Method and system for optimizing distributed resources of multiple transformer areas of power distribution network

Abstract

The invention provides a method and a system for optimizing distributed resources of multiple transformer areas of a power distribution network. The method comprises the following steps of: acquiring historical load data, weather data and network topology parameters of each transformer area of the power distribution network for partitioning; constructing a multi-target resource allocation model according to the partitioning result and the network topology parameters; and based on the real-time load data and the network state, dynamically adjusting the output plan through a reinforcement learning algorithm, and feeding back the output plan to the power distribution management system. By constructing a multi-target resource allocation model taking minimization of operation cost and voltage stability as a joint optimization target, dual targets of minimization of the operation cost and enhancement of the voltage stability are considered, and the limitation of traditional single-target optimization is broken through; and in combination with a real-time feedback mechanism of a reinforcement learning algorithm, load fluctuation and network state change can be dynamically responded, multi-zone-area collaborative optimization operation of the power distribution network is realized, and the economical efficiency, safety and stability of the power distribution network are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution networks, and particularly to an optimization method and system for distributed resources in multiple substations of a distribution network. Background Art

[0002] Traditional distribution network resource optimization methods mostly use static substation division and economic optimization as the target optimization strategies, which have significant defects: on the one hand, the substation clustering based on the load curve at a fixed time period ignores the influence of weather factors on the dynamic characteristics of the load, resulting in insufficient adaptability of the partition result to the actual operation scenario; on the other hand, taking economy as the optimization goal, it is difficult to take into account grid safety constraints such as voltage stability, and at the same time lacks the ability to adaptively adjust to real-time working conditions. In addition, when facing large-scale multi-substation collaborative scheduling, the centralized optimization algorithm is prone to fall into the bottleneck of low calculation efficiency and slow convergence speed, unable to meet the dynamic regulation requirements, resulting in regulation lag and energy efficiency loss. Summary of the Invention

[0003] Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide an optimization method and system for distributed resources in multiple substations of a distribution network, aiming to solve the technical problems in the prior art that the distribution network resource optimization cannot meet the dynamic regulation requirements, resulting in regulation lag and energy efficiency loss.

[0004] To achieve the above purpose, in the first aspect, the present invention provides: an optimization method for distributed resources in multiple substations of a distribution network, including the following steps: Collect historical load data, weather data and network topology parameters of each substation in the distribution network; Based on the dynamic time warping algorithm, calculate the similarity of the load curves between substations according to the historical load data, and perform time series clustering in combination with the temperature difference factor to generate a partition result including the dynamic substation similarity; According to the partition result and the network topology parameters, construct a multi-objective resource allocation model with minimizing the operating cost and voltage stability as the joint optimization goal, and use the genetic algorithm to solve the output plans of distributed power sources and energy storage in each partition; Based on the real-time load data and network status, based on the output plan, dynamically adjust the output of distributed resources through the reinforcement learning algorithm, and feedback it to the distribution management system.

[0005] According to one aspect of the above technical solution, the calculation expression of the multi-objective resource allocation model is: ; ; ; In the formula, is the unit power generation cost of distributed power sources in the k-th partition, and K is the total number of partitions. is the unit power charge and discharge cost of energy storage in the k-th partition. is the charge and discharge power of energy storage in the k-th partition. is the output power value of distributed power sources in the k-th partition. is the unit network loss cost. is the total active power loss of the distribution network lines and transformers. is the node voltage of the n-th node. is the reference voltage, and N is the number of nodes. is the voltage stability weight coefficient. is the output power value of distributed power sources. and are the upper power limit value and the lower power limit value of distributed power sources respectively. is the total output power of all distributed power sources. is the charge and discharge power of energy storage. is the total load demand power of the distribution network. is the total power loss of the distribution network.

[0006] According to one aspect of the above technical solution, the steps of using the genetic algorithm to solve the output plans of distributed power sources and energy storage in each partition specifically include: Step S1: Based on the Pareto solution set output by the multi-objective resource allocation model, mix the optimal solutions of economy and voltage stability according to a preset ratio to generate the core individuals of the initial population, and add neighborhood perturbations. Step S2: Calculate the weighted sum value of economy and voltage stability of each group of solutions as the score to sort each group. Step S3: Retain the top 20% of the solutions, eliminate the bottom 20% of the solutions, and select the middle 60% of the solutions as the parents through roulette wheel selection. Swap the output plans of the preset time periods of the two parents to generate offspring. Step S4: Randomly modify the charge and discharge time periods of energy storage in some offspring or adjust the output values of distributed power sources for random mutation. Combine all offspring including mutated offspring with the parents to form an extended population, and sort the extended population based on the score. Select several individuals with the top scores as the next generation. Step S5: Loop through Step S2 to Step S4 to optimize the solutions generation by generation until the termination condition is met, obtaining the output plans of distributed power sources and energy storage in each partition. After each round of iteration, feedback the cost value and voltage stability of the optimal solution to the model to update the voltage stability weight coefficient for the next cycle.

[0007] According to one aspect of the above technical solution, the rule of random mutation in step S4 is: adjust the charge and discharge power of energy storage within the first preset range, and adjust the output power value of distributed power sources within the second preset range.

[0008] According to one aspect of the above technical solution, the rule of swap crossover in step S3 includes: the parent generations in the same partition exchange the output plans during the peak load periods, and those across partitions only exchange the gene segments of the same type of equipment.

[0009] According to one aspect of the above technical solution, the calculation expression of the dynamic similarity of transformer substations is as follows: ; ; In the formula, D is the dynamic similarity of transformer substations, T is the total time period, and are the load powers of transformer substation i and transformer substation j at the current time period t respectively, is the difference weight coefficient at time period t, is the temperature difference weight, is the temperature difference factor, is the temperature difference, is the humidity difference, and are the temperature and humidity adaptation parameters obtained by regression fitting based on historical load data.

[0010] According to one aspect of the above technical solution, the calculation expression of the difference weight coefficient is: ; In the formula, is the peak time period of the load curve of the distribution network, is the Gaussian attenuation coefficient.

[0011] According to one aspect of the above technical solution, the calculation expression of the reward function in the reinforcement learning algorithm is: ; In the formula, is the change rate of network loss, is the change amount of voltage deviation, is the voltage weight coefficient.

[0012] On the other hand, the present invention provides a distribution network multi-transformer substation distributed resource optimization system, including: An acquisition module, configured to acquire the historical load data, weather data, and network topology parameters of each transformer substation in the distribution network; A partitioning module, configured to calculate the similarity of the load curves between substations based on the dynamic time warping algorithm according to the historical load data, and perform time series clustering by combining the temperature difference factor to generate a partitioning result including the dynamic substation similarity; An output power plan module, configured to construct a multi-objective resource allocation model with minimizing the operating cost and voltage stability as the joint optimization objective according to the partitioning result and the network topology parameters, and use a genetic algorithm to solve the output power plans of distributed power sources and energy storage in each partition; A feedback module, configured to dynamically adjust the output power of distributed resources based on the real-time load data and network status, with the output power plan as the basis, through a reinforcement learning algorithm, and feed it back to the distribution management system.

[0013] According to one aspect of the above technical solution, the output power plan module is specifically configured to: Step S1, based on the Pareto solution set output by the multi-objective resource allocation model, mix the optimal solutions of economy and voltage stability according to a preset ratio to generate the core individuals of the initial population, and add neighborhood perturbations; Step S2, calculate the weighted sum value of economy and voltage stability of each group of solutions as the score to sort each group; Step S3, retain the top 20% of the solutions in terms of score, eliminate the bottom 20% of the solutions in terms of score, and select the middle 60% of the solutions as parents through roulette wheel selection. Exchange the output power plans of the preset time periods of the two parents to generate offspring; Step S4, randomly modify the charge and discharge time periods of the energy storage in some offspring or adjust the output power value of the distributed power source for random mutation. Combine all the offspring including the mutated offspring with the parents to form an extended population, and sort the extended population based on the score, and select several individuals with the top scores as the next generation; Step S5, loop through Step S2 to Step S4 to optimize the solutions generation by generation until the termination condition is met, and obtain the output power plans of the distributed power sources and energy storage in each partition. After each round of iteration, feedback the cost value and voltage stability of the optimal solution to the model to update the voltage stability weight coefficient of the next cycle.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: By using the dynamic time warping algorithm to fuse the temperature difference factor to achieve dynamic clustering of substations, and by constructing a multi-objective resource allocation model with minimizing the operating cost and voltage stability as the joint optimization objective, it takes into account the dual objectives of minimizing the operating cost and enhancing the voltage stability, breaking through the limitations of traditional single-objective optimization; combined with the real-time feedback mechanism of the reinforcement learning algorithm, it can dynamically respond to load fluctuations and network state changes, significantly improve the robustness and self-adaptability of distributed resource regulation, realize the coordinated optimization operation of multiple substations in the distribution network, and enhance the economy, safety and stability of the distribution network. Description of the Drawings

[0015] Figure 1 It is a flowchart of the distribution resource optimization method for multiple substations in the distribution network in the first embodiment of the present invention; Figure 2 It is a structural block diagram of the distribution resource optimization system for multiple substations in the distribution network in the second embodiment of the present invention; The following specific embodiments will further illustrate the present invention in conjunction with the above-mentioned drawings. Specific Embodiments

[0016] To facilitate the understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, these embodiments are provided to make the disclosure of the present invention more thorough and comprehensive.

[0017] It should be noted that when an element is referred to as being "fixedly provided on" another element, it can be directly on the other element or there may also be an intermediate element. When an element is considered to be "connected" to another element, it can be directly connected to the other element or there may be an intermediate element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are only for the purpose of illustration.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0019] Embodiment 1 Please refer to Figure 1 , which shows a flowchart of the distribution resource optimization method for multiple substations in the distribution network in the first embodiment of the present invention. As shown in the figure, the method includes the following steps: Step S100, collect the historical load data, weather data and network topology parameters of each substation in the distribution network.

[0020] In some application scenarios of this embodiment, the historical load data includes the active power curves (15-minute granularity), reactive power curves, and daily / weekly load peak-valley characteristics of each substation area collected in real time through smart meters, SCADA systems, and PMUs (synchronous phasor measurement units). The weather data includes temperature, humidity, etc., which is convenient for correcting the similarity calculation through the temperature-humidity difference factor in the follow-up to avoid the static partition deviation that only relies on historical loads. For example, in the scenario where the air-conditioning load in summer is strongly correlated with high temperatures, the adaptability of the partition result to the actual working conditions is improved, providing accurate input for multi-objective optimization. The network topology parameters include the line impedance matrix (R / X value), transformer capacity and tap position, node connection relationship, distributed power source parameters (such as the maximum output of photovoltaic inverters, charge-discharge efficiency curves of energy storage), load node distribution, and capacity constraints, etc. The line impedance matrix is used to calculate network losses and construct an economic objective function, and the node connection relationship and transformer capacity constraint voltage stability model.

[0021] Step S200, based on the dynamic time warping algorithm, calculate the similarity of the load curves between substations according to the historical load data, and perform time series clustering in combination with the temperature difference factor to generate a partition result including the dynamic similarity of substations.

[0022] Preferably, in this embodiment, the calculation expression of the dynamic similarity of substations is as follows: ; ; In the formula, D is the dynamic similarity of substations, T is the total time period, and are the load powers of substation area i and substation area j at the current time period t respectively, is the difference weight coefficient at time period t, is the temperature difference weight, is the temperature difference factor, is the temperature difference, is the humidity difference, and are the temperature-humidity adaptation parameters obtained by regression fitting according to the historical load data. It should be noted that in this embodiment, the above temperature-humidity parameters and load powers need to be subjected to data standardization processing before calculation to eliminate the dimension difference; by calculating the similarity of the load curves between substation area i and substation area j, the distortion problem caused by time period offset (such as the difference in morning / evening peak times) is solved.

[0023] Furthermore, the calculation expression of the above difference weight coefficient is: ; In the formula, is the peak time period of the distribution network load curve, is the Gaussian attenuation coefficient. Specifically, by setting the differential weight coefficient at the peak period of the load curve, the weight reaches the maximum value near this period, strengthening the influence of the load difference during the peak period on the similarity calculation, which can avoid the noise interference in the non-critical period, and the result is more in line with the actual electricity consumption characteristics. The larger the Gaussian attenuation coefficient, the flatter the weight distribution (covering a wider time range), weakening the difference in the non-peak period; the smaller the Gaussian attenuation coefficient, the more concentrated the weight is near the peak period, focusing on the difference in the critical period.

[0024] Step S300: According to the partition result and the network topology parameters, construct a multi-objective resource allocation model with the joint optimization objective of minimizing the operating cost and voltage stability, and use the genetic algorithm to solve the output plans of distributed power sources and energy storage in each partition.

[0025] Specifically, in this embodiment, the calculation expression of the multi-objective resource allocation model is: ; ; ; In the formula, is the unit power generation cost of the distributed power source in the k-th partition, K is the total number of partitions, is the unit power charge and discharge cost of the energy storage in the k-th partition, is the total active power loss of the distribution network lines and transformers, is the charge and discharge power of the energy storage in the k-th partition, is the output power value of the distributed power source in the k-th partition, is the unit network loss cost, is the node voltage of the n-th node, is the reference voltage, N is the number of nodes, is the voltage stability weight coefficient, is the output power value of the distributed power source, and are the upper power limit value and the lower power limit value of the distributed power source respectively, is the total output power of all distributed power sources, is the charge and discharge power of the energy storage, is the total load demand power of the distribution network, is the total power loss of the distribution network.

[0026] It should be noted that is the economic part of the multi-objective resource allocation model, This is the voltage stability part of the multi-objective resource allocation model. Among them, the economic part aims to minimize the total operating cost of the distribution network. Distributed power sources include low-cost power sources such as photovoltaic and wind power, as well as high-cost power sources such as gas turbines. The unit power charge and discharge cost of energy storage reflects the aging loss of the battery. The voltage stability part ensures the safety of the power grid by minimizing the node voltage deviation. The voltage stability weight coefficient is used for the trade-off between economy and voltage stability. A high value is used to strengthen the voltage stability constraint, which may limit the output of distributed power sources (such as photovoltaic derating operation) and increase the operating cost. A low value gives priority to economy, which may lead to the risk of voltage over-limit.

[0027] Furthermore, in this embodiment, the steps of using the genetic algorithm to solve the output plans of distributed power sources and energy storage in each partition specifically include: Step S1, based on the Pareto solution set output by the multi-objective resource allocation model, mix the optimal solutions of economy and voltage stability according to a preset ratio to generate the core individuals of the initial population, and add neighborhood perturbations. Specifically, the above preset ratio is 7:3, and the neighborhood perturbation is the output power value of the distributed power source ±10%, and the charge and discharge power of the energy storage ±20%.

[0028] Step S2, calculate the weighted sum value of the economy and voltage stability of each group of solutions as the score to sort each group. The score formula is: ; In the formula, S is the score, h1 is the economic part of the above multi-objective resource allocation model, is the voltage stability part of the above multi-objective resource allocation model, where, is the dynamic voltage stability weight coefficient, and its initial value is the voltage stability weight coefficient output by the multi-objective resource allocation model , and is updated through S5.

[0029] Step S3, retain the top 20% of the solutions in terms of score, eliminate the bottom 20% of the solutions in terms of score, and select 60% of the solutions in the middle as parents through roulette wheel selection. Swap the output plans of the preset time periods of the two parents to generate offspring. Specifically, the swapping and crossing rules in Step S3 include: Parents in the same partition exchange the output plans during the load peak period, and those across partitions only exchange the gene segments of the same type of equipment. Since the equipment capacities in different partitions may vary greatly, only exchange the genes of the same type and similar capacity equipment to avoid the charge and discharge power exceeding the equipment capacity. By exchanging the efficient output modes of the same type of equipment, explore the multi-partition coordinated scheduling (such as multiple energy storages discharging at the same time to support the main network voltage).

[0030] The above preset time periods include three consecutive hours with the highest load power (such as 18:00 - 21:00), three consecutive hours with the lowest load power (such as 02:00 - 05:00), the time periods in historical data where the voltage deviation exceeds 5% (such as 12:00 - 14:00 when the photovoltaic output suddenly increases and causes the voltage to rise), and the time periods when the energy storage SOC is near the upper and lower limits (such as the discharge time period when SOC = 20%).

[0031] Step S4, randomly modify the energy storage charge and discharge time periods in some offspring or adjust the output values of distributed power sources for random mutation, combine all offspring including the mutated offspring with the parent generation to form an extended population, and sort the extended population based on the scores, select several individuals with the highest scores as the next generation. Among them, the rule of random mutation in step S4 is: adjust the charge and discharge power of the energy storage within the first preset range, and adjust the output power value of the distributed power source within the second preset range. The above first preset range is preferably ±20% of the current power value, and the second preset range is preferably ±10% of the current output power value. The number of several individuals is the same as the number of core individuals in the initial population.

[0032] Step S5, loop through steps S2 to S4 to optimize the solution generation by generation until the termination condition is met, and obtain the output plans of distributed power sources and energy storage in each partition. After each round of iteration, feedback the cost value of the optimal solution and the voltage stability to the model to update the voltage stability weight coefficient for the next cycle. Specifically, the termination condition is preferably reaching the maximum number of iterations of 100 times or converging stably, that is, the improvement amplitude of the score of the optimal solution is less than 1%.

[0033] Step S400, based on real-time load data and network status, and based on the above output plan, dynamically adjust the output of distributed resources through a reinforcement learning algorithm and feedback it to the distribution management system. Preferably, in this embodiment, the reinforcement learning algorithm uses the Q-Learning algorithm. Among them, the main steps include: capturing the real-time grid operation status to define the state space, which includes the load rate, voltage status, distributed power source and energy storage status; the action is defined as the real-time output adjustment of the distributed power source (DG) and the energy storage (ESS); balance economy and voltage stability through the reward function, and the Q-value update rule is the existing classic strategy.

[0034] Among them, the calculation expression of the reward function is: ; In the formula, is the change rate of network loss, is the change amount of voltage deviation, is the voltage weight coefficient.

[0035] In summary, in the above embodiments of the present invention, the method for optimizing the distribution resources of multiple substations in the distribution network realizes the dynamic clustering of substations through the dynamic time warping algorithm by integrating the temperature difference factor. By constructing a multi-objective resource allocation model with the joint optimization objectives of minimizing the operating cost and voltage stability, it takes into account the dual objectives of minimizing the operating cost and enhancing the voltage stability, breaking through the limitations of traditional single-objective optimization. Combining the real-time feedback mechanism of the reinforcement learning algorithm, it can dynamically respond to load fluctuations and network state changes, significantly improving the robustness and self-adaptability of distributed resource regulation, realizing the collaborative optimization operation of multiple substations in the distribution network, and enhancing the economy, safety, and stability of the distribution network.

[0036] Embodiment 2 The second embodiment of the present application also provides a system for optimizing the distribution resources of multiple substations in the distribution network. This system is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated here. As used hereinafter, terms such as "module", "unit", "sub-unit", etc. can be a combination of software and / or hardware that can achieve a predetermined function. Although the systems described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0037] As Figure 2 shown, the system includes: a collection module 100, a partitioning module 200, an output plan module 300, and a feedback module 400.

[0038] The collection module 100 is used to collect historical load data, weather data, and network topology parameters of each substation in the distribution network; The partitioning module 200 is used to calculate the similarity of the load curves between substations based on the historical load data according to the dynamic time warping algorithm, and perform time series clustering in combination with the temperature difference factor to generate a partitioning result including the dynamic substation similarity; The output plan module 300 is used to construct a multi-objective resource allocation model with the joint optimization objectives of minimizing the operating cost and voltage stability according to the partitioning result and the network topology parameters, and use the genetic algorithm to solve the output plans of distributed power sources and energy storage in each partition; The feedback module 400 is used to dynamically adjust the output of distributed resources based on the real-time load data and network state, taking the output plan as the basis, through the reinforcement learning algorithm, and feed it back to the distribution management system.

[0039] Preferably, in this embodiment, the output plan module 300 is specifically used for: Step S1, mixing the optimal solutions of economy and voltage stability according to a preset ratio based on the Pareto solution set output by the multi-objective resource allocation model to generate the core individuals of the initial population, and adding neighborhood perturbations; Step S2: Calculate the weighted sum of the economy and voltage stability of each group of solutions as the score, and sort each group accordingly. Step S3: Retain the top 20% of the solutions with the highest scores, eliminate the bottom 20% of the solutions, and select the middle 60% of the solutions as parents through roulette wheel selection. Exchange the output plans of the two parents during the preset time period to generate offspring. Step S4: Randomly modify the energy storage charging and discharging time periods in some of the offspring or adjust the output values of distributed power sources for random mutation. Combine all the offspring, including the mutated offspring, with the parents to form an extended population, and sort the extended population based on the scores. Select the top several individuals with the highest scores as the next generation. Step S5: Repeat Steps S2 to S4 iteratively to optimize the solutions generation by generation until the termination condition is met, obtaining the output plans of distributed power sources and energy storage in each partition. After each round of iteration, feedback the cost value of the optimal solution and the voltage stability to the model to update the voltage stability weight coefficient for the next cycle.

[0040] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as within the scope described in this specification.

[0041] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A multi - substation distribution resource optimization method for a distribution network, characterized in that It includes the following steps: Collect historical load data, weather data, and network topology parameters of each substation area in the distribution network; Based on the dynamic time warping algorithm, calculate the similarity of load curves between substations according to the historical load data, and perform time series clustering in combination with the temperature difference factor to generate a partition result including the dynamic similarity of substations; According to the partition result and the network topology parameters, construct a multi-objective resource allocation model with the joint optimization goal of minimizing operating costs and voltage stability, and use the genetic algorithm to solve the output plans of distributed power sources and energy storage in each partition; Based on the real-time load data and network status, and based on the output plan, dynamically adjust the output of distributed resources through the reinforcement learning algorithm and feedback it to the distribution management system.

2. The multi-substation distribution resource optimization method for a distribution network according to claim 1, wherein The calculation expression of the multi-objective resource allocation model is: ； ； ； Wherein, is the unit power generation cost of distributed power sources in the k-th partition, K is the total number of partitions, is the unit power charge and discharge cost of energy storage in the k-th partition, is the total active power loss of the distribution network lines and transformers, is the charge and discharge power of energy storage in the k-th partition, is the output power value of distributed power sources in the k-th partition, is the unit network loss cost, is the node voltage of the n-th node, is the reference voltage, N is the number of nodes, is the voltage stability weight coefficient, is the output power value of distributed power sources, and are respectively the upper power limit value and the lower power limit value of distributed power sources, is the total output power of all distributed power sources, is the charge and discharge power of energy storage, is the total load demand power of the distribution network, is the total power loss of the distribution network.

3. The optimized method for distribution resources of multiple substations in a distribution network according to claim 2, characterized in that The steps of using the genetic algorithm to solve the output plans of distributed power sources and energy storage in each partition specifically include: Step S1, based on the Pareto solution set output by the multi-objective resource allocation model, mix the optimal solutions of economy and voltage stability according to a preset ratio to generate the core individuals of the initial population, and add neighborhood perturbations; Step S2, calculate the weighted sum value of economy and voltage stability of each group of solutions as the score to sort each group; Step S3, retain the top 20% of the solutions with scores, eliminate the bottom 20% of the solutions with scores, and select the middle 60% of the solutions as parents through roulette selection. Exchange the output plans of the preset time periods of the two parents to generate offspring; Step S4, randomly modify the charge and discharge time periods of the energy storage in some offspring or adjust the output value of the distributed power source for random mutation. Combine all the offspring including the mutated offspring with the parents to form an extended population, and sort the extended population based on the scores, and select several individuals with the top scores as the next generation; Step S5, loop through Step S2 to Step S4 to optimize the solutions generation by generation until the termination condition is met, and obtain the output plans of distributed power sources and energy storage in each partition. After each round of iteration, feedback the cost value and voltage stability of the optimal solution to the model to update the voltage stability weight coefficient of the next cycle.

4. The multi-substation area distribution resource optimization method for a distribution network according to claim 3, characterized in that, The rule of random mutation in Step S4 is: adjust the charge and discharge power of the energy storage within the first preset range, and adjust the output power value of the distributed power source within the second preset range.

5. The multi-substation area distributed resource optimization method for a distribution network according to claim 3, characterized in that The rule of exchange and crossover in Step S3 includes: exchange the output plans of the parents at the load peak time period in the same partition, and only exchange the gene segments of the same type of equipment across partitions.

6. The multi-substation area distribution resource optimization method for a distribution network according to claim 1, characterized in that The calculation expression of the dynamic similarity of substations is as follows: ； ； Where D is the similarity of the dynamic power distribution area, T is the total time period, and are the load powers of the power distribution area i and the power distribution area j at the current time period t respectively, is the difference weight coefficient at the time period t, is the temperature difference weight, is the temperature difference factor, is the temperature difference, is the humidity difference, and are the temperature and humidity adaptation parameters obtained by regression fitting based on historical load data.

7. The multi-substation area distribution resource optimization method for a distribution network according to claim 6, characterized in that The calculation expression of the difference weight coefficient is: ； In the formula, is the peak period of the distribution network load curve, is the Gaussian attenuation coefficient.

8. The multi-substation area distribution resource optimization method for a distribution network according to claim 1, characterized in that The calculation expression of the reward function in the reinforcement learning algorithm is: ； In the formula, is the network loss change rate, is the voltage deviation change amount, is the voltage weight coefficient.

9. A multi-substation distribution resource optimization system for a distribution network, characterized in that It includes: A collection module for collecting historical load data, weather data, and network topology parameters of each substation area in the distribution network; A partition module for calculating the similarity of load curves between substations based on the historical load data using the dynamic time warping algorithm, and performing time series clustering in combination with the temperature difference factor to generate a partition result including the dynamic similarity of substations; An output planning module, configured to construct a multi-objective resource allocation model with minimizing operating cost and voltage stability as the joint optimization objective according to the partitioning result and the network topology parameters, and use a genetic algorithm to solve the output plans of distributed power sources and energy storage in each partition; A feedback module, configured to dynamically adjust the output of distributed resources based on real-time load data and network status, with the output plan as the basis, through a reinforcement learning algorithm, and feedback it to the distribution management system.

10. The multi-substation distribution resource optimization system for a distribution network according to claim 9, wherein, The output planning module is specifically configured to: Step S1, mix the optimal solutions of economy and voltage stability according to a preset ratio based on the Pareto solution set output by the multi-objective resource allocation model to generate the core individuals of the initial population, and add neighborhood perturbations; Step S2, calculate the weighted sum value of economy and voltage stability of each group of solutions as the score to sort each group; Step S3, retain the top 20% of the solutions in terms of score, eliminate the bottom 20% of the solutions in terms of score, and select the middle 60% of the solutions as parents through roulette wheel selection. Exchange the output plans of the two parents in a preset time period to generate offspring; Step S4, randomly modify the energy storage charging and discharging time periods in some offspring or adjust the output values of distributed power sources for random mutation. Combine all offspring including mutated offspring with the parents to form an extended population, and sort the extended population based on the score, and select the top several individuals in terms of score as the next generation; Step S5, loop through Step S2 to Step S4 to optimize the solutions generation by generation until the termination condition is met, and obtain the output plans of distributed power sources and energy storage in each partition. After each round of iteration, feedback the cost value and voltage stability of the optimal solution to the model to update the voltage stability weight coefficient for the next cycle.

Images