Low-voltage distributed photovoltaic group regulation and control method and system based on multi-source information fusion

Abstract

The application discloses a low-voltage distributed photovoltaic group regulation and control method and system based on multi-source information fusion, relates to the photovoltaic group regulation and control technical field, and comprises the following steps: obtaining the operation information of each photovoltaic node in a low-voltage distribution network and converting the operation information into a state vector; generating a photovoltaic group control instruction set based on the state vector and a system optimization target, and establishing an expected physical behavior model; monitoring the actual behavior trajectory of each node after execution of the instruction, comparing the actual behavior trajectory with the expected physical behavior model, constructing a response difference atlas, and quantifying a response deviation; for the node with the response deviation, judging the difference type according to the deviation atlas, generating a compensation instruction by reconstructing a local optimization target based on a type matching correction strategy, and outputting a correction path; and inputting historical behavior deviation samples and the correction path into a graph structure time sequence learning model to update node control weights and regulation and control priorities. The application can accurately identify control strategy execution deviation by constructing a response difference atlas based on an actual behavior trajectory.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic group regulation and control, and more particularly to a low-voltage distributed photovoltaic group regulation and control method and system based on multi-source information fusion. BACKGROUND

[0002] With the large-scale access of distributed photovoltaics in low-voltage distribution networks, the demand for group regulation and control is increasingly prominent. Traditional photovoltaic access control strategies are mostly configured statically in single nodes, which are difficult to adapt to overall voltage fluctuations and power imbalances caused by system load fluctuations, inverter dynamic characteristics, and multi-node coupled responses. Therefore, a group regulation and control method that supports distributed sensing, collaborative decision-making, and group execution is urgently needed to achieve unified scheduling and stable operation of photovoltaic resources.

[0003] However, existing photovoltaic group control methods are mostly based on idealized synchronous execution assumptions and do not fully consider the non-ideal characteristics of the "sensing-computing-communication-execution" link in actual operation in low-voltage distribution networks. Especially in multi-source control collaboration scenarios, due to differences in measurement time delay, communication delay, edge computing processing time, and device response speed of each photovoltaic node, the timestamps in the system control closed loop are inconsistent, which may cause the following problems: control actions lag behind disturbance changes, losing the timeliness of regulation; control strategies and disturbance signals are in phase opposition at the same frequency, causing negative feedback failure or even system oscillation; the expected execution state of the control command does not match the actual physical behavior, gradually accumulating "behavior drift" phenomena. The above problems not only reduce control accuracy and group response consistency, but also may damage the voltage stability and power flow collaboration of the entire low-voltage network.

[0004] For example, the invention patent with publication number CN117748569B discloses a low-voltage distributed photovoltaic control method, device, and medium. The method includes: receiving the power grid peak regulation index and monitoring the operation information of photovoltaic users and energy storage devices in the control area in real time; calculating the control target according to the power grid peak regulation index and decomposing it to generate the substation control index of multiple substations; in the case of configuring energy storage devices in the substation, charging or discharging the energy storage devices according to the substation control index and the total charging power of the energy storage devices in the operation information to generate a first day-ahead control scheme; otherwise, determining the control sequence according to the control priority of the photovoltaic users in the operation information and generating a second day-ahead control scheme according to the preset control mechanism and the control sequence; generating an optimal third day-ahead control scheme according to the first day-ahead control scheme and the second day-ahead control scheme, and executing the third day-ahead control scheme to complete the low-voltage distributed photovoltaic control of the control area.

[0005] For example, the invention patent announcement No. CN118367608B announced the maximum access capacity evaluation method of distributed photovoltaic control mode, relates to the field of distributed photovoltaic control, solves the problem of accurate calculation of the maximum access capacity of multi-model multiplex photovoltaic networking, and the maximum access capacity evaluation method is as follows: the data acquisition module acquires the periodic photovoltaic data of each photovoltaic node in the photovoltaic networking; the node discrimination module comprehensively analyzes the node type of each photovoltaic node in the photovoltaic networking, and analyzes the node type of the photovoltaic node to be sent to the control terminal through the server; the control terminal adjusts and controls the control mode of each photovoltaic node according to the node type; the data acquisition module acquires the periodic photovoltaic data of the balance node in the photovoltaic networking; the power flow analysis module analyzes the photovoltaic output of each photovoltaic node in the photovoltaic networking, and analyzes the maximum access capacity of each photovoltaic node and the maximum access capacity of the photovoltaic networking, so that the accurate calculation of the maximum access capacity of multi-model multiplex photovoltaic networking is realized.

[0006] In the above disclosed technical solution, at least the following technical problems exist: due to different response delays of nodes in the perception-computation-communication-execution link, the timestamps in the control closed loop are inconsistent, which may cause the control action to "lag behind the disturbance", and even to be opposite in frequency to the disturbance, and there is inconsistency between the expected execution state of the control strategy and the actual adjustment behavior, resulting in "behavior drift" of the group control system in the execution closed loop.

[0007] In view of the above problems, the present application provides a solution. SUMMARY

[0008] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a low-voltage distributed photovoltaic group regulation and group control method and system based on multi-source information fusion, which fuses the operating state of the photovoltaic node, the control strategy parameters, the boundary disturbance information and the historical response trajectory, constructs a unified state vector model and an expected physical behavior model, and combines the response difference atlas to identify and adjust the control deviation, so as to solve the problems of control response lag, inconsistency between control strategy and physical behavior, and behavior drift of the group control system caused by asynchronous of the perception-computation-communication-execution link in the prior art.

[0009] To achieve the above purpose, the present application provides the following technical solutions:

[0010] The low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion comprises the following steps: obtaining the operation information of each photovoltaic node in the low-voltage distribution network and converting it into a state vector; based on the state vector and the pre-obtained system optimization target, a photovoltaic group control instruction set is generated, and an expected physical behavior model of the control instruction is established; the actual behavior trajectory of each node after the execution of the instruction is monitored, and a response difference atlas is constructed by comparing it with the expected physical behavior model, so as to quantitatively identify the response deviation of the control strategy and the execution state; for the nodes with response deviation, the difference type is judged according to the deviation atlas, the strategy is corrected based on the type matching, the compensation instruction is generated by reconstructing the local optimization target, and the correction path is output; the historical behavior deviation sample and the correction path are input into a graph structure time sequence learning model, the node control weight and the regulation and control priority are updated, and a closed-loop adaptive strategy migration mechanism is formed.

[0011] In a preferred embodiment, the operation information of each photovoltaic node in the low-voltage distribution network is obtained and converted into a state vector, specifically: the operation information of each photovoltaic node in the low-voltage distribution network is obtained, the operation information includes real-time voltage, current, active and reactive power, inverter state, irradiance and predicted load data; the control function information supported by the photovoltaic node is synchronously obtained, the control function information is parsed into a structured capability identification vector, and the node state vector is constructed together with the operation parameters.

[0012] In a preferred embodiment, the state vector and the pre-obtained system optimization target are used to generate a photovoltaic group control instruction set, specifically: the state vectors of a plurality of photovoltaic nodes are combined into a state matrix, the state vectors contain node-level operation parameters and capability identification; a multi-objective regulation and control function is constructed based on the system optimization target, the optimization target includes node voltage stability, active power tracking error minimization, system loss minimization or node voltage coordination; the state matrix is input into a layered optimization engine to output a control instruction set, the engine includes a global coordination layer and a local strategy generation layer, which are respectively used to obtain group-level control boundaries and node-level control solutions.

[0013] In a preferred embodiment, the expected physical behavior model of the control instruction is established, specifically: the control instruction set is divided according to nodes, and control variables of each node are extracted, including the set active power value, the reactive power value and the voltage control coefficient; the adjustment target information corresponding to the control instruction is constructed based on the control variable and the current state vector; the network voltage distribution and the node voltage change trend after control execution are predicted based on the power distribution network topology structure, the electrical impedance parameters and the node state; the active power output change and the reactive power change after the control instruction triggering are predicted in combination with the photovoltaic component output characteristics and the inverter response model; the node state evolution path model after control execution is established based on the node response rate and the historical state transition law; and the expected physical behavior model is generated by fusing the network voltage distribution, the node voltage change trend, the active power output change, the reactive power output change and the node state evolution path.

[0014] In a preferred embodiment, the actual behavior trajectory of each node after the instruction execution is monitored, and a response difference atlas is constructed by comparing the actual behavior trajectory with the expected physical behavior model, specifically: the actual behavior trajectory of each photovoltaic node after the control instruction execution is obtained, and is aligned with the theoretical behavior trajectory generated based on the expected physical behavior model on the time axis; the deviation characteristics between the actual behavior trajectory and the theoretical behavior trajectory are extracted based on the time series analysis method, including the response amplitude difference, the response time lag, the dynamic change rate error and the steady-state offset; the electrical connection graph and the control strategy coupling relationship graph among the photovoltaic nodes are constructed, the response deviation propagation path in the group is modeled in combination with the node space adjacency and the response cooperativity, and the error propagation strength between the nodes is quantified; the deviation characteristics of each node are taken as the graph node attributes, and the error propagation strength is taken as the edge weight, so as to construct the response difference atlas.

[0015] In a preferred embodiment, the response deviation of the control strategy and the execution state is quantified and identified, specifically: the multi-dimensional response deviation characteristics of each photovoltaic node and the error propagation edge weight between the nodes are extracted based on the response difference atlas; the nodes are divided into different response deviation levels by clustering analysis based on the deviation characteristics; the control strategy parameter distribution of the nodes in different deviation levels is analyzed, and the key parameter dimension associated with the deviation level is determined; the state response sensitivity of each node under the change of the key control parameter is quantitatively modeled to generate a regulation sensitivity vector; and the evaluation index is constructed in combination with the response deviation level, the key parameter dimension, the regulation sensitivity vector and the error propagation path, so as to identify the nodes with serious mismatch between the control strategy and the execution state and mark them as deviation source nodes.

[0016] In a preferred embodiment, the compensation instruction is generated by reconstructing the local optimization target, and the output correction path is specifically: according to the response difference graph and the node difference type, a local compensation area containing the bias source node and its adjacent influence nodes is selected; an optimization objective function of the local area is constructed by minimizing the node voltage deviation, reducing the power loss and adjusting the cost constraint; the optimal adjustment compensation amount of each photovoltaic node in the target area is obtained by solving the optimization objective function; the updated control instruction set is generated based on the optimal adjustment compensation amount, and the control instruction set and the node state evolution process are encapsulated as a correction path, and the correction path contains the time sequence corresponding relationship between the adjustment input and the response state.

[0017] In a preferred embodiment, the historical behavior deviation sample and the correction path are input into the graph structure time sequence learning model to update the node control weight and the regulation priority, and a closed-loop adaptive strategy migration mechanism is formed, and the specific steps are as follows: a time sequence sample set is constructed, the response difference graph is trained based on the graph structure time sequence learning model, the dynamic evolution mode of the control strategy deviation propagation between nodes is learned; the edge weight value in the response difference graph is updated according to the training result, and the updated edge weight value is used as the basis for allocating the control weight and the regulation priority between nodes, so as to drive the adaptive migration of the control strategy and the adjustment of the adjustment scheme.

[0018] In a preferred embodiment, the time sequence sample set is constructed, and the specific steps are as follows: based on the response difference graph, the edge weight information between nodes is extracted as the initial representation of the control strategy conduction strength; according to the historical behavior deviation sample and the correction path, a time sequence sample set of the node control input sequence and the state response sequence is constructed, and a mapping relationship with the response difference graph node is established.

[0019] The technical effects and advantages of the low-voltage distributed photovoltaic group regulation and control method and system based on multi-source information fusion are as follows:

[0020] 1. The present application realizes the individualized construction of the state vector and the dynamic mapping of the control interface by introducing the joint modeling mechanism of the photovoltaic node operation parameter and the control ability, significantly improves the instruction adaptability and group control coordination under the condition of heterogeneous control ability, and effectively avoids the problems of instruction failure and behavior conflict caused by inconsistent control ability in the traditional method.

[0021] 2.The method can accurately identify control strategy execution deviation by constructing a desired physical behavior model of control instructions and constructing a response difference atlas based on actual behavior trajectories, and can implement targeted adjustment and compensation in combination with node difference types and error propagation paths; and further utilizes a graph structure time sequence learning method to dynamically learn historical deviation samples and correction paths, update control weights and regulation priorities, and form a continuous adaptive strategy migration mechanism. The method enhances the perception and response ability of the system to disturbances, delays and behavior drift, realizes the transformation of the photovoltaic group control strategy from static configuration to dynamic evolution, and significantly improves the intelligence, robustness and stability of the photovoltaic group operation in the low-voltage distribution network. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A flowchart of the low-voltage distributed photovoltaic group regulation and group control method based on multi-source information fusion of the present application is shown.

[0023] Figure 2 A structure diagram of the low-voltage distributed photovoltaic group regulation and group control system based on multi-source information fusion of the present application is shown. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0025] Embodiment 1, Figure 1 The low-voltage distributed photovoltaic group regulation and group control method based on multi-source information fusion of the present application is given, including the following steps:

[0026] S1, obtaining the operation information of each photovoltaic node in the low-voltage distribution network and converting it into a state vector;

[0027] In this embodiment, the operation information of each photovoltaic node in the low-voltage distribution network is obtained and converted into a state vector, specifically as follows:

[0028] The operation information of the photovoltaic node is obtained, and the operation information includes real-time voltage, current, active and reactive power, inverter state, irradiance and predicted load data;

[0029] Synchronously obtain the control function information supported by the photovoltaic node, including whether to support constant power control, constant voltage control, PQ curve control, reactive power dynamic response control or have power change rate limitation;

[0030] The control function information is parsed into a structured capability identification vector, and the node state vector is constructed together with the operation parameters.

[0031] and based on the capability identification vector, dynamically adjusting the dimension, feature position and control interface mapping mode of the state vector, so that the subsequent control strategy generation module can realize instruction adaptive compatibility for photovoltaic nodes with different control capabilities.

[0032] The capability identification vector is used to dynamically adjust the dimension, feature position and control interface mapping mode of the state vector, specifically:

[0033] According to the control functions supported by each photovoltaic node, a capability identification vector is constructed, and the control functions include but are not limited to: whether to support reactive power regulation, whether to support power change rate limitation, whether to support target voltage setting;

[0034] Based on the capability identification vector, the corresponding feature fields in the enabled or disabled state vector are screened, and the fields without control capability are removed or set to zero, thereby forming a node-specific state vector dimension;

[0035] Each feature field in the state vector is assigned position index information or rearranged in order to ensure the consistency of feature analysis of the input model under the heterogeneous state structure;

[0036] In the process of generating and issuing control instructions, based on the capability identification vector and the state vector structure, a node-level control interface mapping relationship is established to ensure that the control instructions are only issued to nodes with corresponding execution capabilities, avoiding control failure or behavior conflict.

[0037] In the process of constructing the state vector, the system enables or shields the feature fields in the state vector according to the control capability labels of each node, and constructs a variable-dimension input structure in the form of capability mask; at the same time, in order to ensure the subsequent model recognition accuracy, the position of the state feature is rearranged or identified coding; finally, the control strategy generation module dynamically matches the corresponding control interface according to the capability label, thereby realizing the individualization of the node-level strategy structure and the consistency of the group-level control.

[0038] The capability identification vector is specifically:

[0039]

[0040] Among them, is a set of controllable capabilities of the photovoltaic node, is whether to support reactive power regulation, is whether to support constant voltage control, is the capability identification vector.

[0041] The state vector is specifically:

[0042]

[0043]

[0044]

[0045] wherein, is the original operating parameter of the node, is the voltage, is the current, is the capability identification vector is mapped into the function feature field, is the Hadamard product, is the node state vector.

[0046] S2, based on the state vector and the pre-acquired system optimization target, generates a photovoltaic group control control instruction set, and establishes an expected physical behavior model of the control instruction;

[0047] The photovoltaic group control control instruction set is generated based on the state vector and the pre-acquired system optimization target, and specifically:

[0048] The state vectors of a plurality of photovoltaic nodes are combined into a state matrix as an input of the current group operating state, and the state matrix includes node-level operating parameters and capability identification;

[0049] A multi-objective regulation function is constructed based on the system optimization target, and the optimization target includes node voltage stability, active power tracking error minimization, system loss minimization, or node voltage coordination;

[0050] The state matrix is input into a hierarchical optimization engine, and finally a control instruction set including active power output set value, reactive power output set value, and voltage response coefficient control parameter is output, and the engine includes a global coordination layer and a local strategy generation layer, which are respectively used to obtain group-level control boundaries and node-level control solutions.

[0051] Further, in the process of generating each node control solution, the executable control variable combination is selected according to the capability identification vector of the corresponding node, and the uncontrollable or restrictive field is eliminated, so as to ensure the physical feasibility of the control solution.

[0052] The multi-objective regulation function is specifically:

[0053]

[0054]

[0055]

[0056]

[0057]

[0058] wherein, is a multi-objective regulation function, , , , is a weight coefficient (set according to historical experience), is voltage stability, is active power tracking error, is system loss, is voltage coordination between nodes, is the voltage of node i, is the rated voltage, is the actual active power output of the node, is the target active power output of the node, is the line resistance, is the line current, is the line set, is the total number of nodes.

[0059] The expected physical behavior model of the control instruction is established, specifically:

[0060] The control instruction set is divided according to nodes, and the control variables of each node are extracted, including the set active power value, reactive power value, and voltage control coefficient;

[0061] Based on the control variables of the nodes and the current state vector, the regulation target information corresponding to the control instruction is constructed;

[0062] Based on the pre-acquired distribution network topology structure parameters, electrical impedance information, and node state, the network voltage distribution and node voltage change trend after control execution are predicted;

[0063] Combined with the dynamic output model of the photovoltaic component and the response characteristics of the inverter, the active and reactive output changes after control instruction execution are obtained;

[0064] Based on the node response rate and state transition relationship, a node state evolution path model after control execution is established;

[0065] The network voltage distribution, node voltage change trend, active and reactive output changes, and state evolution path are summarized to obtain the expected physical behavior model.

[0066] The regulation target information refers to the voltage, current, or power regulation direction and amplitude that the control instruction intends to achieve;

[0067] State evolution path model: refers to the prediction process of the change of node state over time after control execution, which can be modeled by discrete sequence or continuous differential equation;

[0068] Desired physical behavior model: refers to a theoretical model that comprehensively predicts the control execution results, covers voltage, current, power and system linkage effects, and describes the theoretical voltage response after regulation (network voltage distribution, node voltage change trend), power distribution and node state evolution path.

[0069] S3, monitor the actual behavior trajectory of each node after the execution of the instruction, and compare it with the desired physical behavior model to build a response difference graph, and quantify and identify the response deviation between the control strategy and the execution state;

[0070] The response difference graph built by comparing with the desired physical behavior model is specifically:

[0071] The actual behavior trajectory of each photovoltaic node after the execution of the control instruction is obtained, and is aligned with the theoretical behavior trajectory generated based on the desired physical behavior model on the time axis;

[0072] Based on the multi-scale time series analysis method, the deviation features between the actual behavior trajectory and the theoretical behavior trajectory of the node are extracted, and the deviation features include response amplitude difference, response time lag, dynamic change rate error and steady-state offset;

[0073] An electrical connection graph and a control strategy coupling relationship graph of photovoltaic nodes are built, and based on the spatial adjacency and response cooperativity between nodes, the propagation path of each response deviation in the group range is modeled to obtain the error propagation strength between nodes;

[0074] The deviation features of each node are taken as the node attribute, and the error propagation strength between nodes is taken as the edge weight, to build a response difference graph.

[0075] The error propagation strength is specifically:

[0076]

[0077] Wherein, is the error propagation strength from node i to node j, is the response feature difference value of node i at time t, is the response feature difference value of node j at time t, is the length of the time window, is the starting time of the time window, is the current time.

[0078] The response deviation between the control strategy and the execution state is specifically:

[0079] Based on the response difference graph, multi-dimensional response deviation features of each photovoltaic node are extracted, including voltage, power response error, instruction response delay, fluctuation amplitude, and error propagation edge weight between nodes;

[0080] Based on the deviation characteristics, the node response behavior patterns are identified by a clustering method and divided into different categories, defined as response deviation levels, including: normal response, slight deviation and serious deviation three levels;

[0081] For nodes with different response levels, the distribution of their control strategy parameters (such as PQ instruction type, voltage set value, regulation interface type) is analyzed to identify the key parameter dimensions associated with the deviation level;

[0082] The state response sensitivity of each node under the change of key control parameters is modeled to form a regulation sensitivity vector, which is used to evaluate the stability and regulation risk of the node under the current strategy;

[0083] Combining the response deviation level, key parameter dimension, regulation sensitivity vector and error propagation path, the deviation source node evaluation index is constructed to identify the nodes with serious mismatch between control strategy and execution state, and marked as deviation source nodes.

[0084] S4, for the nodes with response deviation, the difference type is judged according to the deviation map, the strategy is corrected based on the type matching, the compensation instruction is generated by reconstructing the local optimization target, and the correction path is output;

[0085] The node difference class includes control interface execution failure, prediction error amplification and neighborhood interference;

[0086] The node with response deviation is judged according to the deviation map, and the correction strategy is matched based on the type, which is:

[0087] The node response deviation is divided into at least one of the three difference types of control interface execution failure, prediction error amplification and neighborhood interference;

[0088] For the control interface execution failure type, control command retransmission or interface fault switching correction strategy is adopted;

[0089] For the prediction error amplification type, model parameter online correction or prediction error feedback adjustment strategy is adopted;

[0090] For the neighborhood interference type, neighborhood collaborative filtering or local disturbance suppression adjustment strategy is adopted.

[0091] The compensation instruction is generated by reconstructing the local optimization target, and the correction path is output, which is:

[0092] According to the response difference map and the node difference type, the local area containing the deviation source node and its neighborhood influence node which needs to be adjusted and compensated is selected;

[0093] The optimization objective function of the local area is constructed by minimizing the node voltage deviation, reducing power loss and adjusting cost constraint;

[0094] By solving the optimization objective function, the optimal adjustment compensation of each photovoltaic node in the target area is obtained;

[0095] Based on the optimal adjustment compensation, an updated control instruction set is generated, and the control instruction set and its corresponding node state evolution process are encapsulated as a correction path, which describes the control input-response timing chain of the node under adjustment compensation.

[0096] Further, the voltage deviation minimization, power loss reduction and adjustment cost constraint are specifically:

[0097] Voltage deviation minimization: by adjusting the active and reactive power output of each node, the deviation between the node voltage and the set reference value is reduced , is the compensated node voltage, is the rated voltage;

[0098] Power loss reduction : considering the local line power transmission efficiency, the power distribution is optimized to reduce the overall energy loss;

[0099] Adjustment cost constraint: combined with the operation limit and economic cost of energy storage equipment and photovoltaic inverter, the adjustment load is reasonably distributed to avoid excessive adjustment.

[0100] The optimization objective function is specifically:

[0101]

[0102] The constraint condition is:

[0103]

[0104]

[0105]

[0106] wherein, is the compensated node voltage, is the rated voltage, is the branch resistance, is the compensated branch current, is the node set in the selected local area, is the branch set in the area, , is the preset node adjustment cost coefficient, is the active compensation of node i, Let be the reactive power compensation amount for node i. For partial cost budgeting, , These are the minimum and maximum allowable voltages, respectively. , The minimum and maximum allowable active power compensation amounts are... , To compensate for those who contributed effort, whether they achieved merit or not. , These are the minimum and maximum permissible reactive power compensation amounts, respectively. It is the set of adjustable points.

[0107] The correction path is specifically as follows:

[0108]

[0109] in, For the complete corrected path of node i, Let k be the time for the adjustment. For the control input of the k-th adjustment compensation, Let n be the state response of node i after adjustment, and n be the number of adjustment steps experienced during the adjustment and compensation process.

[0110] S5 inputs historical behavior deviation samples and correction paths into the graph structure time-series learning model, updates node control weights and regulation priorities, and forms a closed-loop adaptive policy transfer mechanism.

[0111] The process of inputting historical behavioral deviation samples and correction paths into the graph structure time-series learning model, and updating node control weights and regulation priorities, specifically involves:

[0112] Based on the response difference map, the edge weight information between nodes is extracted as an initial representation of the transmission strength of the control strategy;

[0113] Based on historical behavior deviation samples and correction paths, a time-series sample set of node control input sequences and state response sequences is constructed, and node-level correspondence is established with the response difference map.

[0114] The graph is trained based on a graph structure temporal learning model to learn the dynamic evolution pattern of the propagation of control strategy deviations between nodes.

[0115] Based on the learning results, the edge weights in the response difference graph are dynamically updated, and the updated edge weights are used as the basis for the priority of regulation and the allocation of control weights between nodes, which drives the strategy optimization module to realize the adaptive migration of control strategies and the adjustment of regulation schemes.

[0116] Example 2, Figure 2The low-voltage distributed photovoltaic group regulation and control system based on multi-source information fusion of the application comprises the following modules:

[0117] A state vector construction module is configured to acquire the operation information of each photovoltaic node in the low-voltage distribution network and convert the operation information into a state vector;

[0118] A group control instruction generation module is configured to generate a photovoltaic group control instruction set based on the state vector and a pre-acquired system optimization target, and establish an expected physical behavior model of the control instruction;

[0119] A response difference analysis module is configured to monitor the actual behavior trajectory of each node after the execution of the instruction, compare the actual behavior trajectory with the expected physical behavior model to construct a response difference atlas, and quantitatively identify the response deviation of the control strategy and the execution state;

[0120] A difference type discrimination and compensation module is configured to, for the node with the response deviation, judge the difference type according to the deviation atlas, correct the strategy based on the type matching, generate a compensation instruction by reconstructing a local optimization target, and output a correction path;

[0121] A strategy migration optimization module is configured to input the historical behavior deviation sample and the correction path into a graph structure time sequence learning model, update the node control weight and the regulation and control priority, and form a closed-loop adaptive strategy migration mechanism.

[0122] The above formulas are all dimensionless values, and the formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0123] The above embodiments can be realized wholly or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized in the form of a computer program product wholly or partially.

[0124] Those skilled in the art can realize that the modules and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software depends on the specific application and design constraints of the technical solutions. The person skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0125] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically independently, or two or more modules can be integrated in one module.

[0126] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0127] Finally, the above merely provides the preferred embodiments of the present application, but is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion, characterized in that, The method comprises the following steps: acquiring operation information of each photovoltaic node in a low-voltage power distribution network and converting the operation information into a state vector; based on the state vector and a pre-acquired system optimization target, generating a photovoltaic group control control instruction set, and establishing an expected physical behavior model of the control instruction, specifically: extracting control variables of each node; based on the control variables and the current state vector, constructing adjustment target information; based on the power distribution network topology structure, electrical impedance parameters and node state, predicting network voltage distribution and node voltage change trend after control execution; combining photovoltaic component output characteristics and inverter response model, predicting active and reactive power output change; based on the node response rate and historical state transition law, establishing a node state evolution path model; combining the network voltage distribution, the node voltage change trend, the active and reactive power output change and the node state evolution path, generating the expected physical behavior model; monitoring actual behavior trajectories of each node after execution of the instruction, and comparing the actual behavior trajectories with the expected physical behavior model to construct a response difference atlas, specifically: acquiring actual behavior trajectories of each photovoltaic node after execution of the control instruction, and time-aligning the actual behavior trajectories with theoretical behavior trajectories; extracting deviation characteristics between the actual behavior trajectories and the theoretical behavior trajectories; constructing an electrical connection diagram between photovoltaic nodes and a control strategy coupling relationship diagram, combining node spatial adjacency and response cooperativity, modeling a propagation path of the response deviation in the group, quantifying error propagation strength between nodes; taking the deviation characteristics as node attribute of the diagram, and taking the error propagation strength as edge weight, to construct the response difference atlas; quantifying and identifying response deviation of the control strategy and the execution state; for the nodes with response deviation, judging the difference type according to the deviation atlas, correcting the strategy based on the type matching, generating a compensation instruction through reconstruction of a local optimization target, and outputting a correction path; inputting historical behavior deviation samples and the correction path into a graph structure time sequence learning model, updating node control weight and regulation priority, and forming a closed-loop adaptive strategy migration mechanism. 2.The low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion according to claim 1, characterized in that, The operation information of each photovoltaic node in the low-voltage power distribution network is acquired and converted into a state vector, specifically: acquiring operation information of each photovoltaic node in a low-voltage power distribution network, the operation information including real-time voltage, current, active and reactive power, inverter state, irradiance and predicted load data; synchronously acquiring control function information supported by the photovoltaic node, parsing the control function information into a structured capability identification vector, and constructing a node state vector together with the operation parameters. 3.The low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion according to claim 2, characterized in that, Based on the state vector and the pre-acquired system optimization target, the photovoltaic group control control instruction set is generated, specifically: combining state vectors of multiple photovoltaic nodes into a state matrix, the state vector containing node-level operation parameters and capability identification; based on the system optimization target, constructing a multi-objective regulation function, the optimization target including node voltage stability, active power tracking error minimization, system loss minimization or node voltage coordination; inputting the state matrix into a layered optimization engine to output a control instruction set, the engine including a global coordination layer and a local strategy generation layer, respectively used for acquiring group-level control boundaries and node-level control solutions.

4. The low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion according to claim 3, characterized in that, The control variables include set active power output value, reactive power output value and voltage control coefficient; Based on the control variable and the current state vector, the adjustment target information corresponding to the control instruction is constructed; Based on the node response rate and the historical state transition law, a node state evolution path model after control execution is established.

5. The low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion according to claim 4, characterized in that: Based on the time series analysis method, the deviation characteristics between the actual behavior trajectory and the theoretical behavior trajectory are extracted, including the response amplitude difference, the response time lag, the dynamic change rate error and the steady-state offset.

6. The low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion according to claim 5, characterized in that, The response deviation of the quantized identification control strategy and the execution state is specifically: Based on the response difference graph, the multi-dimensional response deviation characteristics of each photovoltaic node and the error propagation edge weight between nodes are extracted; Based on the deviation characteristics, the nodes are divided into different response deviation levels through cluster analysis; The control strategy parameter distribution of different deviation level nodes is analyzed to determine the key parameter dimension associated with the deviation level; The state response sensitivity of each node to the change of the key control parameter is quantitatively modeled to generate a regulation sensitivity vector; Combining the response deviation level, the key parameter dimension, the regulation sensitivity vector and the error propagation path, an evaluation index is constructed to identify the nodes with serious mismatch between the control strategy and the execution state and mark them as deviation source nodes.

7. The low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion according to claim 6, characterized in that, The compensation instruction is generated by reconstructing the local optimization target, and the correction path is output, specifically: According to the response difference graph and the node difference type, a local compensation area containing the deviation source node and its adjacent influence nodes is selected; The optimization objective function of the local area is constructed by minimizing the node voltage deviation, reducing power loss and constraining the adjustment cost; By solving the optimization objective function, the optimal adjustment compensation amount of each photovoltaic node in the target area is obtained; Based on the optimal adjustment compensation amount, an updated control instruction set is generated, and the control instruction set and the node state evolution process are packaged as a correction path, which contains the time sequence correspondence between the adjustment input and the response state. 8.The low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion according to claim 7, characterized in that, The historical behavior deviation sample and the correction path are input into the graph structure time series learning model to update the node control weight and the regulation priority, forming a closed-loop adaptive strategy migration mechanism, specifically: Construct a time series sample set, train the response difference graph based on the graph structure time series learning model, and learn the dynamic evolution mode of the control strategy deviation propagation between nodes; According to the training result, update the edge weight value in the response difference graph, and use the updated edge weight value as the basis for allocating the regulation priority and the control weight between nodes, to drive the adaptive migration of the control strategy and the adjustment of the regulation scheme. 9.The low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion of claim 8, characterized in that, The steps of constructing the time series sample set are as follows: Based on the response difference graph, the edge weight information between nodes is extracted as the initial representation of the control strategy transmission strength; According to the historical behavior deviation sample and the correction path, a time series sample set of node control input sequence and state response sequence is constructed, and a mapping relationship with the response difference graph node is established.

10. A system using the low-voltage distributed photovoltaic group regulation and control method based on multi-source information fusion according to any one of claims 1-9, characterized in that, The following modules are included: State vector construction module: used to obtain the operation information of each photovoltaic node in the low-voltage distribution network and convert it into a state vector; The group control instruction generation module is configured to generate a photovoltaic group control instruction set based on the state vector and a pre-acquired system optimization target, and establish an expected physical behavior model of the control instruction; The response difference analysis module is configured to monitor actual behavior trajectories of the nodes after execution of the instructions, compare the actual behavior trajectories with the expected physical behavior model, construct a response difference atlas, and quantitatively identify response deviations of the control strategy and the execution state; The difference type discrimination and compensation module is configured to, for the nodes with the response deviations, judge difference types according to the deviation atlas, correct the strategy based on type matching, generate compensation instructions by reconstructing a local optimization target, and output a correction path; The strategy migration optimization module is configured to input the historical behavior deviation samples and the correction path into a graph structure time sequence learning model, update node control weights and regulation priorities, and form a closed-loop adaptive strategy migration mechanism.

Images