County distributed photovoltaic rigid-flexible hybrid control fair adjustment method and system

Abstract

The application discloses a county distributed photovoltaic rigid-flexible hybrid control fair adjustment method and system, which comprises the following steps: dividing the area adjustment resources into rigid areas and flexible areas, the rigid areas participating in reduction through switching or fixed proportion switching, and the flexible areas participating in residual adjustment amount distribution through continuous adjustment weight; constructing a hybrid action space containing rigid switching actions and flexible continuous actions; constructing the area reduction responsibility deviation as a responsibility virtual queue to generate a limited scale rigid candidate set; generating a conditional flexible adjustment weight for each rigid candidate action, evaluating the long-term value of the complete rigid-flexible action through a hybrid action Critic, and selecting the hybrid action with the highest long-term value for execution. The application can meet the requirement of county reverse limit inhibition while promoting the rigid areas and the flexible areas to jointly bear the adjustment responsibility, reducing the long-term responsibility deviation, and improving the engineering executability of the photovoltaic adjustment strategy under the condition of rigid-flexible hybrid resources.

Description

Technical Field

[0001] This invention relates to the fields of distributed photovoltaic regulation, power system operation optimization, and reinforcement learning control technology, specifically a fair regulation method and system for rigid-flexible hybrid control of distributed photovoltaic power in county areas. Background Technology

[0002] After large-scale grid connection of distributed photovoltaic (PV) systems in counties, reverse power transmission pressure is prone to occur during peak PV output periods at midday. Aggregator service providers need to coordinate multiple distribution areas to participate in regulation in order to meet the reverse power limits allowed by the grid connection point.

[0003] In actual county-level power distribution networks, the regulation capabilities of different distribution areas vary. Some distribution areas have continuous power regulation capabilities and can dynamically adjust the reduction weight according to dispatch instructions; others can only perform on-off switching, fixed-ratio reduction, or tiered response. Rigid switching actions are discrete. If the continuous regulation strategy is directly projected onto rigid distribution areas, it may cause overshooting of reduction, residual exceeding limits, or frequent switching, or it may cause flexible distribution areas to bear the main reduction responsibility for a long time.

[0004] Therefore, a fair adjustment method is needed for scenarios where rigid and flexible transformer areas coexist. This method should compress the search space for rigid switching combinations, enable both rigid and flexible transformer areas to share the adjustment responsibility, and select hybrid actions that take into account safety, economy, fairness, and feasibility through a value evaluation mechanism. Summary of the Invention

[0005] The purpose of this invention is to provide a fair regulation method for rigid-flexible hybrid control of distributed photovoltaic power in county-level areas, comprising the following steps:

[0006] Obtain aggregated photovoltaic output, aggregated load power, heterogeneous responsibility share of each transformer area, allowed reverse power boundary of grid connection point, and control type of each transformer area across multiple transformer areas in the county.

[0007] Based on the control type of the distribution area, the distribution area is divided into rigid distribution areas and flexible distribution areas;

[0008] The necessary system adjustments are determined based on the aggregated photovoltaic output, aggregated load power, and the allowable reverse power boundary at the grid connection point.

[0009] Construct a hybrid motion space consisting of rigid transformer switching actions and continuous adjustment weights for flexible transformers;

[0010] The virtual responsibility queue is updated based on the deviation between the base responsibility share of the transformer area and the actual power reduction.

[0011] A rigid candidate action set is generated based on the aforementioned responsibility virtual queue, the necessary system adjustment amount, and the previous switching state.

[0012] For each rigid candidate action in the rigid candidate action set, a corresponding flexible adjustment weight is generated;

[0013] The candidate mixed action, consisting of rigid switching action and flexible adjustment weight, is evaluated using a mixed action value evaluation network. The candidate mixed action with the highest long-term value is selected as the action to be executed in the current period.

[0014] Furthermore, the rigid station area is a station area that can only participate in reduction according to the switching, fixed ratio reduction, or graded response method;

[0015] The flexible distribution area is a distribution area that can continuously adjust its adjustment weight according to the scheduling signal and participate in the allocation of the remaining adjustment amount.

[0016] Furthermore, the necessary adjustment amount of the system is the non-negative portion of the sum of the county-level aggregated photovoltaic output exceeding the county-level aggregated load power and the grid connection point's allowable reverse power boundary;

[0017] When the necessary adjustment amount of the system is greater than zero, the rigid and flexible transformer areas share the task of reduction.

[0018] The necessary system adjustments are as follows:

[0019]

[0020] In the formula: For time period The necessary adjustment amount; The maximum allowable reverse power at the grid connection point; .

[0021] Furthermore, the hybrid action space is represented as , For rigid switching operations, For continuous adjustment of the weight of flexible distribution areas;

[0022] The switching action value for rigid transformer areas is 0 or 1, while the continuous adjustment weight value for flexible transformer areas is between 0 and 1.

[0023] The rigid reduction amount in the rigid distribution zone is determined by the switching action and a fixed reduction ratio, namely:

[0024]

[0025] In the formula: The reduction ratio is fixed for rigid transformer areas;

[0026] The remaining adjustment amount after the rigid reduction is implemented is allocated by the flexible distribution area based on the continuous adjustment weight and heterogeneous responsibility share.

[0027] Furthermore, the responsibility virtual queue is updated recursively based on the baseline reduction responsibility and actual reduction power of the transformer area in the current time period;

[0028] When the actual power reduction in a transformer area is less than the baseline reduction responsibility, the virtual responsibility queue increases; when the actual power reduction in a transformer area is greater than the baseline reduction responsibility, the virtual responsibility queue is released.

[0029] The virtual queue for district responsibilities has been updated as follows:

[0030]

[0031] In the formula: Taiwan District The responsibility of the virtual queue; Taiwan District The baseline responsibility corresponding to the current necessary adjustment amount; Taiwan District The actual power reduction; .

[0032] Furthermore, the rigid candidate action set includes responsibility priority candidates, power matching candidates, and stability maintenance candidates.

[0033] Furthermore, the set of rigid candidate actions satisfies rigid switching constraints, switching variation constraints, and power reduction boundary constraints, namely:

[0034]

[0035] In the formula: A set of rigid actions that can participate in candidate generation; For action The number of changes in the switching state relative to the previous moment; This represents the maximum number of switching operations allowed in a single time period. For action The corresponding rigid reduction amount.

[0036] Furthermore, the flexible adjustment weights are generated by the conditional flexible adjustment Actor;

[0037] The conditional flexible adjustment Actor takes the current state, rigid candidate actions, and corresponding remaining adjustment amounts as inputs and outputs the continuous adjustment weight of the flexible transformer area.

[0038] The continuous adjustment weights for flexible transformer areas are shown below:

[0039] ,

[0040] In the formula: To flexibly adjust the Actor network; These are the parameters of the Actor network; For the first The flexible adjustment weights corresponding to each rigid candidate action;

[0041] The Hybrid Action Critic takes the current state, rigid candidate actions, and flexible adjustment weights as joint inputs and outputs the long-term value of the candidate hybrid actions.

[0042] The long-term value of the candidate mixed actions is as follows:

[0043]

[0044]

[0045]

[0046] In the formula: Index the candidate actions with the highest value; For time period The final combined action.

[0047] Furthermore, the reward function of the hybrid action Critic is used to guide the Critic to learn the long-term value of candidate hybrid actions, so that the strategy can comprehensively consider operating costs, excessive reduction, liability fairness and rigid switching stability while satisfying the grid connection point safety constraints.

[0048] The reward function for the hybrid action Critic is as follows:

[0049]

[0050] In the formula: The reward scaling factor; Operating costs; For excessive reduction of penalties; To constrain the amount of violations; For the sake of instantaneous liability fairness; For intraday liability and fairness items; This is a rigid switching variation term; , , , For the corresponding weights.

[0051] A fair regulation system for rigid-flexible hybrid control of county-level distributed photovoltaic power generation based on the method includes a data acquisition module, a transformer area control type identification module, a necessary regulation amount calculation module, a responsibility virtual queue update module, a rigid candidate action generation module, a conditional flexible regulation module, a hybrid action value evaluation module, and a regulation strategy output module.

[0052] The data acquisition module is used to acquire photovoltaic output, load power, heterogeneous liability share, and grid connection point boundary; the transformer area control type identification module is used to classify rigid transformer areas and flexible transformer areas.

[0053] The necessary adjustment calculation module is used to calculate the necessary adjustment of the system in the current time period;

[0054] The responsibility virtual queue update module is used to recursively update the responsibility deviation of the transformer area.

[0055] The rigid candidate action generation module is used to generate a finite-sized set of rigid candidate actions.

[0056] The conditional flexible adjustment module is used to generate corresponding flexible adjustment weights for each rigid candidate action.

[0057] The hybrid action value evaluation module is used to evaluate the long-term value of candidate hybrid actions.

[0058] The adjustment strategy output module is used to output the current time period's rigid transformer switching status, flexible transformer adjustment weight, and actual power reduction of each transformer.

[0059] This invention incorporates the different adjustment capabilities of rigid and flexible transformer areas into a unified hybrid action space; it uses a responsibility virtual queue to generate candidate actions for each time period to account for long-term responsibility deviations; it compresses rigid switching combinations through three types of rules: responsibility priority, power matching, and stability maintenance; and it achieves rigid-flexible collaborative decision-making through conditional flexible adjustment and hybrid action value evaluation, so that the adjustment strategy simultaneously meets the requirements of grid connection safety, operational economy, responsibility fairness, and action feasibility. Attached Figure Description

[0060] Figure 1 This is a comparison chart of the daily responsibility benchmark and the actual cumulative reduction amount in an embodiment of the present invention;

[0061] Figure 2 This is a ranking curve of daily responsibility completion rate in an embodiment of the present invention;

[0062] Figure 3 This is a diagram showing the power reduction distribution in the time zone of the sunlit island in an embodiment of the present invention.

[0063] Figure 4 This is the system framework of the present invention. Detailed Implementation

[0064] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0065] Example 1:

[0066] A fair regulation method for rigid-flexible hybrid control of distributed photovoltaic power in county-level areas includes the following steps:

[0067] Obtain aggregated photovoltaic output, aggregated load power, heterogeneous responsibility share of each transformer area, allowed reverse power boundary of grid connection point, and control type of each transformer area across multiple transformer areas in the county.

[0068] Based on the control type of the distribution area, the distribution area is divided into rigid distribution areas and flexible distribution areas;

[0069] The necessary system adjustments are determined based on the aggregated photovoltaic output, aggregated load power, and the allowable reverse power boundary at the grid connection point.

[0070] A hybrid action space is constructed, consisting of rigid transformer area switching actions and flexible transformer area continuous adjustment weights. The flexible transformer area continuous adjustment weights are generated by the conditional flexible adjustment Actor based on the current state, rigid candidate actions, and remaining adjustment amounts, and are constrained to the range of 0 to 1. During execution, the remaining adjustment amounts are allocated in conjunction with heterogeneous liability shares.

[0071] The virtual responsibility queue is updated based on the deviation between the base responsibility share of the transformer area and the actual power reduction.

[0072] A rigid candidate action set is generated based on the aforementioned responsibility virtual queue, the necessary system adjustment amount, and the previous switching state.

[0073] For each rigid candidate action in the rigid candidate action set, a corresponding flexible adjustment weight is generated;

[0074] The candidate mixed action, consisting of rigid switching action and flexible adjustment weight, is evaluated using a mixed action value evaluation network. The candidate mixed action with the highest long-term value is selected as the action to be executed in the current period.

[0075] Example 2:

[0076] The county-level distributed photovoltaic rigid-flexible hybrid control fair regulation method is the same as that in Example 1. Furthermore, the rigid area is an area that can only participate in the reduction according to the switching, fixed ratio reduction or graded response method.

[0077] The flexible distribution area is a distribution area that can continuously adjust its adjustment weight according to the scheduling signal and participate in the allocation of the remaining adjustment amount.

[0078] Example 3:

[0079] A county-level distributed photovoltaic rigid-flexible hybrid control fair regulation method, the technical content of which is the same as any one of Embodiments 1-2, further wherein the necessary regulation amount of the system is the non-negative part of the sum of the county-level aggregated photovoltaic output exceeding the county-level aggregated load power and the grid connection point's allowable reverse power boundary;

[0080] When the necessary adjustment amount of the system is greater than zero, the rigid and flexible transformer areas share the task of reduction.

[0081] The necessary system adjustments are as follows:

[0082]

[0083] In the formula: For time period The necessary adjustment amount; The maximum allowable reverse power at the grid connection point; .

[0084] Example 4:

[0085] A county-level distributed photovoltaic rigid-flexible hybrid control fair regulation method, with technical content identical to any one of embodiments 1-3, further wherein the hybrid action space is represented as follows: , For rigid switching operations, For continuous adjustment of the weight of flexible distribution areas;

[0086] The switching action value for rigid transformer areas is 0 or 1, while the continuous adjustment weight value for flexible transformer areas is between 0 and 1.

[0087] The rigid reduction amount in the rigid distribution zone is determined by the switching action and a fixed reduction ratio, namely:

[0088]

[0089] In the formula: The reduction ratio is fixed for rigid transformer areas;

[0090] The remaining adjustment amount after the rigid reduction is implemented is allocated by the flexible distribution area based on the continuous adjustment weight and heterogeneous responsibility share.

[0091] Example 5:

[0092] The county-level distributed photovoltaic rigid-flexible hybrid control fair regulation method has the same technical content as any one of embodiments 1-4. Furthermore, the responsibility virtual queue is updated recursively based on the benchmark reduction responsibility and actual reduction power of the transformer area in the current time period.

[0093] When the actual power reduction in a transformer area is less than the baseline reduction responsibility, the virtual responsibility queue increases; when the actual power reduction in a transformer area is greater than the baseline reduction responsibility, the virtual responsibility queue is released.

[0094] The virtual queue for district responsibilities has been updated as follows:

[0095]

[0096] In the formula: Taiwan District The responsibility of the virtual queue; Taiwan District The baseline responsibility corresponding to the current necessary adjustment amount; Taiwan District The actual power reduction; .

[0097] Example 6:

[0098] The county-level distributed photovoltaic rigid-flexible hybrid control fair regulation method has the same technical content as any one of embodiments 1-5. Furthermore, the rigid candidate action set includes responsibility priority candidates, power matching candidates, and stability maintenance candidates.

[0099] Example 7:

[0100] A county-level distributed photovoltaic rigid-flexible hybrid control fair regulation method, with technical content identical to any one of Examples 1-6, further wherein the rigid candidate action set satisfies rigid switching constraints, switching change constraints, and power reduction boundary constraints, namely:

[0101]

[0102] In the formula: A set of rigid actions that can participate in candidate generation; For action The number of changes in the switching state relative to the previous moment; This represents the maximum number of switching operations allowed in a single time period. For action The corresponding rigid reduction amount.

[0103] Example 8:

[0104] A fair regulation method for rigid-flexible hybrid control of distributed photovoltaic power in county-level areas, with the same technical content as any one of Examples 1-7, further wherein the flexible regulation weight is generated by the conditional flexible regulation Actor;

[0105] The conditional flexible adjustment Actor takes the current state, rigid candidate actions, and corresponding remaining adjustment amounts as inputs and outputs the continuous adjustment weight of the flexible transformer area.

[0106] The continuous adjustment weights for flexible transformer areas are shown below:

[0107] ,

[0108] In the formula: To flexibly adjust the Actor network; These are the parameters of the Actor network; For the first The flexible adjustment weights corresponding to each rigid candidate action;

[0109] The Hybrid Action Critic takes the current state, rigid candidate actions, and flexible adjustment weights as joint inputs and outputs the long-term value of the candidate hybrid actions.

[0110] The long-term value of the candidate mixed actions is as follows:

[0111]

[0112]

[0113]

[0114] In the formula: Index the candidate actions with the highest value; For time period The final combined action.

[0115] Example 9:

[0116] The county-level distributed photovoltaic rigid-flexible hybrid control fair regulation method has the same technical content as any one of Examples 1-8. Furthermore, the reward function of the hybrid action Critic is used to guide the Critic to learn the long-term value of candidate hybrid actions, so that the strategy can comprehensively consider operating costs, excessive reduction, responsibility fairness and rigid switching stability while satisfying the grid connection point safety constraints.

[0117] The reward function for the hybrid action Critic is as follows:

[0118]

[0119] In the formula: The reward scaling factor; Operating costs; For excessive reduction of penalties; To constrain the amount of violations; For the sake of instantaneous liability fairness; For intraday liability and fairness items; This is a rigid switching variation term; , , , For the corresponding weights.

[0120] Example 10:

[0121] A fair regulation system for rigid-flexible hybrid control of county-level distributed photovoltaic power generation based on the methods described in Examples 1-9 includes a data acquisition module, a transformer area control type identification module, a necessary regulation amount calculation module, a responsibility virtual queue update module, a rigid candidate action generation module, a conditional flexible regulation module, a hybrid action value evaluation module, and a regulation strategy output module.

[0122] The data acquisition module is used to acquire photovoltaic output, load power, heterogeneous liability share, and grid connection point boundary; the transformer area control type identification module is used to classify rigid transformer areas and flexible transformer areas.

[0123] The necessary adjustment calculation module is used to calculate the necessary adjustment of the system in the current time period;

[0124] The responsibility virtual queue update module is used to recursively update the responsibility deviation of the transformer area.

[0125] The rigid candidate action generation module is used to generate a finite-sized set of rigid candidate actions.

[0126] The conditional flexible adjustment module is used to generate corresponding flexible adjustment weights for each rigid candidate action.

[0127] The hybrid action value evaluation module is used to evaluate the long-term value of candidate hybrid actions.

[0128] The adjustment strategy output module is used to output the current time period's rigid transformer switching status, flexible transformer adjustment weight, and actual power reduction of each transformer.

[0129] Example 11:

[0130] The county-level distributed photovoltaic rigid-flexible hybrid control fair regulation method includes the following steps:

[0131] Obtain the photovoltaic output, load power, heterogeneous responsibility share of each transformer area, allowed reverse power boundary of grid connection point, and control type of each transformer area in the county.

[0132] Based on the control type of the distribution area, the distribution area is divided into rigid distribution areas and flexible distribution areas;

[0133] The necessary system adjustments are determined based on the aggregated photovoltaic output, aggregated load power, and the allowable reverse power boundary at the grid connection point.

[0134] Construct a hybrid motion space consisting of rigid transformer switching actions and continuous adjustment weights for flexible transformers;

[0135] The virtual responsibility queue is updated based on the deviation between the base responsibility share of the transformer area and the actual power reduction.

[0136] A rigid candidate action set is generated based on the aforementioned responsibility virtual queue, the necessary system adjustment amount, and the previous switching state.

[0137] For each rigid candidate action in the rigid candidate action set, a corresponding flexible adjustment weight is generated;

[0138] The candidate mixed action, consisting of rigid switching action and flexible adjustment weight, is evaluated using a mixed action value evaluation network. The candidate mixed action with the highest long-term value is selected as the action to be executed in the current period.

[0139] The rigid station area is a station area that can only participate in reduction according to the switching, fixed ratio reduction, or graded response method.

[0140] The flexible distribution area is a distribution area that can continuously adjust its adjustment weight according to the scheduling signal and participate in the allocation of the remaining adjustment amount.

[0141] The necessary adjustment amount of the system is the non-negative part of the sum of the county-level aggregated photovoltaic output exceeding the county-level aggregated load power and the grid connection point's allowable reverse power boundary;

[0142] When the necessary adjustment amount of the system is greater than zero, the rigid and flexible transformer areas share the task of reduction.

[0143] The hybrid action space includes rigid station switching action and flexible station continuous adjustment weight;

[0144] The switching action value for rigid transformer areas is 0 or 1, while the continuous adjustment weight value for flexible transformer areas is between 0 and 1.

[0145] The rigid reduction amount is formed by the switching action and the fixed reduction ratio in the rigid transformer area;

[0146] After the rigid reduction is executed, the remaining adjustment amount is obtained, which is further allocated by the flexible distribution area based on the continuous adjustment weight and heterogeneous responsibility share.

[0147] The responsibility virtual queue is updated recursively based on the baseline reduction of responsibility and the actual reduction power of the transformer area in the current time period;

[0148] When the actual power reduction of a transformer area is lower than its baseline reduction responsibility, the virtual responsibility queue increases; when the actual power reduction of a transformer area is higher than its baseline reduction responsibility, the virtual responsibility queue is released.

[0149] The rigid candidate action set is formed by merging responsibility-priority candidates, power-matching candidates, and stability-maintaining candidates.

[0150] Among them, the responsibility priority candidate priority includes rigid transformer areas with higher responsibility virtual queue values, the power matching candidate priority makes the rigid reduction amount close to the necessary system adjustment amount, and the stability of the candidate priority reduces the switching changes of rigid transformer areas in adjacent time periods.

[0151] The set of rigid candidate actions satisfies rigid switching constraints, switching variation constraints, and power reduction boundary constraints.

[0152] When the number of candidates exceeds the set limit, priority is given to retaining actions that appear repeatedly in multiple candidate rules or that simultaneously have good responsibility release, power matching and switching stability characteristics.

[0153] The flexible adjustment weights are generated by the conditional flexible adjustment Actor; the conditional flexible adjustment Actor takes the current state, rigid candidate actions and corresponding remaining adjustment amounts as inputs, and outputs the continuous adjustment weights of the flexible station area.

[0154] The Hybrid Action Critic takes the current state, rigid candidate actions, and flexible adjustment weights as joint inputs and outputs the long-term value of the candidate hybrid actions.

[0155] The long-term value assessment of the hybrid action Critic simultaneously considers operating costs, excessive reduction, violation of grid connection point constraints, instantaneous liability fairness, intraday liability fairness, and rigid switching changes.

[0156] The final action to be executed is determined by the action with the highest value among the candidate mixed actions.

[0157] Example 12:

[0158] A fair regulation method for rigid-flexible hybrid control of distributed photovoltaic (PV) systems in county-level areas includes the following: obtaining PV output, load power, heterogeneous responsibility share, allowable reverse power boundary at grid connection point, and control type of each PV area; classifying PV areas into rigid and flexible PV areas based on their control type; calculating the necessary system regulation; constructing a hybrid action space consisting of rigid switching actions and flexible continuous regulation weights; characterizing cross-time period responsibility deviation of PV areas using a responsibility virtual queue; generating a rigid candidate set based on responsibility compensation, power matching, and switching stability; and selecting the final action to be executed through conditional flexible regulation and hybrid action value evaluation.

[0159] The necessary system adjustment for time period t is:

[0160]

[0161] In the formula: For time period The necessary adjustment amount; The maximum allowable reverse power at the grid connection point; .when At that time, there is no need to actively reduce photovoltaic power; when At that time, rigid and flexible transformer areas need to share the task of reduction.

[0162] Let the set of rigid transformer areas be denoted as The set of flexible transformer areas is denoted as Mixed actions are represented as:

[0163]

[0164] in, For rigid transformer area switching operations, the flexible transformer area operation is denoted as...

[0165] Rigid reduction areas are formed by a fixed reduction ratio:

[0166]

[0167] In the formula: The reduction ratio is fixed for the rigid distribution area. After the rigid action is executed, the remaining adjustment amount that needs to be adjusted by the flexible distribution area is:

[0168]

[0169] The virtual queue for transformer area responsibilities is updated recursively based on the difference between the baseline responsibility and the actual power reduction:

[0170]

[0171] In the formula: Taiwan District The responsibility of the virtual queue; Taiwan District The baseline responsibility corresponding to the current necessary adjustment amount; Taiwan District The actual power reduction; The larger the queue value, the higher the historical reduction responsibility arrears in that area, and the higher its priority should be in participating in reduction in the subsequent candidate action generation.

[0172] The set of rigid actions that can be entered into the candidate generation satisfies:

[0173]

[0174] In the formula: A rigid set of actions that can participate in candidate generation; For action The number of changes in the switching state relative to the previous moment; This represents the maximum number of switching operations allowed in a single time period. For action The corresponding rigid reduction amount. The formula is used to exclude three types of actions: first, actions that violate the rigid 0 / 1 control type; second, actions with excessive switching changes between adjacent time periods; and third, actions where the rigid control zone alone has already exceeded the necessary adjustment amount. This can reduce switching combinations that clearly lack engineering rationality before entering the value evaluation, making the candidate actions with adjustment significance in the subsequent Critic network centralized evaluation.

[0175] The three candidate categories correspond to different adjustment intentions: the responsibility-priority candidate allows rigid transformer areas with high responsibility deficits to participate; the power matching candidate ensures that the rigid reduction amount is close to the current adjustment demand; and the stability-maintaining candidate reduces switching changes between adjacent time periods. The three candidate categories are then combined to obtain the time period. Rigid candidate set:

[0176] ,

[0177] In the formula: It is a rigid set of candidate actions; For the first One rigid candidate action; This represents the number of candidate actions after merging and deduplication. The three types of candidate rules correspond to three adjustment intentions: responsibility compensation, power matching, and switching stability. Responsibility priority candidates allow rigid transformer areas with high historical responsibility deficits to participate; power matching candidates ensure that the rigid reduction amount is as close as possible to the current necessary adjustment amount; and stability maintenance candidates reduce execution disturbances caused by frequent switching between adjacent time periods. After the three types of candidates are generated, this paper merges and deduplicates the candidate actions; when the candidate size exceeds the set upper limit, actions that repeatedly appear in multiple rule types or simultaneously possess good responsibility release, power matching, and switching stability characteristics are prioritized for retention. Therefore, It is not a definite action given by a single rule, but a finite set of candidates for subsequent value evaluation.

[0178] For the k-th rigid candidate action in the rigid candidate set, the conditional flexible adjustment Actor outputs the flexible adjustment weights based on the current state, the rigid candidate action, and the remaining adjustment amount:

[0179] ,

[0180] In the formula: To flexibly adjust the Actor network; These are the parameters of the Actor network; For the first The flexible adjustment weights corresponding to the rigid candidate actions. The corresponding candidate hybrid actions are represented as follows:

[0181]

[0182] The Critic evaluates the long-term value of candidate hybrid actions:

[0183]

[0184]

[0185]

[0186] In the formula: Index the candidate actions with the highest value; For time period The final combined action.

[0187] The reward function takes the following form:

[0188]

[0189] In the formula: The reward scaling factor; Operating costs; For excessive reduction of penalties; To constrain the amount of violations; For the sake of instantaneous liability fairness; For intraday liability and fairness items; This is a rigid switching variation term; , , , For the corresponding weights.

[0190] When the environment is executed, it must satisfy the following constraints: rigid transformer area switching constraints, flexible transformer area continuous weight constraints, transformer area power reduction upper limit constraints, grid connection point reverse power boundary constraints, and rigid transformer area switching frequency constraints.

[0191] Example 13:

[0192] The verification of the fair regulation method for rigid-flexible hybrid control of distributed photovoltaic power in county-level areas includes the following:

[0193] This embodiment uses a county-level aggregation system comprising 75 distributed photovoltaic (PV) distribution areas as its object, with a data time resolution of 1 hour and 24 time periods per day. In the baseline calculation, the top 20% of distribution areas with higher reverse transmission pressure are designated as rigid distribution areas, corresponding to 15 rigid distribution areas and 60 flexible distribution areas. During the training phase, 90 operating days are selected to construct reinforcement learning interaction samples, and during the testing phase, 30 consecutive operating days are selected to evaluate the overall performance.

[0194] This embodiment sets up a rule-based rigid-flexible baseline, a continuous policy mapping baseline, a continuous projection reinforcement learning baseline, and compares them with the method of this invention. Table 1 shows a comparison of the overall performance of different methods.

[0195] Table 1. Comparison of overall performance of different methods

[0196] Regular rigid-flexible baseline 7.7566 0.1099 0.9336 90.286 1.6622 69.03 Continuous policy mapping baseline 7.7452 0.4267 0.5424 191.43 0.7289 89.84 CAP-RL 7.7478 0.4533 0.4995 190.66 0.8178 89.84 This article's method 7.7569 0.0274 0.9940 34.881 2.0000 56.58

[0197] As shown in Table 1, the responsibility completion rate of the method of this invention has a Gini coefficient of 0.0274, a Jain index of 0.9940, and an average responsibility cohort of 34.881 kWh at the end of the period, all of which are significantly better than the comparative methods. The operating cost is not much different from the baseline methods, indicating that the improvement in responsibility fairness has not brought about significant economic costs.

[0198] like Figure 1 As shown, the actual cumulative reduction in each transformer area under the method of this invention is generally close to the responsibility benchmark; as Figure 2 As shown, the responsibility completion curve of the method of the present invention is closer to 1, and the responsibility sharing among platforms is more balanced; as Figure 3 As shown, the power reduction is mainly concentrated during the midday peak photovoltaic period, and is distributed among multiple transformer substations. These results demonstrate that the present invention can coordinate the necessary regulation tasks under conditions where rigid and flexible transformer substations coexist, and reduce the likelihood of flexible substations bearing the primary responsibility for power reduction for extended periods.

[0199] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0200] 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 it. Those skilled in the art can modify or make equivalent substitutions to the specific embodiments of the present invention. Modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A fair regulation method for rigid-flexible hybrid control of distributed photovoltaic power in county-level areas, characterized in that, Includes the following steps: Obtain aggregated photovoltaic output, aggregated load power, heterogeneous responsibility share of each transformer area, allowed reverse power boundary of grid connection point, and control type of each transformer area across multiple transformer areas in the county. Based on the control type of the distribution area, the distribution area is divided into rigid distribution areas and flexible distribution areas; The necessary system adjustments are determined based on the aggregated photovoltaic output, aggregated load power, and the allowable reverse power boundary at the grid connection point. Construct a hybrid motion space consisting of rigid transformer switching actions and continuous adjustment weights for flexible transformers; The virtual responsibility queue is updated based on the deviation between the base responsibility of the transformer area and the actual power reduction. A rigid candidate action set is generated based on the aforementioned responsibility virtual queue, the necessary system adjustment amount, and the previous switching state. For each rigid candidate action in the set of rigid candidate actions, a corresponding flexible adjustment weight is generated; The candidate mixed action, consisting of rigid switching action and flexible adjustment weight, is evaluated using a mixed action value evaluation network. The candidate mixed action with the highest long-term value is selected as the action to be executed in the current period.

2. The fair regulation method for rigid-flexible hybrid control of county-level distributed photovoltaic power generation according to claim 1, characterized in that: The rigid station area is a station area that can only participate in reduction according to the switching, fixed ratio reduction, or graded response method. The flexible distribution area is a distribution area that can continuously adjust its adjustment weight according to the scheduling signal and participate in the allocation of the remaining adjustment amount.

3. The fair regulation method for rigid-flexible hybrid control of county-level distributed photovoltaic power generation according to claim 1, characterized in that: The necessary adjustment amount of the system is the non-negative part of the sum of the county-level aggregated photovoltaic output exceeding the county-level aggregated load power and the grid connection point's allowable reverse power boundary; When the necessary adjustment amount of the system is greater than zero, the rigid and flexible transformer areas share the task of reduction. The necessary system adjustments are as follows: In the formula: For time period The necessary adjustment amount; The maximum allowable reverse power at the grid connection point; .

4. The fair regulation method for rigid-flexible hybrid control of county-level distributed photovoltaic power generation according to claim 1, characterized in that: The hybrid action space is represented as: , For rigid switching operations, For continuous adjustment of the weight of flexible distribution areas; The switching action value for rigid transformer areas is 0 or 1, while the continuous adjustment weight value for flexible transformer areas is between 0 and 1. The rigid reduction amount is formed according to the switching action and fixed reduction ratio in the rigid distribution area, that is: In the formula: The reduction ratio is fixed for rigid transformer areas; The remaining adjustment amount after the rigid reduction is implemented is allocated by the flexible distribution area based on the continuous adjustment weight and heterogeneous responsibility share.

5. A fair regulation method for rigid-flexible hybrid control of distributed photovoltaic power in a county as described in claim 1, characterized in that: The responsibility virtual queue is updated recursively based on the baseline reduction of responsibility and the actual reduction power of the transformer area in the current time period; When the actual power reduction in a transformer area is less than the baseline reduction responsibility, the virtual responsibility queue increases; when the actual power reduction in a transformer area is greater than the baseline reduction responsibility, the virtual responsibility queue is released. The virtual queue for district responsibilities has been updated as follows: In the formula: Taiwan District The responsibility of the virtual queue; Taiwan District The baseline responsibility corresponding to the current necessary adjustment amount; Taiwan District The actual power reduction; .

6. The fair regulation method for rigid-flexible hybrid control of county-level distributed photovoltaic power generation according to claim 1, characterized in that: The rigid candidate action set includes responsibility priority candidates, power matching candidates, and stability maintenance candidates.

7. A fair regulation method for rigid-flexible hybrid control of distributed photovoltaic power in a county-level area according to claim 7, characterized in that: The set of rigid candidate actions satisfies rigid switching constraints, switching variation constraints, and power reduction boundary constraints, namely: In the formula: A rigid set of actions that can participate in candidate generation; For action The number of changes in the switching state relative to the previous moment; This represents the maximum number of switching operations allowed in a single time period. For action The corresponding rigid reduction amount.

8. A fair regulation method for rigid-flexible hybrid control of distributed photovoltaic power in a county as described in claim 1, characterized in that: The flexible adjustment weights are generated by the conditional flexible adjustment Actor; The conditional flexible adjustment Actor takes the current state, rigid candidate actions, and corresponding remaining adjustment amounts as inputs and outputs the continuous adjustment weight of the flexible transformer area. The continuous adjustment weights for flexible transformer areas are shown below: , In the formula: To flexibly adjust the Actor network; These are the parameters of the Actor network; For the first The flexible adjustment weights corresponding to each rigid candidate action; The Hybrid Action Critic takes the current state, rigid candidate actions, and flexible adjustment weights as joint inputs and outputs the long-term value of the candidate hybrid actions. The long-term value of the candidate mixed actions is as follows: In the formula: Index the candidate actions with the highest value; For time period The final combined action.

9. A fair regulation method for rigid-flexible hybrid control of distributed photovoltaic power in a county as described in claim 8, characterized in that: The reward function for the hybrid action Critic is used to guide the Critic to learn the long-term value of candidate hybrid actions. The reward function for the hybrid action Critic is as follows: In the formula: The reward scaling factor; Operating costs; For excessive reduction of penalties; To constrain the amount of violations; For the sake of instantaneous liability fairness; For intraday liability and fairness items; This is a rigid switching variation term; , , , For the corresponding weights.

10. A fair regulation system for rigid-flexible hybrid control of distributed photovoltaic power in a county-level area based on the method of any one of claims 1-9, characterized in that: It includes a data acquisition module, a transformer area control type identification module, a necessary adjustment amount calculation module, a responsibility virtual queue update module, a rigid candidate action generation module, a conditional flexible adjustment module, a mixed action value evaluation module, and an adjustment strategy output module; The data acquisition module is used to acquire photovoltaic output, load power, heterogeneous liability share, and grid connection point boundary; the transformer area control type identification module is used to classify rigid transformer areas and flexible transformer areas. The necessary adjustment calculation module is used to calculate the necessary adjustment of the system in the current time period; The responsibility virtual queue update module is used to recursively update the responsibility deviation of the transformer area. The rigid candidate action generation module is used to generate a finite-sized set of rigid candidate actions. The conditional flexible adjustment module is used to generate corresponding flexible adjustment weights for each rigid candidate action. The hybrid action value evaluation module is used to evaluate the long-term value of candidate hybrid actions. The adjustment strategy output module is used to output the current time period's rigid transformer switching status, flexible transformer adjustment weight, and actual power reduction of each transformer.

Images