve the following beneficial effects: (1) This invention analyzes and derives the three core reliability indicators from the dual dimensions of line and node faults, transforms them into mixed integer linear constraints and embeds them into the planning model. This breaks through the limitation of the separation of reliability assessment and planning model in traditional planning, realizes the deep coupling of reliability and planning decision, and can accurately quantify the power supply guarantee level in the planning stage, guiding the scheme to achieve the optimal balance between economy and reliability.

[0068] (2) The two-stage robust planning model established in this invention can effectively deal with the strong randomness of new energy output by using budget-constrained uncertainty set to characterize the uncertainty of distributed power output, ensuring the feasibility and robustness of planning schemes in extreme scenarios, while achieving flexible control of robustness and economy through conservative parameters.

[0069] (3) The present invention adopts the C&CG algorithm enhanced by machine learning, which replaces the difficult inner optimization and blind scene search links in the traditional algorithm with the fitting and prediction capabilities of the neural network. Under the premise of ensuring the accuracy of the optimal solution, the solution efficiency is increased by 60 to 80 times, which greatly shortens the planning scheme generation time and perfectly adapts to the engineering rapid planning scenario of AC and DC hybrid distribution network with multiple nodes and multiple branches.

[0070] To more clearly illustrate the technical solution and advantages of the present invention, the following will provide a more detailed description of this embodiment in conjunction with the accompanying drawings and an improved 54-node system as an example. The examples described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0071] The overall process flowchart provided in this example is as follows: Figure 1 As shown, the C&CG algorithm framework based on machine learning enhancement is explained as follows: Figure 2 As shown, taking the improved 54-node system as an example, the topology diagram is as follows: Figure 3 As shown, the steps in this example have been fully described above and will not be repeated here. The planned topology is as follows. Figure 4 As shown.

[0072] The comparison of this example method with the ordinary C&CG algorithm is shown in the table below.

[0073]

[0074] As can be easily seen from Table 1, compared with the ordinary C&CG algorithm, the machine learning-driven robust planning method proposed in this invention can bring better results and significantly improve the solution efficiency.

[0075] It should be noted that, depending on the implementation needs, the various steps / components described in this application can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0076] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A robust machine learning planning method for reliability-constrained AC / DC hybrid distribution networks, characterized in that, include: S1. Obtain and initialize the basic operating parameters of the AC / DC hybrid distribution network; The basic operating parameters serve as the input basis for subsequent model construction, constraint derivation, and solution calculation; S2. Based on the aforementioned basic operating parameters, construct a target planning model for the AC / DC hybrid distribution network, and add basic operating constraints for the distribution network to the target planning model; S3. The preset core reliability indicators of the distribution network are analytically derived from the dual dimensions of line faults and node faults, and then transformed into mixed integer linear constraints and embedded into the target planning model to obtain the reliability coupled planning model. S4. To address the uncertainty of distributed power output, a budget-constrained uncertainty set is constructed. This budget-constrained uncertainty set is used as the fluctuation constraint boundary of distributed power output, and the reliability coupling planning model is transformed into a two-stage robust planning model. S5. Employ a machine learning-enhanced column constraint generation algorithm to decompose the two-stage robust programming model into a main problem and sub-problems. The model is then solved and optimized through iterative interaction between the main problem and sub-problems until it converges. S6. Call the commercial solver to calculate the two-stage robust programming model after iterative convergence and output the optimal planning result of the AC / DC hybrid distribution network.

2. The robust planning method for machine learning in a reliability-constrained AC / DC hybrid distribution network according to claim 1, characterized in that, The basic operating parameters of AC / DC hybrid distribution networks include distribution network line parameters, distributed power generation capacity and node information, and load baseline values.

3. The robust planning method for machine learning in a reliability-constrained AC / DC hybrid distribution network according to claim 1, characterized in that, The objective function of the goal programming model is to minimize the overall cost, which includes investment and construction costs, operation and maintenance costs, power production costs, and reliability costs.

4. The robust machine learning planning method for reliability-constrained AC / DC hybrid distribution networks according to claim 1, characterized in that, The basic operating constraints of the distribution network include power flow constraints, node voltage constraints, branch current constraints, and radial operating constraints.

5. The robust planning method for machine learning in a reliability-constrained AC / DC hybrid distribution network according to claim 1, characterized in that, The preset core reliability indicators for the power distribution network include the expected power shortage value, the average power outage frequency index, and the average power outage duration index.

6. The robust machine learning planning method for reliability-constrained AC / DC hybrid distribution networks according to claim 1, characterized in that, The two-stage robust planning model includes a first-stage planning model and a second-stage operation model. The first-stage planning model is used to determine the distribution network node types and branch construction schemes before the uncertainty is realized, while the second-stage operation model is used to adjust the distribution network operation parameters to cope with the output fluctuations of distributed power sources after the uncertainty is realized.

7. The robust machine learning planning method for reliability-constrained AC / DC hybrid distribution networks according to claim 6, characterized in that, The machine learning-enhanced column constraint generation algorithm employs a pre-trained neural network, which is used to predict the optimal value of the running model in the second stage based on the decision variables and uncertainty parameters in the first stage.

8. The robust machine learning planning method for reliability-constrained AC / DC hybrid distribution networks according to claim 1, characterized in that, The decomposed main problem is used to make distribution network planning and optimization decisions based on the adverse scenarios output by the sub-problems. The decomposed sub-problems are used to solve the worst-case scenario of distributed power generation output and feed it back to the main problem to complete iterative optimization.

9. The robust machine learning planning method for reliability-constrained AC / DC hybrid distribution networks according to claim 1, characterized in that, The commercial solver used is the Gurobi solver.

10. A robust machine learning programming system for reliability-constrained AC / DC hybrid distribution networks, characterized in that, include: The data initialization module is used to acquire and initialize the basic operating parameters of the AC / DC hybrid distribution network; The basic operating parameters serve as the input basis for subsequent model construction, constraint derivation, and solution calculation; The basic model building module is used to construct a target planning model for the AC / DC hybrid distribution network based on the basic operating parameters, and to add basic operating constraints of the distribution network to the target planning model. The reliability coupling module is used to analyze and derive the preset core reliability indicators of the distribution network from the dual dimensions of line faults and node faults, transform them into mixed integer linear constraints, and then embed them into the target planning model to obtain the reliability coupling planning model. The robust model construction module is used to construct a budget-constrained uncertainty set for the output uncertainty of distributed power sources. The budget-constrained uncertainty set is used as the fluctuation constraint boundary of the output of distributed power sources to transform the reliability coupled planning model into a two-stage robust planning model. The model iterative solution module is used to decompose the two-stage robust programming model into a main problem and sub-problems using a machine learning-enhanced column constraint generation algorithm. The solution optimization is completed through iterative interaction between the main problem and sub-problems until the model converges. The results output module is used to call a commercial solver to calculate the two-stage robust programming model after iterative convergence and output the optimal planning results for the AC / DC hybrid distribution network.

Images