Power distribution district collaborative planning method and system based on neural network agent model

Abstract

The application discloses a power distribution area collaborative planning method and system based on a neural network agent model, and the method comprises the following steps: constructing a double-layer optimization model of flexible interconnection and energy storage collaborative planning of a power distribution area; randomly generating a candidate site and capacity scheme satisfying the constraint conditions of the double-layer optimization model, and performing optimization calculation of a running model on each candidate site and capacity scheme to obtain a real total cost of the whole life cycle corresponding to each candidate scheme, thereby forming a training sample set; performing supervised training on a preset neural network agent model; generating and iteratively updating a site and capacity scheme, evaluating the predicted total cost of the site and capacity scheme by using the neural network agent model, optimizing the site and capacity scheme according to the predicted total cost, and outputting a final target site and capacity scheme and a corresponding collaborative running strategy until a convergence condition is satisfied. The method effectively solves the problems of disconnection of 'planning-construction-operation', low calculation efficiency and the like in traditional planning, and realizes collaborative improvement of safety and solution efficiency.

Description

Technical Field

[0001] This invention belongs to the field of power system planning technology, and in particular relates to a method and system for collaborative planning of distribution substations based on a neural network proxy model. Background Technology

[0002] Traditional AC distribution network-based transformer substation planning and operation modes face severe challenges: First, the uncontrollable unidirectional power flow results in limited power exchange capacity between adjacent substations, easily leading to localized heavy overloads and voltage exceeding limits; second, the local absorption capacity of renewable energy is insufficient, and peak photovoltaic power generation can easily cause voltage rises and power backflow; third, there are bottlenecks in improving power supply reliability, as loads in fault-affected areas cannot be quickly transferred, relying on manual operation such as generator trucks, resulting in slow response speeds and low power supply reliability.

[0003] To address these challenges, employing flexible DC interconnection devices with rapid and flexible power flow control capabilities to flexibly interconnect multiple adjacent AC distribution areas, forming a low-voltage flexible DC interconnection system, or deploying distributed energy storage on the low-voltage side of the distribution area, has become an important direction for technological development. However, current research and practice mainly focus on the operation and control level of the system, while there are significant gaps and deficiencies in the early planning stage: First, existing planning methods mostly use deterministic scenarios or simple time-series production simulations, failing to fully consider the coupled effects of multiple uncertainties among the source, load, and storage, resulting in poor robustness of the planning scheme; second, the planning process fails to be deeply coupled with the refined operation strategies of the flexible DC interconnection devices (such as switching between multiple control modes and loss characteristics), causing a disconnect between the three stages of "planning, construction, and operation," resulting in low investment efficiency; finally, there is a lack of a fully automated planning tool covering the entire process from global optimization to equipment selection, relying on manual experience, which is inefficient and makes it difficult to find the global or near-global optimal solution.

[0004] Therefore, there is an urgent need to propose an intelligent planning method that can deeply integrate planning and operation and significantly improve solution efficiency, so as to output a collaborative planning scheme with optimal economy, security and flexibility. Summary of the Invention

[0005] This invention aims to address the problems of poor scenario adaptability, disconnect between planning and operation, high computational costs due to complex models, and low decision-making efficiency in existing distribution network planning methods when facing high proportions of renewable energy integration, particularly in the planning of inter-distribution network flexibility and energy storage coordination. This invention provides a distribution network inter-distribution network collaborative planning method and system based on a neural network surrogate model. It can output economical, robust, and executable construction schemes for flexible DC interconnection and energy storage systems. Specifically, it includes the optimal placement (location), capacity (power / capacity) schemes for flexible DC interconnection devices and energy storage systems, as well as matching collaborative operation strategies, thereby maximizing the overall carrying capacity, renewable energy absorption rate, power supply reliability, and life-cycle economic efficiency of the regional distribution network.

[0006] In a first aspect, the present invention provides a distribution area collaborative planning method based on a neural network agent model, comprising:

[0007] Acquire historical operation monitoring data and power grid structure parameters of each distribution substation within the target area;

[0008] Based on the historical operation monitoring data and the power grid structure parameters, a two-layer optimization model for the coordinated planning of flexible interconnection and energy storage in distribution substations is constructed. The upper layer of the two-layer optimization model is the planning model, and the decision variables of the planning model are the site selection and capacity setting schemes of the flexible interconnection device and the energy storage system. The optimization objective is to minimize the annualized investment cost.

[0009] The lower layer of the two-layer optimization model is the operation model. The operation model takes the scheme decided by the upper layer as input, the decision variables are the operation status of each device in multiple typical time-series scenarios, and the optimization objective is to minimize the total operating cost of the system in all scenarios.

[0010] Candidate site selection and sizing schemes that satisfy the constraints of the planning model are randomly generated, and the optimization calculation of the running model is performed on each candidate site selection and sizing scheme to obtain the true total life cycle cost of each candidate scheme, which constitutes a training sample set.

[0011] Using the training sample set, a pre-defined neural network proxy model is trained under supervision, enabling the neural network proxy model to learn the mapping relationship from candidate addressing and sizing schemes to the true total cost of candidate addressing and sizing schemes;

[0012] An optimization algorithm is used to generate and iteratively update the addressing and sizing scheme. The predicted total cost of the addressing and sizing scheme is evaluated using the neural network surrogate model. The addressing and sizing scheme is optimized based on the predicted total cost until the convergence condition is met. Finally, the target addressing and sizing scheme and the corresponding cooperative operation strategy are output.

[0013] Secondly, the present invention provides a distribution area collaborative planning system based on a neural network agent model, comprising:

[0014] The acquisition module is configured to acquire historical operation monitoring data and power grid structure parameters of each distribution substation within the target area;

[0015] The construction module is configured to build a two-layer optimization model for the coordinated planning of flexible interconnection and energy storage in distribution substations based on the historical operation monitoring data and the power grid structure parameters. The upper layer of the two-layer optimization model is the planning model, and the decision variables of the planning model are the site selection and capacity setting scheme of the flexible interconnection device and the energy storage system. The optimization objective is to minimize the annualized investment cost.

[0016] The lower layer of the two-layer optimization model is the operation model. The operation model takes the scheme decided by the upper layer as input, the decision variables are the operation status of each device in multiple typical time-series scenarios, and the optimization objective is to minimize the total operating cost of the system in all scenarios.

[0017] The calculation module is configured to randomly generate candidate site selection and capacity determination schemes that satisfy the constraints of the planning model, and perform optimization calculations of the running model for each candidate site selection and capacity determination scheme to obtain the true total lifecycle cost of each candidate scheme, thus forming a training sample set.

[0018] The training module is configured to use the training sample set to perform supervised training on a preset neural network proxy model, so that the neural network proxy model learns the mapping relationship from the candidate addressing and sizing scheme to the true total cost of the candidate addressing and sizing scheme;

[0019] The output module is configured to generate and iteratively update the addressing and sizing scheme using an optimization algorithm, evaluate the predicted total cost of the addressing and sizing scheme using the neural network surrogate model, optimize the addressing and sizing scheme based on the predicted total cost until the convergence condition is met, and output the final target addressing and sizing scheme and the corresponding cooperative operation strategy.

[0020] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the distribution area collaborative planning method based on a neural network agent model according to any embodiment of the present invention.

[0021] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the distribution area collaborative planning method based on a neural network agent model according to any embodiment of the present invention.

[0022] This application presents a method and system for collaborative planning of distribution substations based on a neural network surrogate model. First, it collects source-load and topology data of the distribution substations and establishes a two-layer optimization model that includes upper-level investment decisions and lower-level multi-scenario operation simulations. To overcome the problem of high computational cost when directly solving the two-layer optimization model, a neural network surrogate model is innovatively trained. By learning from a large number of "planning scheme-simulation cost" samples, it achieves millisecond-level rapid prediction of the total cost of new schemes. Then, combined with intelligent optimization algorithms, the neural network surrogate model efficiently searches for the optimal site selection and capacity determination scheme for flexible interconnection devices and energy storage systems. The accuracy and adaptability of the scheme are ensured through a closed-loop process of "optimization-verification-online update." This invention effectively solves the problems of disconnect between "planning-construction-operation" and low computational efficiency in traditional planning, achieving a synergistic improvement in security and solution efficiency. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating a distribution area collaborative planning method based on a neural network agent model, provided in an embodiment of the present invention;

[0025] Figure 2 This is a structural block diagram of a distribution area collaborative planning system based on a neural network agent model, provided in an embodiment of the present invention.

[0026] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see Figure 1 The diagram shows a flowchart of a distribution area collaborative planning method based on a neural network agent model.

[0029] like Figure 1As shown, the distribution area collaborative planning method based on the neural network agent model specifically includes the following steps:

[0030] Step S101: Obtain historical operation monitoring data and power grid structure parameters of each distribution substation within the target area.

[0031] In this step, historical operation monitoring data and power grid structure parameters of each distribution substation within the target planning area are acquired. The historical operation monitoring data includes at least historical load data and photovoltaic output data, and the power grid structure parameters include at least topology connections and line impedance parameters. The acquired electrical quantity data is then standardized to facilitate subsequent model training and calculation.

[0032] Step S102: Based on the historical operation monitoring data and the power grid structure parameters, construct a two-layer optimization model for the coordinated planning of flexible interconnection of distribution substations and energy storage.

[0033] In this step, the upper layer of the two-layer optimization model is the planning model. The decision variables of the planning model are the site selection and capacity setting scheme of the flexible interconnection device and energy storage system, and the optimization objective is to minimize the annualized investment cost.

[0034] Specifically, the objective function of the planning model is:

[0035] ,

[0036] In the formula, The total investment cost, The unit capacity cost of flexible DC interconnect devices, For the set of all nodes in the system, This is a binary decision variable, indicating whether a flexible vertical coupling device should be installed at node i. The rated power of the flexible DC device at node i. The fixed installation cost of a single flexible DC interconnect device, For the discount rate, For equipment planning cycle, The unit energy cost of an energy storage system This is a binary decision variable, indicating whether energy storage should be installed at node i. Install the rated capacity of energy storage for node i. The unit power cost of the energy storage system, The rated power of the energy storage installed at node i, For the newly built line collection, The investment cost per unit length of newly added line. This refers to the length of the newly constructed line.

[0037] The constraints of the planning model include:

[0038] The constraint relating equipment installation status to capacity upper and lower limits is expressed as follows:

[0039] ,

[0040] In the formula, This represents the lower limit of the capacity of flexible DC interconnect devices that can be installed at node i. The upper limit of the capacity of the flexible DC interconnect device that can be installed at node i;

[0041] The ratio constraint between rated power and rated energy of energy storage is expressed as follows:

[0042] ,

[0043] ,

[0044] In the formula, This represents the lower limit of the energy capacity of the energy storage system that can be installed at node i. Let i be the upper limit of the energy capacity of the energy storage system that can be installed at node i. This represents the maximum permissible power-to-energy ratio for the energy storage system.

[0045] The maximum total number of flexible interconnect devices that can be installed is constrained by the following expression:

[0046] ,

[0047] In the formula, The maximum total number of flexible DC interconnect devices that can be installed;

[0048] The lower layer of the two-layer optimization model is the operation model. The operation model takes the scheme decided by the upper layer as input, the decision variables are the operating status of each device in multiple typical time-series scenarios, and the optimization objective is to minimize the total operating cost of the system in all scenarios.

[0049] Specifically, the objective function of the running model is:

[0050] ,

[0051] In the formula, This is a collection of all typical scenarios. For equipment planning cycle, The price at which electricity is purchased from the upper-level power grid at time t. Let be the active power purchased from the upstream power grid at time t under scenario s. This is the unit network loss cost coefficient. The total number of branch roads, For branch resistance, Let be the current amplitude flowing through branch l at time t in scenario s. These are the weighting coefficients. To represent the exceedance of various safety constraints at time t in scenario s. For time step.

[0052] The constraints for running the model include:

[0053] The active and reactive power balance constraint at a node is expressed as follows:

[0054] ,

[0055] In the formula, Let be the active power flowing from node j to its downstream node k at time t in scenario s. Let k be the set of all k nodes from node j to node k. Let j be the set of all nodes j from node i to node j. Let be the current flowing from node i to node j at time t in scenario s. Let be the active power flowing from node i to node j at time t in scenario s. branch road The resistance, For the photovoltaic power prediction at node j at time t in scenario s, This represents the predicted load value for node j at time t in scenario s. The active power of the AC network is injected into the flexible DC device at node j at time t under scenario s. Let be the net charging and discharging power of the energy storage system at node j at time t under scenario s;

[0056] The branch power flow and voltage drop constraints based on second-order cone relaxation are expressed as follows:

[0057] ,

[0058] In the formula, Let be the reactive power flowing from node j to its downstream node k at time t in scenario s. Let be the reactive power flowing from node i to its downstream node j at time t in scenario s. branch road Reactance, Let the reactive load of node j at time t be given in scenario s. Inject reactive power from the AC network into the flexible DC device at node j at time t under scenario s;

[0059] The safe operating upper and lower limit constraints for node voltage amplitude and branch current amplitude are expressed as follows:

[0060] ,

[0061] ,

[0062] ,

[0063] ,

[0064] In the formula, Let be the voltage amplitude at node j at time t in scenario s. Let i be the voltage amplitude at node i at time t in scenario s. This represents the lower limit of the allowable node voltage. This represents the upper limit of the allowed node voltage;

[0065] The active and reactive power output limits and capacity circle constraints of flexible interconnected devices are expressed as follows:

[0066] ,

[0067] ,

[0068] In the formula, The rated apparent power capacity of the flexible DC device at node i;

[0069] The charging and discharging power, dynamic state of charge, and periodic operating constraints of the energy storage system are expressed as follows:

[0070] ,

[0071] ,

[0072] ,

[0073] ,

[0074] ,

[0075] ,

[0076] In the formula, Let i be the charging power of energy storage at time t in scenario s. for, for, Let i be the discharge power of stored energy i at time t in scenario s. This is a binary variable representing whether energy storage i is in a charging state at time t in scenario s. The rated power capacity for energy storage to be installed at node i, Let i be the net power of energy storage at time t in scenario s. Let i be the state of charge of energy storage at time t in scenario s. Let i be the state of charge of energy storage at time t-1 in scenario s. For energy storage charging efficiency, For energy storage and discharge efficiency, The rated energy capacity for installing energy storage at node i, For time step, This is the lower limit of the allowable state of charge for energy storage. This represents the upper limit of the allowable state of charge for energy storage. Let i be the initial state of charge of energy storage i in scenario s. Let i be the state of charge of energy storage i at the last moment in scenario s.

[0077] Step S103: Randomly generate candidate site selection and capacity determination schemes that satisfy the constraints of the planning model, and perform optimization calculations of the running model on each candidate site selection and capacity determination scheme to obtain the true total lifecycle cost of each candidate scheme, forming a training sample set.

[0078] In this step, the physical topology of the distribution network is modeled as a graph. , For the set of vertices, Let it be the set of edges;

[0079] Graph attention networks are used to encode and aggregate node and edge features, and global average pooling is used to obtain graph-level feature vectors. The node features include at least historical average load and photovoltaic installed capacity, and the edge features include at least line resistance and reactance. The graph-level feature vector... The expression is:

[0080] ,

[0081] In the formula, It is the mean pooling function. To enhance the features of subsequent nodes, Let i be the set of nodes i. Let i be the feature vector of node i;

[0082] The graph-level feature vector The eigenvectors of planning variables obtained by flattening candidate site selection and sizing schemes and external condition eigenvectors The features are concatenated to form a comprehensive feature vector. The external condition feature vector It includes at least the statistical values ​​of total annual photovoltaic power generation and total annual load. The expression for the comprehensive feature vector is:

[0083] ,

[0084] In the formula, This is the composite feature vector obtained by concatenating the various feature vectors;

[0085] The comprehensive feature vector Input a multilayer perceptron, perform forward propagation calculations, and output a predicted value of the total cost of the planning scheme. The expression is:

[0086] ,

[0087] In the formula, The weights of the linear rectified function, The weights of the integrated feature vectors, The bias is used to integrate the feature vectors. For the bias of the predicted value, It is a linear rectified function;

[0088] The neural network proxy model is trained using the training sample set and a composite loss function combining the main task and auxiliary task until the prediction accuracy of the neural network proxy model reaches the preset requirement.

[0089] It should be noted that the expression for the composite loss function is:

[0090] ,

[0091] In the formula, For compound loss, The weight of the mean squared error of total cost. The mean square error of the total cost. This is the total cost forecast. This is the actual value of the total cost. This is the weight of the mean square error of the new energy consumption rate. The mean square error of the renewable energy consumption rate. This is the predicted value for the renewable energy absorption rate. This represents the actual value of the renewable energy absorption rate. This represents the weight of the mean square error of the average voltage deviation. This represents the mean square error of the average voltage deviation. This is the predicted value of the average voltage deviation. This represents the actual value of the average voltage deviation.

[0092] Step S104: Using the training sample set, supervised training is performed on the preset neural network proxy model, so that the neural network proxy model learns the mapping relationship from the candidate addressing and sizing scheme to the true total cost of the candidate addressing and sizing scheme.

[0093] Step S105: Use an optimization algorithm to generate and iteratively update the addressing and sizing scheme. Use the neural network surrogate model to evaluate the predicted total cost of the addressing and sizing scheme. Optimize the addressing and sizing scheme based on the predicted total cost until the convergence condition is met. Output the final target addressing and sizing scheme and the corresponding cooperative operation strategy.

[0094] In this step, a set of candidate site selection and gradation schemes that satisfy the basic constraints of the planning model are generated randomly to form the initial population;

[0095] Input the feature vector corresponding to each candidate site selection and capacity quantification scheme in the current population into the trained neural network proxy model to obtain the total cost prediction value of each candidate site selection and capacity quantification scheme;

[0096] Using the total cost prediction value as the fitness, a genetic algorithm is used to perform selection, crossover, and mutation operations on the current population to generate a new generation of candidate schemes.

[0097] Repeat the iteration until the preset maximum number of iterations is reached to obtain the optimal candidate solution;

[0098] The optimal candidate solution is taken as input, and the running model is called to perform optimization calculations to obtain the true total cost and running strategy of the optimal candidate solution.

[0099] If the relative error between the actual total cost and the predicted total cost is within the allowable range, the optimal candidate solution and operating strategy are output as the final planning result. Otherwise, the optimal candidate solution and the corresponding actual total cost are used as new training samples to update the neural network proxy model, and the optimization process is restarted.

[0100] In summary, the method of this application establishes a two-layer stochastic optimization model that deeply couples upper-layer planning (site selection and capacity determination) with lower-layer operation (multi-scenario scheduling). This model jointly optimizes long-term decision-making and short-term refined operation strategies, ensuring that the final equipment configuration scheme (how to install the maximum capacity of flexible DC-DC devices and energy storage) and the future optimal operation mode (how to adjust power according to real-time electricity prices and source-load fluctuations) achieve a globally optimal match. Compared with traditional methods, the planning scheme generated by this invention is not only more economical in initial investment, but also can adaptively cope with multiple uncertainties such as the randomness of photovoltaic output and load fluctuations throughout the entire life cycle. This significantly improves the renewable energy absorption capacity, voltage quality, and power supply reliability, fundamentally ensuring the effectiveness of investment and the long-term adaptability of the scheme.

[0101] A proxy model integrating graph neural networks and multilayer perceptrons was trained. By learning from a massive number of "planning scheme-simulation cost" sample pairs, it accurately captures the complex mapping relationship between power grid topology, equipment configuration, and total lifecycle cost. During the optimization process, the model can replace the precise simulation that originally required hours of computation within milliseconds, enabling real-time evaluation of the merits of candidate schemes. This makes it possible to use intelligent optimization techniques such as genetic algorithms to perform efficient and global search of the huge solution space, shortening the planning decision time from the traditional "days" to "hours", and providing key technical support for online rolling planning and large-scale scenario analysis.

[0102] Please see Figure 2 The diagram shows a structural block diagram of a distribution area collaborative planning system based on a neural network agent model according to this application.

[0103] like Figure 2 As shown, the distribution area collaborative planning system 200 includes an acquisition module 210, a construction module 220, a calculation module 230, a training module 240, and an output module 250.

[0104] The acquisition module 210 is configured to acquire historical operation monitoring data and power grid structure parameters of each distribution substation within the target area; the construction module 220 is configured to construct a two-layer optimization model for flexible interconnection and energy storage collaborative planning of distribution substations based on the historical operation monitoring data and the power grid structure parameters. The upper layer of the two-layer optimization model is a planning model, where the decision variables are the site selection and capacity determination schemes of the flexible interconnection devices and energy storage systems, and the optimization objective is to minimize the annualized investment cost; the lower layer of the two-layer optimization model is an operation model, where the operation model takes the schemes decided by the upper layer as input, the decision variables are the operating states of each device under multiple typical time-series scenarios, and the optimization objective is to minimize the total operating cost of the system under all scenarios; the calculation module 230 is configured to randomly generate a calculation module that satisfies the specified conditions. The system generates candidate location and capacity sizing schemes based on the constraints of the planning model, and performs optimization calculations on each candidate scheme using the running model to obtain the true total lifecycle cost of each candidate scheme, forming a training sample set. A training module 240 is configured to use the training sample set to supervise the training of a preset neural network proxy model, enabling the neural network proxy model to learn the mapping relationship from candidate location and capacity sizing schemes to the true total cost of the candidate location and capacity sizing schemes. An output module 250 is configured to generate and iteratively update location and capacity sizing schemes using an optimization algorithm, evaluate the predicted total cost of the location and capacity sizing schemes using the neural network proxy model, optimize the location and capacity sizing schemes based on the predicted total cost, until convergence conditions are met, and output the final target location and capacity sizing scheme and the corresponding collaborative operation strategy.

[0105] It should be understood that Figure 2The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0106] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the distribution area collaborative planning method based on a neural network proxy model in any of the above method embodiments.

[0107] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0108] Acquire historical operation monitoring data and power grid structure parameters of each distribution substation within the target area;

[0109] Based on the historical operation monitoring data and the power grid structure parameters, a two-layer optimization model for the coordinated planning of flexible interconnection and energy storage in distribution substations is constructed. The upper layer of the two-layer optimization model is the planning model, and the decision variables of the planning model are the site selection and capacity setting schemes of the flexible interconnection device and the energy storage system. The optimization objective is to minimize the annualized investment cost.

[0110] The lower layer of the two-layer optimization model is the operation model. The operation model takes the scheme decided by the upper layer as input, the decision variables are the operation status of each device in multiple typical time-series scenarios, and the optimization objective is to minimize the total operating cost of the system in all scenarios.

[0111] Candidate site selection and sizing schemes that satisfy the constraints of the planning model are randomly generated, and the optimization calculation of the running model is performed on each candidate site selection and sizing scheme to obtain the true total life cycle cost of each candidate scheme, which constitutes a training sample set.

[0112] Using the training sample set, a pre-defined neural network proxy model is trained under supervision, enabling the neural network proxy model to learn the mapping relationship from candidate addressing and sizing schemes to the true total cost of candidate addressing and sizing schemes;

[0113] An optimization algorithm is used to generate and iteratively update the addressing and sizing scheme. The predicted total cost of the addressing and sizing scheme is evaluated using the neural network surrogate model. The addressing and sizing scheme is optimized based on the predicted total cost until the convergence condition is met. Finally, the target addressing and sizing scheme and the corresponding cooperative operation strategy are output.

[0114] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the distribution area collaborative planning system based on a neural network agent model, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely located relative to a processor, which can be connected to the distribution area collaborative planning system based on a neural network agent model via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0115] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3 Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the distribution area collaborative planning method based on a neural network surrogate model as described in the above method embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the distribution area collaborative planning system based on the neural network surrogate model. The output device 340 may include a display screen or other display device.

[0116] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0117] In one implementation, the above-described electronic device is applied to a distribution area collaborative planning system based on a neural network agent model, serving as a client, and includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0118] Acquire historical operation monitoring data and power grid structure parameters of each distribution substation within the target area;

[0119] Based on the historical operation monitoring data and the power grid structure parameters, a two-layer optimization model for the coordinated planning of flexible interconnection and energy storage in distribution substations is constructed. The upper layer of the two-layer optimization model is the planning model, and the decision variables of the planning model are the site selection and capacity setting schemes of the flexible interconnection device and the energy storage system. The optimization objective is to minimize the annualized investment cost.

[0120] The lower layer of the two-layer optimization model is the operation model. The operation model takes the scheme decided by the upper layer as input, the decision variables are the operation status of each device in multiple typical time-series scenarios, and the optimization objective is to minimize the total operating cost of the system in all scenarios.

[0121] Candidate site selection and sizing schemes that satisfy the constraints of the planning model are randomly generated, and the optimization calculation of the running model is performed on each candidate site selection and sizing scheme to obtain the true total life cycle cost of each candidate scheme, which constitutes a training sample set.

[0122] Using the training sample set, a pre-defined neural network proxy model is trained under supervision, enabling the neural network proxy model to learn the mapping relationship from candidate addressing and sizing schemes to the true total cost of candidate addressing and sizing schemes;

[0123] An optimization algorithm is used to generate and iteratively update the addressing and sizing scheme. The predicted total cost of the addressing and sizing scheme is evaluated using the neural network surrogate model. The addressing and sizing scheme is optimized based on the predicted total cost until the convergence condition is met. Finally, the target addressing and sizing scheme and the corresponding cooperative operation strategy are output.

[0124] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for collaborative planning of distribution radio areas based on a neural network surrogate model, characterized in that, include: Acquire historical operation monitoring data and power grid structure parameters of each distribution substation within the target area; Based on the historical operation monitoring data and the power grid structure parameters, a two-layer optimization model for the coordinated planning of flexible interconnection and energy storage in distribution substations is constructed. The upper layer of the two-layer optimization model is the planning model, and the decision variables of the planning model are the site selection and capacity setting schemes of the flexible interconnection device and the energy storage system. The optimization objective is to minimize the annualized investment cost. The lower layer of the two-layer optimization model is the operation model. The operation model takes the scheme decided by the upper layer as input, the decision variables are the operation status of each device in multiple typical time-series scenarios, and the optimization objective is to minimize the total operating cost of the system in all scenarios. Candidate site selection and sizing schemes that satisfy the constraints of the planning model are randomly generated, and the optimization calculation of the running model is performed on each candidate site selection and sizing scheme to obtain the true total life cycle cost of each candidate scheme, which constitutes a training sample set. Using the training sample set, a pre-defined neural network surrogate model is subjected to supervised training, enabling the neural network surrogate model to learn the mapping relationship from candidate site selection and capacity sizing schemes to the true total cost of candidate site selection and capacity sizing schemes. Specifically, this includes: modeling the physical topology of the distribution network as a graph. , For the set of vertices, Let it be the set of edges; Graph attention networks are used to encode and aggregate node and edge features, and global average pooling is used to obtain graph-level feature vectors. The node features include at least historical average load and photovoltaic installed capacity, and the edge features include at least line resistance and reactance. The graph-level feature vector... The expression is: ， In the formula, It is the mean pooling function. To enhance the features of subsequent nodes, Let i be the set of nodes i. Let i be the feature vector of node i; The graph-level feature vector The eigenvectors of planning variables obtained by flattening candidate site selection and sizing schemes and external condition eigenvectors The features are concatenated to form a comprehensive feature vector. The external condition feature vector It includes at least the statistical values ​​of total annual photovoltaic power generation and total annual load. The expression for the comprehensive feature vector is: ， In the formula, This is the composite feature vector obtained by concatenating the various feature vectors; The comprehensive feature vector Input a multilayer perceptron, perform forward propagation calculations, and output a predicted value of the total cost of the planning scheme. The expression is: ， In the formula, The weights of the linear rectified function, The weights of the integrated feature vectors, The bias is used to integrate the feature vectors. For the bias of the predicted value, It is a linear rectified function; Using the training sample set, the neural network surrogate model is trained with a composite loss function combining the main task and auxiliary task until the prediction accuracy of the neural network surrogate model reaches a preset requirement. The expression for the composite loss function is: ， In the formula, For compound loss, The weight of the mean squared error of total cost. The mean square error of the total cost. This is the total cost forecast. This is the actual value of the total cost. This is the weight of the mean square error of the new energy consumption rate. The mean square error of the renewable energy consumption rate. This is the predicted value for the renewable energy absorption rate. This represents the actual value of the renewable energy absorption rate. This represents the weight of the mean square error of the average voltage deviation. This represents the mean square error of the average voltage deviation. This is the predicted value of the average voltage deviation. This represents the actual value of the average voltage deviation. An optimization algorithm is used to generate and iteratively update the addressing and sizing scheme. The predicted total cost of the addressing and sizing scheme is evaluated using the neural network surrogate model. The addressing and sizing scheme is optimized based on the predicted total cost until the convergence condition is met. Finally, the target addressing and sizing scheme and the corresponding cooperative operation strategy are output.

2. The distribution area collaborative planning method based on a neural network surrogate model according to claim 1, characterized in that, The constraints of the planning model include: the correlation between equipment installation status and capacity upper and lower limits, the ratio of rated power of energy storage to rated energy, and the maximum total number of flexible interconnection devices. The objective function of the planning model is: ， In the formula, The total investment cost, The unit capacity cost of flexible DC interconnect devices, For the set of all nodes in the system, This is a binary decision variable, indicating whether a flexible vertical coupling device should be installed at node i. The rated power of the flexible DC device at node i. The fixed installation cost of a single flexible DC interconnect device, For the discount rate, For equipment planning cycle, The unit energy cost of an energy storage system This is a binary decision variable, indicating whether energy storage should be installed at node i. Install the rated capacity of energy storage for node i. The unit power cost of the energy storage system, The rated power of the energy storage installed at node i, For the newly built line collection, The investment cost per unit length of newly added line. This refers to the length of the newly constructed line.

3. The distribution area collaborative planning method based on a neural network surrogate model according to claim 1, characterized in that, The constraints of the operating model include: node active and reactive power balance constraints, branch power flow and voltage drop constraints based on second-order cone relaxation, upper and lower limits constraints for safe operation of node voltage amplitude and branch current amplitude, active and reactive power output limits and capacity circle constraints of flexible interconnection devices, charging and discharging power of energy storage systems, and dynamic and periodic operation constraints of state of charge. The objective function of the operating model is: ， In the formula, This is a collection of all typical scenarios. For equipment planning cycle, The price at which electricity is purchased from the upper-level power grid at time t. Let be the active power purchased from the upstream power grid at time t under scenario s. This is the unit network loss cost coefficient. The total number of branch roads, For branch resistance, Let be the current amplitude flowing through branch l at time t in scenario s. These are the weighting coefficients. To represent the exceedance of various safety constraints at time t in scenario s. For time step.

4. The distribution area collaborative planning method based on a neural network surrogate model according to claim 1, characterized in that, The process of generating and iteratively updating the addressing and sizing scheme using an optimization algorithm, evaluating the predicted total cost of the addressing and sizing scheme using the neural network surrogate model, and optimizing the addressing and sizing scheme based on the predicted total cost includes: A set of candidate site selection and gradation schemes that satisfy the basic constraints of the planning model are generated randomly to form an initial population; Input the feature vector corresponding to each candidate site selection and capacity quantification scheme in the current population into the trained neural network proxy model to obtain the total cost prediction value of each candidate site selection and capacity quantification scheme; Using the total cost prediction value as the fitness, a genetic algorithm is used to perform selection, crossover, and mutation operations on the current population to generate a new generation of candidate schemes. Repeat the iteration until the preset maximum number of iterations is reached to obtain the optimal candidate solution; The optimal candidate solution is taken as input, and the running model is called to perform optimization calculations to obtain the true total cost and running strategy of the optimal candidate solution. If the relative error between the actual total cost and the predicted total cost is within the allowable range, the optimal candidate solution and operating strategy are output as the final planning result. Otherwise, the optimal candidate solution and the corresponding actual total cost are used as new training samples to update the neural network proxy model, and the optimization process is restarted.

5. A distribution area collaborative planning system based on a neural network surrogate model, characterized in that, include: The acquisition module is configured to acquire historical operation monitoring data and power grid structure parameters of each distribution substation within the target area; The construction module is configured to build a two-layer optimization model for the coordinated planning of flexible interconnection and energy storage in distribution substations based on the historical operation monitoring data and the power grid structure parameters. The upper layer of the two-layer optimization model is the planning model, and the decision variables of the planning model are the site selection and capacity setting scheme of the flexible interconnection device and the energy storage system. The optimization objective is to minimize the annualized investment cost. The lower layer of the two-layer optimization model is the operation model. The operation model takes the scheme decided by the upper layer as input, the decision variables are the operation status of each device in multiple typical time-series scenarios, and the optimization objective is to minimize the total operating cost of the system in all scenarios. The calculation module is configured to randomly generate candidate site selection and capacity determination schemes that satisfy the constraints of the planning model, and perform optimization calculations of the running model for each candidate site selection and capacity determination scheme to obtain the true total lifecycle cost of each candidate scheme, thus forming a training sample set. The training module is configured to use the training sample set to perform supervised training on a preset neural network proxy model, enabling the neural network proxy model to learn the mapping relationship from candidate site selection and capacity sizing schemes to the true total cost of candidate site selection and capacity sizing schemes. Specifically, this includes: modeling the physical topology of the distribution network as a graph. , For the set of vertices, Let it be the set of edges; Graph attention networks are used to encode and aggregate node and edge features, and global average pooling is used to obtain graph-level feature vectors. The node features include at least historical average load and photovoltaic installed capacity, and the edge features include at least line resistance and reactance. The graph-level feature vector... The expression is: ， In the formula, It is the mean pooling function. To enhance the features of subsequent nodes, Let i be the set of nodes i. Let i be the feature vector of node i; The graph-level feature vector The eigenvectors of planning variables obtained by flattening candidate site selection and sizing schemes and external condition eigenvectors The features are concatenated to form a comprehensive feature vector. The external condition feature vector It includes at least the statistical values ​​of total annual photovoltaic power generation and total annual load. The expression for the comprehensive feature vector is: ， In the formula, This is the composite feature vector obtained by concatenating the various feature vectors; The comprehensive feature vector Input a multilayer perceptron, perform forward propagation calculations, and output a predicted value of the total cost of the planning scheme. The expression is: ， In the formula, The weights of the linear rectified function, The weights of the integrated feature vectors, The bias is used to integrate the feature vectors. For the bias of the predicted value, It is a linear rectified function; Using the training sample set, the neural network surrogate model is trained with a composite loss function combining the main task and auxiliary task until the prediction accuracy of the neural network surrogate model reaches a preset requirement. The expression for the composite loss function is: ， In the formula, For compound loss, The weight of the mean squared error of total cost. The mean square error of the total cost. This is the total cost forecast. This is the actual value of the total cost. This is the weight of the mean square error of the new energy consumption rate. The mean square error of the renewable energy consumption rate. This is the predicted value for the renewable energy absorption rate. This represents the actual value of the new energy consumption rate. This represents the weight of the mean square error of the average voltage deviation. This represents the mean square error of the average voltage deviation. This is the predicted value of the average voltage deviation. This represents the actual value of the average voltage deviation. The output module is configured to generate and iteratively update the addressing and sizing scheme using an optimization algorithm, evaluate the predicted total cost of the addressing and sizing scheme using the neural network surrogate model, optimize the addressing and sizing scheme based on the predicted total cost until the convergence condition is met, and output the final target addressing and sizing scheme and the corresponding cooperative operation strategy.

6. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1 to 4.

Images