A distributed photovoltaic cluster cooperative control method based on edge computing

Abstract

The application discloses a kind of based on edge computing's distributed photovoltaic cluster collaborative control method, belong to distributed photovoltaic power generation control field, including the following steps: S1, constructs distributed photovoltaic cluster system model and multilayer edge computing architecture;S2, based on scene clustering photovoltaic cluster partition and uncertainty modeling;S3, based on distributed model predictive control edge layer collaborative optimization;S4, cloud layer global optimization and long-term scheduling;S5, fog layer aggregation optimization and global coordination.The above-mentioned based on edge computing's distributed photovoltaic cluster collaborative control method is used, through edge-fog-cloud three-layer collaborative architecture, the efficient control of distributed photovoltaic cluster under uncertainty scene is realized, the control precision, stability and economy of distributed photovoltaic cluster are significantly improved, and key support is provided for high proportion renewable energy grid connection.

Description

Technical Field

[0001] This invention relates to the field of distributed photovoltaic power generation control technology, and in particular to a collaborative control method for distributed photovoltaic clusters based on edge computing. Background Technology

[0002] As an important component of renewable energy, the large-scale application of distributed photovoltaic (PV) systems places higher demands on the flexibility and stability of power systems. Existing technologies mainly employ the following solutions:

[0003] 1. Centralized Cloud Platform Control Architecture: Traditional methods directly collect photovoltaic unit data and issue control commands through a remote cloud platform. For example, some systems upload all inverter operating data to the cloud for centralized optimization and then transmit control parameters back via the communication network. This solution relies on high-bandwidth communication links, and cloud computing resources need to process massive amounts of data simultaneously, resulting in insufficient real-time performance (end-to-end latency typically exceeds 200ms), failing to meet the dynamic response requirements of inverters (such as adjusting the modulation ratio within 50ms when irradiance changes abruptly). Furthermore, the centralized architecture is prone to system paralysis when communication is interrupted and cannot effectively address the geographical dispersion and electrical coupling of distributed photovoltaic systems.

[0004] 2. Single-layer edge computing collaborative control: To improve real-time performance, some studies have attempted to deploy edge computing nodes in photovoltaic clusters for local optimization. For example, edge servers are used to achieve regional data aggregation and simple power balance regulation. However, such solutions only achieve local control at the edge layer and lack a collaborative mechanism with upper layers (fog layer, cloud layer).

[0005] 3. Deterministic Model Predictive Control (MPC): Existing MPC methods mostly build photovoltaic output models based on historical data and achieve power point tracking through rolling optimization. For example, some systems use MPC algorithms to adjust the inverter duty cycle to track the maximum power point (MPP). However, such methods do not fully consider the uncertainties of meteorological disturbances (such as irradiance fluctuations and temperature changes), resulting in large model prediction errors (prediction deviations exceeding 15% in typical scenarios). In addition, traditional MPC does not introduce robust control mechanisms, which can easily lead to frequency limit violations when inverter parameters drift or communication delays (such as 5G network delays exceeding 10ms).

[0006] 4. Static Partitioning and Single-Objective Optimization: Existing cluster control typically employs fixed partitioning strategies (such as division by geographical location) without considering electrical coupling characteristics (such as line impedance differences), resulting in low correlation of power fluctuations within the same zone (correlation coefficient below 0.6). Furthermore, optimization objectives often focus on short-term power tracking (such as minute-level metrics), lacking long-term economic considerations (such as annual maintenance plans and curtailment costs). Summary of the Invention

[0007] The purpose of this invention is to provide a collaborative control method for distributed photovoltaic clusters based on edge computing, thereby solving the aforementioned technical problems.

[0008] To achieve the above objectives, this invention provides a distributed photovoltaic cluster collaborative control method based on edge computing, comprising the following steps:

[0009] S1. Construct a distributed photovoltaic cluster system model and a multi-layer edge computing architecture: Considering the output characteristics of photovoltaic units and the deployment requirements of edge computing nodes, establish a dynamic model of a distributed photovoltaic cluster with meteorological disturbances, and design a three-layer edge-fog-cloud architecture to achieve distributed data processing and control coordination.

[0010] S2. Photovoltaic cluster partitioning and uncertainty modeling based on scenario clustering: Taking into account the spatiotemporal fluctuation characteristics of photovoltaic power and meteorological disturbances, scenario analysis is used to generate multi-disturbance scenarios, and K-means clustering is used to divide the distributed photovoltaic cluster dynamic model constructed in step S1 into cluster partitions.

[0011] S3. Edge layer collaborative optimization based on distributed model predictive control: Within each cluster partition divided in step S2, distributed model predictive control is executed through edge computing nodes. Combined with robust control barrier function constraints on inverter parameters, the optimal control sequence is obtained and then corrected.

[0012] S4. Cloud Global Optimization and Long-Term Scheduling: Based on historical data, the annual power grid scheduling plan, and the multi-disturbance scenarios divided in step S2, the cloud layer with an edge-fog-cloud three-layer architecture is used to optimize the capacity planning and maintenance strategy of distributed photovoltaic clusters and generate annual power targets.

[0013] S5. Fog Layer Aggregation Optimization and Global Coordination: Based on the corrected optimal control sequence obtained in step S3, fog layer nodes in the edge-fog-cloud three-layer architecture are used to aggregate the edge layer data of the cluster partitions. Combined with the annual power target generated in step S4, power allocation coordination between cluster partitions is achieved through multi-objective optimization, and edge layer control instructions for each cluster partition are generated.

[0014] Therefore, the above-mentioned collaborative control method for distributed photovoltaic clusters based on edge computing, as described in this invention, has the following beneficial effects:

[0015] 1. Multi-layer edge computing architecture: Design a three-layer collaborative architecture of edge-fog-cloud, in which the edge layer realizes real-time data acquisition and local optimization (response time ≤50ms), the fog layer performs inter-regional collaboration and power balancing (control cycle 0.1s), and the cloud layer provides long-term scheduling and policy solidification (update cycle 1h), thereby reducing the communication bandwidth requirement to below 100Mbps, while supporting network outage autonomy (the edge layer can maintain local control for ≥30 minutes).

[0016] 2. Uncertainty Quantification and Dynamic Zoning: Monte Carlo simulation is used to generate multiple meteorological disturbance scenarios, and Gaussian kernel density estimation is used to calculate the scenario probability. Simultaneously, a K-means clustering algorithm based on electrical-geographic hybrid distance is proposed, which improves the correlation of power fluctuations in units within the same region to above 0.8, significantly enhancing the collaborative optimization effect.

[0017] 3. Distributed Model Predictive Control (DMPC) and Robust Constraints: A DMPC algorithm with RCBF constraints is designed to achieve power co-optimization within partitions at the edge layer, reducing the standard deviation of frequency deviation from 0.3Hz to below 0.2Hz. At the same time, the PSO algorithm reduces the computing power requirements of edge computing nodes (reducing the number of iterations from 100 to 50). In addition, a robust control mechanism is introduced to ensure frequency safety margin even with communication latency (≤10ms).

[0018] 4. Multi-objective dynamic optimization and closed-loop verification: The fog layer constructs a weighted function with multiple objectives such as network loss and control smoothness, and uses an improved NSGA-II algorithm to solve for the Pareto optimal solution, which reduces the network loss rate by more than 15%; the cloud layer formulates an annual maintenance plan through dynamic programming and adjusts the strategy parameters in combination with quarterly evaluation, ultimately achieving an annual net profit increase of more than 8%.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is a flowchart of a distributed photovoltaic cluster collaborative control method based on edge computing according to the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0022] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0023] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] like Figure 1 As shown, a collaborative control method for a distributed photovoltaic cluster based on edge computing includes the following steps:

[0025] S1. Construct a distributed photovoltaic cluster system model and a multi-layer edge computing architecture: Considering the output characteristics of photovoltaic units and the deployment requirements of edge computing nodes, establish a dynamic model of a distributed photovoltaic cluster with meteorological disturbances, and design a three-layer edge-fog-cloud architecture to achieve distributed data processing and control coordination.

[0026] Step S1 specifically includes the following steps:

[0027] S11. Considering the coupling effects of irradiance, temperature, and dust deposition on photovoltaic output, establish a photovoltaic unit output model:

[0028] ;

[0029] In the formula, Indicates the first Real-time power of each photovoltaic unit; Indicates the dust obstruction attenuation coefficient; This indicates the maximum output power under standard test conditions; Indicates real-time irradiance; Indicates the power temperature coefficient; Indicates the real-time ambient temperature; Indicates the optical depth of the aerosol;

[0030] S12. Based on distributed edge computing technology, construct an edge-fog-cloud three-layer architecture adapted to the geographically dispersed characteristics of photovoltaic clusters to achieve a balance between local real-time control and global optimization;

[0031] S13. Construct a dynamic model of a distributed photovoltaic (PV) cluster to integrate the interaction between each PV unit and the edge computing node, and quantify the power fluctuation and communication coupling characteristics of the distributed PV cluster. The expression of the dynamic model of the distributed PV cluster is as follows:

[0032] ;

[0033] In the formula, Represents the state vector The time derivative, and , This indicates the communication latency of edge computing nodes. This indicates the transpose operation. Indicates the first The actual output voltage of each photovoltaic unit; Represents the control vector, and , Indicates the first The modulation ratio of each inverter; Represents the perturbation vector; and Let these represent the state matrix and the control matrix, respectively. , , Indicates the first The photovoltaic unit and the first Line impedance between photovoltaic units Indicates the first The inductance of each line; The time constant representing the communication delay of edge computing nodes; Indicates the first The gain coefficient of each inverter.

[0034] The edge-fog-cloud three-layer architecture described in step S12 includes an edge layer, a fog layer, and a cloud layer. The edge layer is built on an edge gateway used to perceive the operating status of the distributed photovoltaic cluster. The edge gateway includes multiple photovoltaic units. Deployed and used for real-time power acquisition ,Voltage and current Edge computing nodes The edge layer uses the MQTT protocol.

[0035] The fog layer is built on industrial servers and multiple edge computing nodes. Corresponding to 1 fog node Fog Node Local area network communication between them; the fog layer uses the Modbus TCP protocol;

[0036] The cloud layer is built on cloud servers and uses the HTTPS protocol.

[0037] S2. Photovoltaic cluster partitioning and uncertainty modeling based on scenario clustering: Taking into account the spatiotemporal fluctuation characteristics of photovoltaic power and meteorological disturbances, scenario analysis is used to generate multi-disturbance scenarios, and K-means clustering is used to divide the distributed photovoltaic cluster dynamic model constructed in step S1 into cluster partitions.

[0038] Step S2 specifically includes the following steps:

[0039] S21. Uncertainty Scenario Generation: Based on historical meteorological data and photovoltaic power fluctuation characteristics, Monte Carlo simulation is used to generate multi-disturbance scenarios;

[0040] S211. Probability Distribution Modeling: Fitting with Beta Distribution ,in, Both represent shape parameters, and , , and These represent the mean and standard deviation of irradiance, respectively. Represents the Beta distribution;

[0041] Simultaneously, normal distribution fitting is used. ,in, and These represent the temperature mean and temperature standard deviation, respectively. Represents a normal distribution;

[0042] S212. Based on the probability distribution constructed in step S211, the Monte Carlo method is used to generate... Group sampled values , They represent the first The irradiance and ambient temperature collected in this study;

[0043] S213, will Convert to parameters for actual disturbance scenarios:

[0044] ;

[0045] ;

[0046] In the formula, and These represent the disturbance scenarios. Irradiance and ambient temperature; and These represent standard irradiance and standard temperature, respectively. , ;

[0047] S214. Scene Probability Calculation: Kernel density estimation is used to assign probability values ​​to the perturbation scene generated in step S213.

[0048] ;

[0049] In the formula, Indicates a disturbance scenario The probability value; and These represent the bandwidth parameters for irradiance and temperature scenarios, respectively. This indicates the total number of irradiance and temperature samples generated by the Monte Carlo simulation; This represents the total number of cluster partitions in the K-means clustering partitioning;

[0050] S215, to Perform normalization processing;

[0051] S22. K-means clustering partitioning based on electrical-geographic distance: Integrating the electrical and geographic distances between photovoltaic units, and dividing the clustering partitions using K-means clustering;

[0052] S221, Adjust line impedance Considering the electrical distance, and integrating the electrical and geographical distances between photovoltaic units, a weighted distance matrix is ​​constructed. :

[0053] ;

[0054] ;

[0055] In the formula, Indicates the first The photovoltaic unit and the first The mixing distance between individual photovoltaic units; Indicates the first The geographical distance of the photovoltaic unit and the first Geographical distance between individual photovoltaic units; Indicates weight;

[0056] S222, Clustering;

[0057] S2221, K-means clustering initialization: From Randomly selected from photovoltaic units Non-repeating edge computing nodes As initial centroids, and ensuring that the mixing distance between each initial centroid is greater than a set threshold. ,in The initial set of centroids is obtained. , Indicates the first The initial centroids of each cluster partition are determined, with each cluster partition corresponding to an inverter.

[0058] S2222, Node allocation and cluster update;

[0059] S22221, Calculate the value of each photovoltaic unit. With the initial centroid Mixed distance between and photovoltaic unit Edge computing nodes are assigned to the cluster with the smallest distance:

[0060] ;

[0061] In the formula, Indicates the first The initial centroid set of each cluster partition;

[0062] S22222 Calculate the average mixture distance vector of all nodes within each cluster partition, and select the edge node closest to this mixture distance vector as the new centroid. :

[0063] ;

[0064] In the formula, Let represent the average mixed distance vector, and ; Indicates the first The edge computing node and the first Mixed distance between edge computing nodes;

[0065] S22223, Iteration step S22222: Stop iterating and output the clustering partitioning results when the change in the mixing distance between all new centroids and old centroids is less than the set threshold or the maximum number of iterations is reached. , This represents the final set of clustering partitioning results.

[0066] S3. Edge layer collaborative optimization based on distributed model predictive control: Within each cluster partition divided in step S2, distributed model predictive control is executed through edge computing nodes. Combined with robust control barrier function constraints on inverter parameters, the optimal control sequence is obtained and then corrected.

[0067] Step S3 specifically includes the following steps:

[0068] S31. Design a distributed model predictive control objective function with the optimization objectives of minimizing power point tracking error, control quantity fluctuation, and node coordination deviation. :

[0069] ;

[0070] In the formula, Indicates the prediction time domain; , and These represent the power error weighting coefficient, the control quantity change weighting coefficient, and the coordination error weighting coefficient, respectively. Indicates the power tracking error, and , Indicates the first Each photovoltaic unit in Time prediction Constant output power Indicates the first Each cluster partition in Reference power at any given time; This represents the rate of change of the control quantity, and , Indicates the first One inverter in Time prediction Control the amount at all times. Indicates the first One inverter in The actual control quantity at any given moment; Indicates cooperative bias, and , and They represent the first The first inverter and the first One inverter in The actual control quantity at any given moment; Indicates the first The first inverter and the first Communication delay time between inverters;

[0071] S32. Construct robust control barrier function constraints:

[0072] ;

[0073] in,

[0074] ;

[0075] ;

[0076] In the formula, Indicates the inverter modulation ratio; and These represent the lower and upper limits of the inverter modulation ratio constraint, respectively. and These represent the maximum and minimum modulation ratios of the inverter, respectively. Indicates the inverter gain; Represents the system's inertial constant; and Both represent attenuation coefficients; Indicates the first The frequency deviation of each edge computing node, and , Indicates the first The actual frequency of each edge computing node Indicates the rated frequency; Indicates the maximum frequency deviation; Indicates the maximum power disturbance; Indicates the first Real-time power of each photovoltaic unit; Indicates the first The system damping coefficient corresponding to each photovoltaic unit; Indicates the first The backup power of each photovoltaic unit;

[0077] S33. Distributed Optimization Solution: The particle swarm optimization algorithm is used to solve the constrained optimization problem of distributed model predictive control, and the optimal control sequence is searched iteratively through swarm intelligence. ;

[0078] S34. Data Synchronization and Collaborative Correction among Edge Computing Nodes: Edge computing nodes exchange local optimization results through real-time communication and correct the optimal control sequence based on the power balance requirements within the clustered partitions. The corrected optimal control sequence is obtained. .

[0079] Step S34 specifically includes the following steps:

[0080] S341. Configure edge computing nodes to use the MQTT protocol for data transmission. ,in This indicates the local power deviation, and ; Indicates the first The data set transmitted by each edge computing node; Indicates the first Communication trigger threshold for each edge computing node; Indicates time;

[0081] Simultaneously, each edge computing node is configured to publish a local control sequence. and power And collect information from other edge computing nodes in the same partition;

[0082] S342, Power Correction: When the total power deviation of the cluster partition... hour, This indicates that a power deviation threshold is set, and the optimal control sequence is corrected based on the proportional allocation principle:

[0083] ;

[0084] In the formula, This represents the corrected optimal control sequence.

[0085] S4. Cloud Global Optimization and Long-Term Scheduling: Based on historical data, the annual power grid scheduling plan, and the multi-disturbance scenarios divided in step S2, the cloud layer with an edge-fog-cloud three-layer architecture is used to optimize the capacity planning and maintenance strategy of distributed photovoltaic clusters and generate annual power targets.

[0086] Step S4 specifically includes the following steps:

[0087] S41. Seasonal and periodic characteristics of photovoltaic power are extracted through time series analysis, and a trend model is constructed using multiple linear regression to fit the trend:

[0088] ;

[0089] In the formula, This represents the long-term trend forecast of photovoltaic power based on multiple linear regression fitting. , , , , All represent the fitting coefficients; , , and These represent the month vector, hour vector, monthly average irradiance, and monthly average temperature, respectively.

[0090] S42. To maximize annual net revenue, integrate electricity revenue, maintenance costs, and curtailment losses to construct an annual optimization objective function. :

[0091] ;

[0092] In the formula, This indicates the unit price of revenue per kilowatt-hour; This indicates the actual annual power generation of the photovoltaic cluster; Indicates unit maintenance cost; Indicates cluster partitioning Maintenance cycle; Indicates the annual plan for the power grid; This indicates the cost of penalties for abandoning light;

[0093] S43. Add the following annual optimization constraints:

[0094] Equipment lifespan constraints: ;

[0095] Maintain window constraints: ;

[0096] In the formula, Indicates the first The equipment maintenance cycle for each cluster partition; and They represent the first The maximum and minimum values ​​of equipment maintenance cycles for each cluster partition; Indicates the maintenance window time; Indicates the annual maximum trend power;

[0097] S44. Solve the annual optimization objective function with added annual optimization constraints using dynamic programming. The annual optimization is decomposed into monthly subproblems, and then solved iteratively using the Bellman equation to obtain the optimal annual plan. ;in, This represents the annual power target curve obtained through cloud optimization;

[0098] S45. Optimal annual plan The targets are broken down into quarterly goals and distributed to the fog layer. The targets for the next quarter are then revised based on the actual data from the current quarter, resulting in the revised quarterly targets. :

[0099] ;

[0100] In the formula, and These represent the power target values ​​for the current quarter and the next quarter, respectively. This represents quarterly feedback data on the fog layer.

[0101] S5. Fog Layer Aggregation Optimization and Global Coordination: Based on the corrected optimal control sequence obtained in step S3, fog layer nodes in the edge-fog-cloud three-layer architecture are used to aggregate the edge layer data of the cluster partitions. Combined with the annual power target generated in step S4, power allocation coordination between cluster partitions is achieved through multi-objective optimization, and edge layer control instructions for each cluster partition are generated.

[0102] Step S5 specifically includes the following steps:

[0103] S51. Data aggregation and global state assessment: Fog layer nodes receive clustered partition data uploaded by each edge computing node, integrate it into a global state matrix, and assess the deviation between the current cluster and the power grid command.

[0104] S511. Data verification and cleaning: Perform anomaly detection on the received clustered partition data, remove outliers, and fill in missing data using linear interpolation.

[0105] S512. Calculate the total power of the distributed photovoltaic cluster based on the cleaned clustered partition data. Power deviation and network loss rate :

[0106] ;

[0107] ;

[0108] ;

[0109] In the formula, Indicates cluster partitioning Cleaning power; Indicates the reference power of the power grid; This represents the transmission loss between cluster partitions. , Indicates cluster partitioning The line resistance; Indicates cluster partitioning To cluster partition Transmission power, ; This represents the total number of cluster partitions in the K-means clustering partitioning;

[0110] Obtain the global state vector , Indicates cluster partitioning To cluster partition A set of transmission power sequences;

[0111] S52. Constructing a multi-objective optimization function for the fog layer: With the objectives of tracking the grid power command, minimizing transmission losses, and adjusting the optimal control sequence, construct a fog layer optimization function. :

[0112] ;

[0113] In the formula, , and These represent the objective function for tracking power command, minimizing transmission loss, and minimizing the modified optimal control sequence change, respectively. , and All represent weighting coefficients; This represents the total power deviation within the prediction time domain; This represents the predicted transmission loss value, and , Indicates the first The cluster partition to the Each cluster partition in Time prediction Transmission power at any given moment; Indicates control over the time domain; This represents the change in the corrected optimal control sequence, and , and They represent the first Each cluster partition in Time prediction Time and The optimal control sequence at time 1;

[0114] S53. Add the following fog layer optimization constraints:

[0115] Transmission capacity constraints: ;

[0116] Corrected optimal control sequence variation constraints: ;

[0117] Frequency stability constraints: ;

[0118] Power balance constraints: ;

[0119] Grid acceptance constraints: ;

[0120] In the formula, Indicates cluster partitioning Clustering partitioning The maximum transmission power between them, and , Indicates the rated capacity of the line. Indicates the power factor; This represents the maximum change in the corrected optimal control sequence; Indicates the control cycle; Represents the system's inertial constant; Indicates the maximum permissible deviation; Indicates cluster partitioning Frequency deviation; Indicates the damping coefficient; Represents the local photovoltaic power of the cluster partition, and ; Indicates cluster partitioning Local load; This represents the actual output power of the distributed photovoltaic cluster;

[0121] S54. An improved non-dominated sorting genetic algorithm is used to solve the multi-objective constrained optimization problem of fog layer. Then, through fast non-dominated sorting and crowding calculation, the global coordination instructions in the Pareto optimal solution set are found under the premise of satisfying the constraints.

[0122] S541. Population Initialization: Set each individual to represent a set of cluster partition power adjustment amounts. And randomly generated within the feasible region. ,in, This represents the power adjustment for all cluster partitions. They represent the first Power adjustment for each cluster partition Follow the revised quarterly targets ; express Uniformly distributed random numbers within the interval; and These represent the maximum and minimum values ​​of the cluster partition power adjustment, respectively. ;

[0123] S542. Verify whether each individual meets the fog layer optimization constraints. If it does, proceed to step S543; otherwise, perform the following repair adjustments:

[0124] when At that time, the transmission power is reduced proportionally:

[0125] ;

[0126] In the formula, Indicates the repaired cluster partitions Clustering partitioning Transmission power between;

[0127] when Adjustment Until the frequency stability constraint is met;

[0128] S543. Classify individuals within the population according to the corresponding objective function based on non-dominance relations, and determine the dominance level of each individual;

[0129] S544. Calculate individual crowding level:

[0130] ;

[0131] In the formula, Represents an individual The degree of congestion; and Each represents an individual within its respective level. The maximum and minimum values; and They represent the first In terms of the objective function dimension, individual The objective function values ​​of the next and previous neighboring individuals within the non-dominated hierarchy;

[0132] S545. Parent generations are selected based on roulette wheel selection combined with crowding sorting, and offspring are generated through simulated binary crossover and polynomial mutation, where the crossover operation expression is as follows:

[0133] ;

[0134] In the formula, and These represent the results after the crossover operation, i.e., the first... The power adjustment of two offspring individuals generated from each cluster partition; The morphological control parameters for the cross-distribution are represented, and , Indicates the interval The numbers inside follow a uniform distribution. Indicates the cross-distribution index. ; and These represent the first and second parent individuals involved in the crossover, respectively. Power adjustment for each cluster partition;

[0135] The expression for the mutation operation is as follows:

[0136] ;

[0137] In the formula, Indicates the result after the mutation operation, the first Power adjustment of mutated offspring in each cluster partition; and They represent the first The maximum and minimum allowable power adjustment for each cluster partition; The morphological control parameters representing the variation distribution, and , Indicates the distribution index of variation. ; A unique identifier number representing an edge computing node;

[0138] S546. Merge the parent and offspring populations, reorder the non-dominated populations, calculate crowding, and then add individuals to the population according to their rank until the population size reaches the set size, resulting in a new population composed of the best individuals. This allows for the acquisition of the optimal power adjustment amount;

[0139] S55, Fog Node will adjust the optimal power amount. Transformed into edge layer control commands:

[0140] ;

[0141] In the formula, Indicates fog layer towards the first Real-time control commands issued by each cluster partition; This indicates that after local optimization at the edge layer, the first... Initial control sequences for each cluster partition; Indicates the conversion factor;

[0142] S56. Determine if the power deviation after executing S45 at the edge layer is satisfied. If yes, return to step S54 to recalculate; otherwise, output... ,in Indicates the real-time power of the edge layer. It is a unit of power.

[0143] Step S543 specifically includes the following steps:

[0144] S5431. For each individual, obtain the set of individuals it governs. and dominance count ;

[0145] S5432, Remove After classifying the individuals, update the remaining individuals and return to step S5431 until all individuals have been classified.

[0146] Step S5 is followed by step S6, which verifies the performance of the control strategy through edge-fog-cloud data interaction to form a closed-loop optimization mechanism.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for edge-computing-based distributed photovoltaic cluster cooperative control, characterized in that: Comprising the following steps: S1, constructing a distributed photovoltaic cluster system model and a multi-layer edge computing architecture: considering the output characteristics of photovoltaic units and the deployment requirements of edge computing nodes, a dynamic model of distributed photovoltaic cluster with meteorological disturbance is established, and a three-layer architecture of edge-fog-cloud is designed to realize distributed data processing and control coordination; S2, photovoltaic cluster partitioning and uncertainty modeling based on scene clustering: in view of the temporal and spatial fluctuation characteristics of photovoltaic power combined with meteorological disturbance, a scene analysis method is used to generate multiple disturbance scenes, and the dynamic model of distributed photovoltaic cluster constructed in step S1 is divided into clustering partitions through K-means clustering; S3, edge layer collaborative optimization based on distributed model predictive control: in each clustering partition divided in step S2, distributed model predictive control is performed through edge computing nodes, the inverter parameters are constrained by robust control barrier function, the optimal control sequence is solved, and the optimal control sequence is modified; S4, cloud layer global optimization and long-term scheduling: based on historical data, annual scheduling plan of power grid and multiple disturbance scenes divided in step S2, the cloud layer of edge-fog-cloud three-layer architecture is used to optimize the capacity planning and maintenance strategy of distributed photovoltaic cluster, and annual power target is generated; S5, fog layer aggregation optimization and global coordination: based on the modified optimal control sequence obtained in step S3, the fog layer nodes of edge-fog-cloud three-layer architecture are used to converge the edge layer data of clustering partitions, and combined with the annual power target generated in step S4, the power distribution coordination among clustering partitions is realized through multi-objective optimization, and the edge layer control instructions of each clustering partition are generated. 2.The method of claim 1, wherein: Step S1 specifically includes the following steps: S11, considering the coupling effects of irradiance, temperature and sand deposition on photovoltaic output, a photovoltaic unit output model is established: ； In the formula, represents the real-time power of the nth photovoltaic unit; represents the sand and dust shielding attenuation coefficient; represents the maximum output power under standard test conditions; represents the real-time irradiance; represents the power temperature coefficient; represents the real-time ambient temperature; represents the aerosol optical depth;​ S12, based on distributed edge computing technology, an edge-fog-cloud three-layer architecture suitable for the geographical dispersion characteristics of photovoltaic cluster is constructed to realize the balance between local real-time control and global optimization; S13, a dynamic model of distributed photovoltaic cluster is constructed to integrate the interaction between photovoltaic units and edge computing nodes, and to quantify the power fluctuation and communication coupling characteristics of distributed photovoltaic cluster, wherein the expression of the dynamic model of distributed photovoltaic cluster is as follows: ； wherein denotes the time derivative of the state vector and , denotes the edge computing node communication delay time, denotes the transpose operation, denotes the actual output voltage of the th photovoltaic unit; denotes the control vector, and , denotes the modulation ratio of the th inverter; denotes the disturbance vector; and denote the state matrix and the control matrix, respectively, and , , denotes the actual output voltage of the th photovoltaic unit; denotes the line impedance between the th photovoltaic unit and the th photovoltaic unit; denotes the time constant of the edge computing node communication delay; denotes the gain coefficient of the th inverter. 3.The method of claim 2, wherein: The edge-fog-cloud three-layer architecture described in step S12 includes an edge layer, a fog layer and a cloud layer, wherein the edge layer is constructed based on an edge gateway for sensing the running state of a distributed photovoltaic cluster, the edge gateway includes a plurality of photovoltaic units Arranged and used for collecting real-time power , voltage and current of the edge computing node ; the edge layer adopts MQTT protocol The fog layer is based on an industrial server, and multiple edge computing nodes Corresponding to 1 fog layer node , fog layer node Local area network communication between them; the fog layer adopts Modbus TCP protocol; The cloud layer is constructed based on cloud server, and the cloud layer uses HTTPS protocol.

4. The method according to claim 2, wherein: Step S2 specifically includes the following steps: S21, uncertainty scenario generation: based on historical meteorological data and photovoltaic power fluctuation characteristics, multiple disturbance scenarios are generated by Monte Carlo simulation; S211, probability distribution modeling: Beta distribution fitting is adopted wherein, both represent shape parameters, and , , and respectively represent irradiance mean and irradiance standard deviation, Beta distribution; simultaneously fitting a normal distribution wherein, and represent the temperature mean and temperature standard deviation, respectively, denotes a normal distribution; S212, based on the probability distribution constructed in step S211, generating by using a Monte Carlo method Group sampling value , respectively represent the irradiance and ambient temperature of the first acquisition; S213、to convert to actual perturbation scenario parameters: ； ； wherein and denote the irradiance and the ambient temperature of the perturbed scenario , respectively; and denote the standard irradiance and the standard temperature, respectively, and , ; S214, scene probability calculation: the disturbance scenarios generated in step S213 are assigned with probability by kernel density estimation: ； wherein denotes the probability value of the perturbation scenario and denote the bandwidth parameters of the irradiance and temperature scenarios, respectively; denotes the total number of irradiance and temperature sample sets generated by the Monte Carlo simulation; denotes the total number of cluster partitions of the K-means clustering partitioning;​ S215、to normalization processing; S22, K-means clustering partitioning based on electrical-geographical distance: the electrical distance and geographical distance between photovoltaic units are fused, and the clustering partitions are divided by K-means clustering; S221、the line impedance The weighted distance matrix is constructed by considering the electrical distance and merging the electrical distance and geographical distance between the photovoltaic units : ； ； wherein denotes the mixing distance between the th photovoltaic unit and the th photovoltaic unit; denotes the geographical distance between the th photovoltaic unit and the th photovoltaic unit; denotes the weight; S222, clustering; S2221、K-means clustering initialization: randomly select one photovoltaic unit from one non-repeating edge computing node as initial centroids, and make the hybrid distance between each initial centroid greater than a set threshold wherein , get the initial centroid set , represent the initial centroid of the first cluster partition, and each cluster partition corresponds to an inverter; S2222, node allocation and clustering update; S22221, compute each photovoltaic unit with the initial centroid between the mixed distances and assign the edge computing node of the photovoltaic unit to the cluster with the minimum distance: ； In the formula, denotes the initial centroid set of the th cluster partition; S22222, compute the average hybrid distance vector of all nodes within each cluster partition, and select the edge computing node closest to the hybrid distance vector as the new centroid : ； In the formula, denotes the average mixing distance vector, and ; denotes the mixing distance between the first edge computing node and the first edge computing node; S22223, iterate step S22222 until the variation of the mixed distance of all new centroids and old centroids is less than a set threshold value or a maximum iteration number is reached, stop iteration and output the clustering partition result , represents the final output clustering partition result set.

5. The edge computing-based distributed photovoltaic cluster cooperative control method according to claim 4, characterized in that: Step S3 specifically includes the following steps: S31, design a distributed model predictive control objective function with the optimization goal of minimizing power tracking error, control quantity fluctuation and node coordination deviation : ； In the formula, Indicates the prediction time domain; , and These represent the power error weighting coefficient, the control quantity change weighting coefficient, and the coordination error weighting coefficient, respectively. Indicates the power tracking error, and , Indicates the first Each photovoltaic unit in Time prediction Constant output power Indicates the first Each cluster partition in Reference power at any given time; This represents the rate of change of the control quantity, and , Indicates the first One inverter in Time prediction Control the amount at all times. Indicates the first One inverter in The actual control quantity at any given moment; Indicates cooperative bias, and , and They represent the first The first inverter and the first One inverter in The actual control quantity at any given moment; Indicates the first The first inverter and the first Communication delay time between inverters; S32, constructing a robust control barrier function constraint condition: ； wherein, ； ； wherein denotes the inverter modulation ratio; and denote the lower and upper bound of the inverter modulation ratio constraint, respectively; and denote the maximum and minimum inverter modulation ratio, respectively; denotes the inverter gain; denotes the system inertia constant; and both denote the damping coefficient; denotes the frequency deviation of the th edge computing node, and , denotes the actual frequency of the th edge computing node, denotes the rated frequency; denotes the maximum frequency deviation; denotes the maximum power disturbance; denotes the real-time power of the th photovoltaic unit; denotes the system damping coefficient corresponding to the th photovoltaic unit; denotes the reserve power of the th photovoltaic unit; S33, distributed optimization solution: adopt particle swarm optimization algorithm to solve the constraint optimization problem of distributed model predictive control, and search the optimal control sequence through group intelligence iteration ; S34, data synchronization and collaborative correction between edge computing nodes: edge computing nodes exchange local optimization results through real-time communication, and correct the optimal control sequence based on the power balance demand within the clustering partition , to obtain the corrected optimal control sequence .

6. The edge computing-based distributed photovoltaic cluster cooperative control method according to claim 5, characterized in that: Step S34 specifically includes the following steps: S341、Set up the transmission of data between edge computing nodes using MQTT protocol wherein represents a local power deviation, and ; represents a transmission data set of the th edge computing node; represents a communication trigger threshold of the th edge computing node; represents time; Simultaneously setting each edge computing node to publish a local control sequence and power and collect information of other edge computing nodes in the same partition; S342, power correction: when the total power deviation of the clustering partition is greater than the power deviation threshold represents setting a power deviation threshold, correcting the optimal control sequence based on the proportional distribution principle:​ ； In the formula, denotes the modified optimal control sequence.

7. The edge computing-based distributed photovoltaic cluster cooperative control method according to claim 6, characterized in that: Step S4 Specifically includes the following steps: S41, extract the seasonal and periodic characteristics of photovoltaic power through time series analysis, and construct a trend model using multiple linear regression to perform trend fitting; ； wherein represents a long-term trend photovoltaic power prediction value based on a multiple linear regression fit; , , , , each represent a fit coefficient; , , and represent a month vector, an hour vector, a monthly average irradiance and a monthly average temperature, respectively. S42, integrate the electricity income, maintenance cost and light loss to build an annual optimization objective function with the goal of maximizing the annual net income : ； wherein, represents the degree electric yield unit price; represents the actual annual power generation of the photovoltaic cluster; represents the unit maintenance cost; represents the maintenance period of the clustering partition ; represents the grid annual plan; represents the light abandonment penalty cost; S43, add the following annual optimization constraint conditions: Device lifetime constraints: ; Maintain window constraints: ; In the formula, represents the equipment maintenance period of the first cluster partition; and respectively represent the maximum value and the minimum value of the equipment maintenance period of the first cluster partition; represents the maintenance window time; represents the annual maximum trend power; S44, solving the annual optimization objective function with the annual optimization constraint by dynamic programming method Decomposing the annual optimization into monthly sub-problems, and solving the optimal annual plan by Bellman equation iteration ; wherein, represents the annual power target curve obtained by cloud layer optimization S45, the optimal annual plan is determined decomposed into quarterly targets issued to the fog layer, and the next quarter's target is corrected based on the current quarter's actual data to obtain the corrected quarterly target : ； wherein and represent the power target value for the next and current quarter, respectively; represents the fog layer quarter feedback data.

8. The edge computing-based distributed photovoltaic cluster cooperative control method according to claim 7, characterized in that: Step S5 specifically includes the following steps: S51, data aggregation and global state evaluation: the fog layer node receives the clustering partition data uploaded by each edge computing node, integrates it into a global state matrix, and evaluates the deviation of the current cluster from the grid instruction; S511, data verification and cleaning: perform anomaly detection on the received clustering partition data, eliminate outliers, and use linear interpolation to fill in missing data; S512, based on the cleaned clustering partition data, calculate the total power of the distributed photovoltaic cluster , power deviation , and network loss rate : ； ； ； wherein represents the cleaning power of the cluster partition ; represents the grid reference power; represents the transmission loss between the cluster partitions, , represents the line resistance of the cluster partition ; represents the transmission power of the cluster partition to the cluster partition , ; represents the total number of cluster partitions of the K-means cluster division; obtaining a global state vector , representing a cluster partition to a cluster partition a set of transmission power sequences; S52, construct a multi-objective optimization function of the fog layer: track the power instruction of the power grid, minimize the transmission loss, and correct the optimal control sequence change to construct the optimization function of the fog layer : ； In the formula, , and These represent the objective function for tracking power command, minimizing transmission loss, and minimizing the modified optimal control sequence change, respectively. , and All represent weighting coefficients; This represents the total power deviation within the prediction time domain; This represents the predicted transmission loss value, and , Indicates the first The cluster partition to the Each cluster partition in Time prediction Transmission power at any given moment; Indicates control over the time domain; This represents the change in the corrected optimal control sequence, and , and They represent the first Each cluster partition in Time prediction Time and The optimal control sequence at time 1; S53, add the following fog layer optimization constraint conditions: Transmission capacity constraints: ; The modified optimal control sequence change constraint: ; Frequency stability constraint: ; Power balance constraint: ; Grid acceptance constraints: ; In the formula, represents the maximum transmission power between the cluster partition and the cluster partition , and , represents the line rated capacity, represents the power factor; represents the maximum change amount of the modified optimal control sequence; represents the control period; represents the system inertia constant; represents the maximum frequency deviation; represents the frequency deviation of the cluster partition ; represents the damping coefficient; represents the local photovoltaic power of the cluster partition, and ; represents the local load of the cluster partition ; represents the actual output power of the distributed photovoltaic cluster; S54, solve the fog layer multi-objective constraint optimization problem using an improved non-dominated sorting genetic algorithm, and then find the global coordinated instruction in the Pareto optimal solution set under the premise of meeting the constraint conditions through fast non-dominated sorting and congestion calculation; S541、population initialization: set each individual to represent a set of cluster partition power adjustment amount , and randomly generate in the feasible region , wherein, represents the power adjustment amount of all cluster partitions, respectively represents the power adjustment amount of the follows the corrected quarterly target ; represents a uniformly distributed random number in the interval; and respectively represent the maximum and minimum values of the cluster partition power adjustment amount, and ; S542, verify whether each individual meets the fog layer optimization constraint condition one by one, if it meets, execute step S543, otherwise, perform the following repair adjustment: When the transmission power is scaled down: ; In the formula, represents the cluster partition after repair and the transmission power between the cluster partitions ; When the adjustment is made until the frequency stability constraint is met; S543, classify individuals in the population according to the corresponding objective function through non-dominated relationship to determine the dominance level of each individual; S544, calculate the individual congestion degree: ； wherein denotes the crowding distance of the individual ; and denote the maximum and minimum values of the individuals in the respective rank; and denote the objective function values of the next and previous adjacent individual of the individual in the respective non-dominated rank in the th objective function dimension; S545, select the parent based on the roulette wheel selection combined with the congestion degree sorting, and generate the offspring through simulated binary crossover and polynomial mutation, where the crossover operation expression is as follows: ； In the formula, and respectively represent the power adjustment amount of the two offspring individuals generated by the first and second cluster partitions after the crossover operation; represents the shape control parameter of the crossover distribution, and , represents a random number obeying a uniform distribution in the interval , represents the crossover distribution index, ; and respectively represent the power adjustment amount of the first and second cluster partitions in the two parent individuals participating in the crossover. The mutation operation expression is as follows: ； In the formula, represents the variation of the power adjustment amount of the variation offspring of the first cluster partition after the variation operation; and respectively represent the maximum allowed power adjustment amount and the minimum allowed power adjustment amount of the first cluster partition; represents the shape control parameter of the variation distribution, and , represents the variation distribution index, ; represents the unique identification number of the edge computing node; S546, merge the parent and offspring populations, and re-non-dominated sort, calculate the crowding distance, and then rank the individuals according to the ranks to include the individuals in the population until the population size reaches the set size, to obtain a new population composed of optimal individuals , thereby obtaining an optimal power adjustment amount S55, the fog layer node adjusts the optimal power adjustment amount converts into edge layer control instructions: ； In the formula, indicates the real-time control instruction issued by the fog layer to the first cluster partition; indicates the initial control sequence of the first cluster partition after local optimization by the edge layer; indicates the conversion coefficient; S56, judging whether the power deviation after the execution of S45 in the edge layer is satisfied , if yes, returning to step S54 to recalculate, otherwise output , wherein represents the real-time power of the edge layer, is a power unit.

9. The edge computing-based distributed photovoltaic cluster cooperative control method according to claim 8, characterized in that: Step S543 specifically includes the following steps: S5431. For each individual, obtain the set of individuals over which it is dominated and domination count ; S5432, move out The remaining individuals are updated, and step S5431 is returned until the classification of all individuals is completed.

10. The method according to claim 9, wherein the method is characterized in that: Step S5 also includes step S6, which verifies the performance of the control strategy through edge-fog-cloud data interaction to form a closed-loop optimization mechanism.

Images