Distributed photovoltaic regulation and control method and device based on feature clustering and network access

Abstract

The invention discloses a distributed photovoltaic regulation and control method and device based on feature clustering and network access. The regulation and control method comprises the following steps: grouping communication networking characteristics of a plurality of inverters of a distributed photovoltaic field station, and adaptively controlling active power and reactive power of the inverters according to a set power voltage regulation and control distribution mode, so as to realize adaptive closed-loop regulation and control; on the basis of distributed photovoltaic regulation and control remote communication requirements, a communication implementation method integrating wired and wireless network access is provided, and a communication processing module is provided to realize automatic network switching according to main station channel selection, network communication conditions and user access point adjustment, so that the reliability of main station communication is ensured. According to the invention, the problems of multiple collection control objects of a distributed photovoltaic field station, communication access in a wired and wireless complex communication environment, remote regulation and control reliability and the like are well solved, and the improvement of distributed photovoltaic grid-connected regulation and control operation stability is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of distributed photovoltaic station of supporting arbitrary inverter acquisition control, wired / wireless network access, according to the change of communication quality, adaptive realization of distributed photovoltaic regulation and control technology according to the group of inverter characteristics, access network, belongs to the new energy regulation and control technical field. BACKGROUND

[0002] In recent years, distributed photovoltaic power generation has developed rapidly, by installing photovoltaic power generation equipment at user side, direct power supply to distribution network load is realized, distributed photovoltaic power generation is consumed locally to reduce the dependence on traditional power grid, and new energy utilization rate is improved. With large-scale grid-connected distributed photovoltaic, the distribution network power flow in different periods has changed, causing problems such as reverse power flow, voltage fluctuation, power quality and protection configuration. In order to ensure the safe and stable operation of distribution network, it is necessary to strengthen the monitoring of distributed photovoltaic power generation and realize the regulation and control of distributed photovoltaic.

[0003] Distributed photovoltaic power stations are located in industrial parks or user factory areas, and are numerous and scattered. In the communication scheme for collecting and monitoring them in the early stage, there are multiple communication methods such as wireless, EPON network, dispatching data network, etc. to access the master station. The early collection terminal interface cannot adapt to the above multiple communication interface methods, and the field terminal configuration is complex and inflexible. In the early stage, independent active power regulation AGC (Automatic Generation Control) devices and reactive power voltage regulation AVC (Automatic Voltage Control) devices are configured in some distributed photovoltaic stations with conditions, which realize the power and voltage control of distributed photovoltaic stations. However, the communication between independent AGC / AVC devices and the master station needs to be forwarded through the collection terminal or remote communication management machine. The stability and overall reliability of the master station communication are affected by multiple factors. The implementation cost is high, the regulation is not flexible, the stability of the master station communication is poor, the maintenance workload is large, and the independent regulation scheme cannot meet the needs of flexible access of distributed photovoltaic stations to the master station and integrated regulation.

[0004] Distributed photovoltaic stations develop rapidly, and the existing distributed photovoltaic stations have many and widely distributed inverters. Single-inverter regulation mode is complex and inflexible, and cannot meet the current station regulation. The traditional distributed photovoltaic station terminal cannot adapt to the change of network access, and the online rate of closed-loop regulation with the master station cannot be improved. The reliability of distributed photovoltaic regulation cannot be guaranteed, and it is difficult to meet the goal of reliable and stable operation of distribution network and efficient consumption of distributed photovoltaic. SUMMARY

[0005] The purpose of the present application is to provide a kind of distributed photovoltaic regulation based on feature clustering, adaptive fusion network access, network anomaly and communication quality change, improve the flexibility of distributed photovoltaic data collection and regulation, the communication reliability between distributed photovoltaic station and master station, realize photovoltaic flexible regulation and guarantee the stability of remote communication method and device.

[0006] The first aspect of the present application provides a kind of distributed photovoltaic regulation based on feature clustering and network access. By the communication networking feature clustering of multiple inverters of distributed photovoltaic station, according to the set power voltage regulation distribution mode, the control of active power and reactive power of inverter is adaptively carried out, to realize closed-loop control. Support distributed photovoltaic AGC, AVC function realization integrated in photovoltaic terminal.

[0007] The number of inverters of distributed photovoltaic station is not configured, and even up to hundreds, the data acquisition of inverter can come from inverter, data collector, box transformer monitoring and control, remote device and other equipment, and the acquisition of inverter active power, reactive power, voltage, current and operating state data is affected by the state of each device, communication. All photovoltaic inverters are in group unit, and the feature clustering can reduce the number of acquisition control objects, reduce the regulation difficulty, simplify inverter data operation and distribution control, and adapt to different field requirements.

[0008] A kind of distributed photovoltaic regulation based on feature clustering and network access, it includes: Step S101, based on the dynamic group of all inverters of station based on grid feature and inverter feature, generate sparse label matrix; Step S103, the real-time active power and reactive power of each inverter are collected, power feature matrix is constructed, and grid node voltage sensitivity matrix is synchronously acquired; Step S105, based on the sparse label matrix and voltage sensitivity matrix, obtain group power limit value; Step S107, obtain dynamic weight coefficient based on voltage sensitivity matrix; Step S109, through the dual-mode channel of supporting wireless and wired, based on communication quality evaluation value dynamic channel switching; Step S1011, receive power instruction issued by master station through switched channel, and combine dynamic weight coefficient and group power limit value to distribute group power.

[0009] Among them, the step S101 specifically includes: Based on grid topology impedance matrix and inverter electrical distance, all inverters of station are divided into multiple dynamic groups by improving spectral clustering algorithm, and sparse label matrix is generated based on the division result.

[0010] Further, the step S101 specifically includes: Based on the grid topology impedance matrix and the electrical distance of the inverter, the m inverters in the whole station are divided into n dynamic groups by improving the spectral clustering algorithm, and a n*m sparse label matrix Z is generated, wherein Z ik =1 indicates that the inverter k belongs to the group i, and satisfies the real-time electrical distance constraint D ek ≤D thr , D ek is the electrical distance of the inverter k, and D thr is the distance threshold.

[0011] The step S103 specifically includes: collecting real-time active power and reactive power of the inverter, constructing corresponding power feature matrix, and synchronously acquiring grid node voltage sensitivity matrix.

[0012] Further, the maximum and minimum values (P max / P min ) of real-time active power and the maximum and minimum values (Q max / Q min ) of reactive power are collected; based on the maximum and minimum values of active power and the maximum and minimum values of reactive power, m*m power feature matrix is constructed respectively, and grid node voltage sensitivity matrix is synchronously acquired. Wherein The power feature matrix includes active power matrix P and reactive power matrix Q.

[0013] The step S105 specifically includes: dynamically calculating the group power limit value by the modified function Γ(t) containing the Hermite polynomial expansion on the sparse label matrix, the power feature matrix and the voltage sensitivity matrix. v The input includes the integral value of the group historical power and the gradient norm of the voltage sensitivity matrix S

[0014] The step S107 specifically includes: generating a dynamic weight coefficient by combining the integral value of the group historical power determined by the sparse label matrix and the power feature matrix and the exponential decay function of the voltage sensitivity gradient.

[0015] The step S1011 specifically includes: receiving the power instruction issued by the master station, distributing the group power according to the dynamic weight coefficient, adjusting the group power distribution value according to the priority proportion of the voltage sensitivity when the distributed group power exceeds the group power limit value, and injecting compensation power to suppress the delay oscillation.

[0016] Further, the step S1011 specifically includes: receiving the power instruction P t issued by the master station, and distributing the group power P it according to the dynamic weight coefficient R i . When P it When the group power limit is exceeded (i.e., the group power P it When the group power limit is exceeded (i.e., the group power P

[0017] The improved spectral clustering algorithm divides all inverters in the station into multiple dynamic groups, specifically including: A weight matrix is constructed based on the current harmonic feature vectors of the inverters; A corresponding degree matrix is determined based on the weight matrix, and a Laplacian matrix is generated; The Laplacian matrix is subjected to eigenvalue decomposition to determine the embedding vectors of the inverters; Based on the embedding vectors of the inverters and the electrical distance constraint within the group, the cluster centers are dynamically adjusted by Mahalanobis distance.

[0018] Further, the improved spectral clustering algorithm divides all inverters in the station into multiple dynamic groups, i.e., the improved spectral clustering algorithm divides m inverters in the station into n dynamic groups, specifically including: A weight matrix W is constructed, and the elements of the weight matrix W are where I k is the current harmonic feature vector of the inverter k, k represents the current inverter to be calculated for similarity, and j represents other inverters compared with k; A Laplacian matrix L = D-W is generated, where D is a degree matrix, and L is subjected to eigenvalue decomposition; Based on Mahalanobis distance, the cluster centers are dynamically adjusted to ensure that the inverters within the group satisfy the electrical distance constraint of Z ik =1.

[0019] where the compensation power ΔP satisfies the following relationship:

[0020] where ΔP is the compensation power, P it is the group power, t delay is calculated by the delay parameter in the communication quality data set, η is the compensation gain, and τ is the delay attenuation constant.

[0021] The step S109 specifically includes: dynamically switching the channel / link based on the communication quality evaluation value Q com (t) through a dual-mode channel supporting wireless and wired communication.

[0022] In the step S109, the determination of the communication quality evaluation value comprises: monitoring a feature parameter related to the channel communication quality in real time, and establishing a communication quality comprehensive data set; determining an average signal strength and a packet loss rate coefficient based on the communication quality comprehensive data set; and obtaining the channel communication quality based on the average signal strength and the packet loss rate coefficient and in combination with a preset communication condition; wherein the feature parameter comprises a signal strength, a signal delay, and a packet loss rate.

[0023] The method further comprises: issuing the modified power instruction to the group inverter to realize AGC / AVC closed-loop regulation.

[0024] The second aspect of the application provides a distributed photovoltaic regulation device based on feature grouping and network access, which comprises: a processing unit for executing the above method, a parallel computing unit is built in to accelerate the matrix operation containing Z ik ; a dual-mode communication module for integrating a dual-mode channel of wireless and wired modes, and switching of a primary channel and a backup channel; a dynamic memory for storing a sparse matrix Z and a voltage sensitivity matrix S v .

[0025] The device further comprises: an overvoltage protection circuit for locking reactive power injection when a node voltage > 1.1 p.u.

[0026] According to the first and second aspects of the application, a regulation device is provided, which comprises a processor, a memory, and a computer program stored in the memory and executable on the processor, and the processor implements the distributed photovoltaic regulation method and the fusion communication access implementation method as described above when executing the program.

[0027] According to the first and second aspects of the application, a computer readable storage medium is provided, which stores a computer program, and the computer program is executable on a processor to implement the distributed photovoltaic regulation method and the fusion communication access implementation method as described above.

[0028] The application has the following beneficial effects: The application provides a regulation method and a network access implementation method for a distributed photovoltaic station, solves the problems of a large number of collected control objects, complex operating conditions, real-time distribution, and communication access and remote regulation reliability in a complex wired and wireless communication environment, and improves the distributed photovoltaic regulation capability and the online rate of main station control, thereby improving the distributed photovoltaic construction, grid connection regulation, and distribution network operation stability. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 A distributed photovoltaic station acquisition control terminal and main station connection schematic diagram for an embodiment; Figure 2 A flowchart of a distributed photovoltaic regulation method based on feature grouping and network access for an embodiment; Figure 3 A wireless communication access and quality evaluation flowchart for an embodiment; Figure 4 An electronic device internal structure diagram for an embodiment. DETAILED DESCRIPTION

[0030] To make the technical solutions, technical features, purposes and effects of the present application easy to understand, the present application is further described below in conjunction with specific embodiments and drawings.

[0031] The embodiment of the present application provides a distributed photovoltaic regulation method and device based on feature grouping and network access, which can be applied in a distributed photovoltaic station acquisition control terminal, and the method comprises the steps of Figure 1 The schematic diagram of the distributed photovoltaic station acquisition control terminal and the main station connection for the embodiment, the distributed photovoltaic acquisition control terminal is a data acquisition control device in the distributed photovoltaic station, in actual application, the regulation and control implementation method of the present application is built-in as an independent module in the acquisition control terminal, the data in the distributed photovoltaic station is collected through the distributed photovoltaic acquisition control terminal in the station, and is sent to the main station in a wireless and wired fusion communication mode; the instructions issued by the main station are transmitted to the module of the present application in the acquisition control terminal through the wireless and wired communication mode, and then the corresponding equipment is processed and controlled by the related modules; wherein the controlled equipment can include an inverter, a data collector, a box transformer control, a motion device, an electric meter, a weather monitoring device, etc.

[0032] The embodiment of the present application provides a distributed photovoltaic regulation method based on feature grouping. By grouping the communication network features of multiple inverters of the distributed photovoltaic station, the active power and the reactive power of the inverters are adaptively controlled according to the set power voltage regulation and distribution mode, so that the closed-loop regulation is realized. The distributed photovoltaic AGC and AVC functions integrated in the photovoltaic terminal are supported.

[0033] The number of inverters of the distributed photovoltaic station is not the same, and even reaches hundreds, the data acquisition of the inverter can come from the inverter, the data collector, the box transformer control, the remote device and other equipment, and the data such as the active power, the reactive power, the voltage, the current and the running state of the inverter are affected by each equipment and communication.

[0034] Specifically, as shown in the figure, Figure 2 The present application provides a distributed photovoltaic regulation method based on feature grouping and network access, which comprises: Step S101, dynamically grouping all inverters in the whole station based on grid characteristics and inverter characteristics, generating a sparse label matrix; Step S103, collecting real-time active power and reactive power of each inverter, constructing a power characteristic matrix, and synchronously acquiring a grid node voltage sensitivity matrix; Step S105, acquiring group power limit value based on the sparse label matrix and the voltage sensitivity matrix; Step S107, acquiring dynamic weight coefficient based on the voltage sensitivity matrix; Step S109, dynamically switching channels based on communication quality evaluation value through dual-mode channels supporting wireless and wired; Step S1011, receiving power instructions issued by the master station through the switched channel, and distributing group power in combination with dynamic weight coefficient and group power limit value.

[0035] The step S101 specifically comprises: Based on the grid topology impedance matrix and the electrical distance of the inverter, the m inverters in the whole station are divided into n dynamic groups by improving the spectral clustering algorithm, and an n*m-dimensional sparse label matrix Z is generated, wherein Z ik =1 indicates that the inverter k belongs to the group i, and satisfies the real-time electrical distance constraint D ek ≤D thr , D ek is the electrical distance of the inverter k, and D thr is the distance threshold. The grid characteristics include the grid topology impedance matrix, and the inverter characteristics include the electrical distance of the inverter.

[0036] The grouping of the inverter and the power calculation divide the m inverters in the whole station into n groups. The rule of the group is that one inverter can only be in one group, and one group can have one or more inverters. The indirectly controlled inverter group is one group, and the directly controlled single inverter is one group, to obtain an n*m-dimensional sparse matrix.

[0037] Spectral clustering is a clustering method based on graph theory, which realizes clustering by calculating the similarity between data points (such as similarity matrix, Laplacian matrix). The core idea is to map data to a low-dimensional space, and then use other clustering algorithms (such as k-means) for clustering. Spectral clustering algorithm has good performance in processing nonlinear data and high-dimensional data. The improved spectral clustering algorithm in the invention helps to improve the accuracy and adaptability of clustering.

[0038] The grid topology impedance matrix describes the electrical connection and impedance characteristics between nodes in the grid. The electrical distance reflects the physical or electrical distance between inverters. These information is used to construct the similarity matrix, which further affects the clustering result.

[0039] The inverters are divided into dynamic groups by a clustering algorithm, and each group represents a set of inverters with similar characteristics. This division helps to optimize power grid operation and improve system stability. A sparse label matrix Z is generated to represent the affiliation of each inverter. The elements of the matrix Z ik =1 indicate that the kth inverter belongs to the ith group.

[0040] During the clustering process, the real-time electrical distance constraint D ek ≤D thr must be met to ensure that the clustering results meet the actual operating conditions. This constraint helps to improve the practicality and reliability of the clustering results.

[0041] The step S103 specifically includes: collecting real-time active power P max / P min and reactive power Q max / Q min of the inverters, constructing m×m-dimensional power feature matrices P and Q, and synchronously obtaining the grid node voltage sensitivity matrix . The power feature matrix includes the active power matrix P and the reactive power matrix Q.

[0042] The group label corresponding to the kth inverter in the ith group is Z ik (i=1~n, k=1~m), and the unconfigured bits in a group are 0.

[0043] The m inverter power data form an m×m-dimensional feature matrix P (P max , P min ), Q (Q max , Q min ).

[0044] The dot product of the above two matrices can calculate the power data P i , Q i of the ith group, and the group power limit P imax , P imin , Q imax , Q imin . According to the n group power, the total power P c emitted by the inverters can be calculated.

[0045] Similarly, the group allocation capacity weight coefficient, i.e., the dynamic weight coefficient R i , can be calculated according to the group capacity.

[0046] The maximum and minimum values of active power and reactive power of each inverter need to be collected (P max , Q max ) and (P min , Q min ). These data can be obtained in the following ways: Real-time monitoring and recording, through the real-time data acquisition system of the power system monitoring system or the inverter, the real-time values of the active power and reactive power of each inverter are recorded. These data can be stored in the database for subsequent analysis and processing.

[0047] Historical data and statistical analysis, through statistical analysis of historical data, the maximum and minimum values can be calculated. For example, by analyzing the power data in a period of time, the maximum and minimum values can be determined.

[0048] The control strategy of the inverter (such as droop control, virtual impedance regulation, etc.) can affect the range of its output power. For example, through the adaptive virtual impedance regulation method, the equalization and optimization of the active power of the inverter can be realized.

[0049] In this invention, m x m dimensional feature matrices P and Q need to be constructed. These matrices are used to represent the distribution or characteristics of inverter power. The specific construction methods include: Data normalization and standardization, the collected power data is normalized or standardized to eliminate dimensional differences and facilitate subsequent analysis.

[0050] Feature extraction and dimension reduction, through principal component analysis (PCA), wavelet transform, etc., the key features are extracted to construct the feature matrix.

[0051] Matrix construction, the power data is organized into matrix form, for example, the power data of each inverter is organized into a matrix, where the rows or columns represent different inverters or time points.

[0052] The grid node voltage sensitivity matrix S v describes the relationship between node voltage and power change. Its calculation method can include: Through sensitivity analysis, the sensitivity of node voltage to active power and reactive power can be calculated. For example, the sensitivity of voltage to active power (spv) and reactive power (sqv) can be calculated through node admittance matrix, phase angle difference, etc. In power system analysis, the Jacobian matrix is used to describe the relationship between system state variables and input variables. Through the Jacobian matrix, the relationship between voltage change and power change can be calculated. Specifically, through the node admittance matrix, the phase angle difference, etc. to construct the Jacobian matrix of the power flow equation; then through the inverse matrix of the Jacobian matrix to give the sensitivity of voltage to active power and reactive power, realize the sensitivity analysis; According to the sensitivity analysis results, construct the voltage sensitivity matrix S v , for describing the response of node voltage to power changes.

[0053] Through real-time monitoring of the system, obtain the power data and grid node voltage data of each inverter.

[0054] Use software tools for data processing and analysis, construct the power feature matrix and sensitivity matrix.

[0055] Among them, the step S105 specifically includes: through the modified function containing the Hermite polynomial expansion, the sparse marker matrix, the power feature matrix and the voltage sensitivity matrix are dynamically calculated to determine the group power limit value , the input includes the integral value of Z⊙P and the gradient norm of S v .

[0056] Among them, R i is generated by the exponential decay function of the group historical power integral value and the voltage sensitivity gradient. Among them, the group historical power integral value is determined by the cumulative value of the active power injection in the group historical predetermined time period; the group historical power integral value can also be determined by the Hadamard product norm of the sparse marker matrix and the power feature matrix; the voltage sensitivity gradient is determined by the rate of change of the voltage sensitivity matrix over time.

[0057] Among them, the dynamic weight coefficient R i is calculated by the following formula: In the formula, γ is the gradient correction factor, T is the time window, Sv is the voltage sensitivity matrix, and exp(·) is the exponential decay function.

[0058] Among them, the step S1011 specifically includes: receiving the power instruction P t issued by the master station, and distributing the group power P i according to the dynamic weight coefficient R it .

[0059] Among them, the inverter exits or stops and other abnormal treatments, the inverter exit or stop state is represented by another m×m feature matrix, the corresponding marker feature changes from 1 to 0, and the group power data can be obtained by matrix point multiplication.

[0060] The power distribution of the group is according to the received power instruction (voltage instruction is converted to reactive power instruction through reactive voltage algorithm) P t , according to the group capacity weight coefficient R i , and is distributed to each group P it , Q it .

[0061] .

[0062] The expression of compensation power ΔP is: where t delay Calculated by the delay parameter in the communication quality dataset, η is the compensation gain, and τ is the delay attenuation constant.

[0063] Receive power instructions issued by the master station. The master station (Master Station) sends power instructions to the system or device, which are used to control or adjust the power distribution of the system.

[0064] Dynamic weight coefficients are used to dynamically adjust group power distribution, and their values involve two key parts: Group historical power integral value, which refers to the accumulation or integration of historical power data, used to reflect historical power usage.

[0065] Voltage sensitivity gradient, which refers to the degree of influence of voltage change on power distribution, i.e. the sensitivity of voltage change to power distribution.

[0066] Exponential decay function is a common data weighting processing method, which is used to gradually weaken the influence of past data and make recent data more influential. In this context, an exponential decay function can be used to calculate or adjust its value to reflect the decay effect of historical data.

[0067] Group power distribution, according to the dynamic weight coefficient, the system allocates power to each group (Group). This process may involve dynamic adjustment of power to adapt to changes in the system or load.

[0068] where P it When the limit is exceeded, adjust the allocation value according to the sensitivity priority ratio and inject compensation power ΔP to suppress delay oscillation. The sensitivity priority can be determined by the absolute value of the voltage sensitivity, i.e. the larger the absolute value of the voltage sensitivity, the higher the priority.

[0069] In actual application scenarios, the improved spectral clustering algorithm described in this embodiment divides the m inverters of the entire station into n dynamic groups, which specifically includes: Construct a weight matrix W through the current harmonic feature vector of the inverter; The elements of the weight matrix are where I k is the current harmonic feature vector of inverter k, I j is the current harmonic feature vector of inverter j, k represents the current inverter to be calculated for similarity, j represents other inverters compared with k, I j is the current harmonic feature vector of other inverter j, and δ is the bandwidth parameter of the Gaussian kernel. A corresponding degree matrix is determined based on the weight matrix, and a Laplacian matrix L=D-W is generated, where D is the degree matrix; The Laplacian matrix L is subjected to eigenvalue decomposition to determine the embedding vector of the inverter; Based on the embedding vector of the inverter and the electrical distance constraint of the inverters in the group, the cluster center is dynamically adjusted by Mahalanobis distance; it can be ensured that the inverters in the group satisfy the electrical distance constraint of Z ik =1.

[0070] Wherein, the current harmonic eigenvector of the inverter k can be determined by the grid topology impedance matrix, which can be determined by harmonic injection method based on frequency domain impedance scanning, harmonic separation method based on port network, harmonic source positioning method based on state estimation, inverter parameter backstepping method based on harmonic impedance matching, harmonic feature learning method, etc. Further, the harmonic injection method based on frequency domain impedance scanning obtains the impedance of the grid at the harmonic frequency by offline or online frequency domain scanning and measures the node voltage harmonic, combines the harmonic source characteristics of the inverter, and directly calculates the harmonic current according to Ohm's law to determine the current harmonic eigenvector. The harmonic separation method based on port network regards the inverter and its connected grid as a port network, separates the harmonic current of the inverter by measuring the harmonic components of the port voltage and current, and determines the current harmonic eigenvector. The harmonic source positioning method based on state estimation establishes a harmonic state space model of the grid, uses the measurement data (voltage, current harmonic) and the grid topology impedance, and solves the harmonic current of the inverter by weighted least squares (WLS) or extended Kalman filter (EKF) algorithm to determine the current harmonic eigenvector. The inverter parameter backstepping method based on harmonic impedance matching analyzes the matching relationship between the inverter output impedance and the grid impedance, backstepping the harmonic current of the inverter, and determining the current harmonic eigenvector. The harmonic feature learning method is based on machine learning (such as neural network, random forest), which trains the model based on historical data, establishes the mapping relationship between the grid topology impedance and the harmonic current of the inverter, realizes the real-time prediction of the harmonic current, and determines the current harmonic eigenvector. Wherein, the current harmonic eigenvector can be determined by the statistical and relative measurement of the harmonic current in amplitude, distribution and comprehensive distortion degree. The current harmonic eigenvector I k =[THD k , H3 k , H5 k ] contains the normalized values of total harmonic distortion, 3rd harmonic component and 5th harmonic component.

[0071] Generating a Laplacian matrix L, including: calculating the total weight of each inverter with other inverters through the elements of the weight matrix W; determining the diagonal elements of the degree matrix based on the total weight of each inverter with other inverters, giving the degree matrix of the inverter; generating the Laplacian matrix based on the degree matrix and the weight matrix. Eigen decomposition of L, including: eigen decomposition of L, extracting the eigenvectors corresponding to the first n1 smallest non-zero eigenvalues to form an eigenvector matrix; normalizing the column vectors of the eigenvector matrix to obtain a low-dimensional embedding matrix, each row of the low-dimensional embedding matrix being the embedding vector of the inverter in n1-dimensional space.

[0072] Based on the embedding vector of the inverter and the electrical distance constraint of the inverter within the group, the cluster center is dynamically adjusted by Mahalanobis distance, including: randomly selecting a plurality of inverters as initial cluster centers and determining their embedding vectors; calculating the Mahalanobis distance between the embedding vector of each inverter and the embedding vector of each cluster center, and assigning it to the group where the cluster center with the smallest Mahalanobis distance is located; updating the cluster center after all assignments are completed; calculating the average or maximum Mahalanobis distance between the inverters within each group to ensure that it does not exceed the distance threshold; if there is an inverter that exceeds the distance threshold, it is re-assigned to the nearest group or the center position is adjusted; repeat the steps of updating the cluster center and re-assigning until the cluster center no longer changes significantly or the predetermined number of repetitions is reached, giving a group that satisfies the electrical distance constraint, i.e. a group that satisfies the electrical distance constraint of Z ik =1.

[0073] wherein the correction function is defined as: wherein:‖Z⊙P‖ i represents the i-th row norm of the Hadamard product of the sparse label matrix Z and the active power matrix P; H n (t) is an n-order Hermite polynomial.

[0074] wherein, represents the power attenuation coefficient, which quantifies the contribution weight of the historical power to the current limit value (κ∝1 / T, the larger the time window T, the weaker the historical influence).

[0075] is the group historical power integral value, which represents the cumulative modulus length (L2 norm) of the power vector of the group i in the time period [t0, t0+T], and t0 is the starting time, wherein: Z is the sparse label matrix of the group (Z ik =1 indicates that the inverter k belongs to the group i); P is the active power matrix (P kj represents the power of inverter k at time j); ⊙ is the Hadamard product (element-by-element multiplication); ‖·‖ iEuclidean norm of the i-th row vector.

[0076] Sensitivity weight factor, adjusts the strength of the grid voltage stability constraint on power limits.

[0077] ‖▽S v (i)‖2 is the gradient norm of the voltage sensitivity matrix, reflecting the voltage fluctuation risk of the grid node where group i is located (the larger the gradient, the more power needs to be limited).

[0078] ρ is the light fluctuation correction factor, calibrated in real time by the irradiance sensor.

[0079] H n (t) is an n-order Hermite polynomial (usually 4), used to fit the power fluctuation characteristics under irradiance mutation scenarios, which is in the form of: .

[0080] The role of the Hermite polynomial here is to convert random light disturbance into an explicit mathematical expression, transcending the traditional static limit model.

[0081] The selection of the Hermite polynomial is based on the theory of random processes, and its fitting accuracy for non-Gaussian disturbances (such as cloud cover) is better than that of Fourier series. While the traditional PSO algorithm only optimizes instantaneous power, the present invention introduces historical state memory through the integral term, avoiding local oscillation.

[0082] The step S109 specifically includes: through a dual-mode channel supporting wireless and wired communication, based on the communication quality evaluation value Q com (t) dynamically switches the channel / link.

[0083] Specifically, the system of the present invention supports dual-mode communication of wireless communication and wired communication. This dual-mode communication can support both wireless and wired communication at the same time to adapt to different communication needs and environmental conditions. The system dynamically evaluates the quality of the communication link corresponding to the current communication channel through the communication quality evaluation value Q com (t), and dynamically switches the communication channel according to the evaluation result to ensure the stability and efficiency of communication.

[0084] In practical applications, dual-mode communication technology (such as 4G / 5G dual-mode) has been widely used in various communication devices and networks, such as 5G mobile phones, industrial routers and communication terminals, etc. These devices can automatically switch between different communication modes through dual-mode communication technology to provide more stable and efficient communication services.

[0085] The communication quality evaluation value Q com(t) is an indicator used to evaluate the quality of a communication link, which can be calculated based on factors such as signal strength, network load, data transmission speed, etc. By dynamically switching communication channels, the system can adjust the communication mode according to the real-time communication quality to ensure the continuity and reliability of communication.

[0086] where Q com The calculation method of (t) is: where Θ is the channel capacity, β is the fading factor, μ is the mean of the packet loss rate, σ is the standard deviation, LPLR is the observed value of the packet loss rate, and x is the packet loss rate.

[0087] where the parameters μ and σ are dynamically updated by the Kalman filter: State equation: μ k =Aμ k-1 +Bu k +w k ; Observation equation: σ k =Cσ k-1 +v k ; where u k is the real-time collected packet loss rate LPLR sequence, w k is the process noise, μ k-1 represents the state estimate of the previous time, σ k-1 represents the observation value or the covariance of the state estimate of the previous time, v k is the observation noise.

[0088] The Kalman filter is a recursive algorithm for estimating the state of a system, whose core idea is to continuously optimize the estimation of the system state through prediction and update steps.

[0089] In the Kalman filter, the evolution of the system state and the generation of the observation value are described by the state equation and the observation equation, respectively. The state equation describes how the system state transfers from one time point to the next time point. The observation equation describes how to generate the observation value from the system state.

[0090] The Kalman filter performs state estimation through the following two steps: Prediction step, based on the state estimate and input of the previous time, to predict the state and covariance of the current time.

[0091] Update step, according to the observation value to correct the predicted state to improve the estimation accuracy.

[0092] The parameters μ and σ represent the estimates of the system state and observation value, respectively. Through the Kalman filter, these parameters can be dynamically updated through the state equation and observation equation. Specifically: μk represents the state estimation value at the current time, which is constantly adjusted through the update of the state equation and the observation equation.

[0093] σ k represents the covariance of the observation value or state estimation at the current time, which is constantly optimized through the adjustment of the Kalman gain.

[0094] The parameters μ and σ are dynamically updated by the Kalman filter, and their state equations and observation equations describe the evolution of the system state and the generation of the observation value. The Kalman filter constantly optimizes the estimation of the system state through the prediction and update steps.

[0095] where A is the state transition matrix, which is a p x p matrix describing the spontaneous evolution of the state variable μ from k-1 to k.

[0096] B is the control input matrix, which is a p x p matrix, and p is the input dimension, which maps the external control quantity u k to the state space.

[0097] C is the observation matrix, which is a m x p matrix, and m is the observation dimension, which describes how the state σ is mapped to the observation space.

[0098] The determination of the communication quality evaluation value includes: monitoring characteristic parameters related to channel communication quality in real time, and establishing a communication quality comprehensive data set; determining the average signal strength and packet loss rate coefficient based on the communication quality comprehensive data set; based on the average signal strength and packet loss rate coefficient, and combined with the preset communication condition, the channel communication quality is obtained; wherein the characteristic parameters include signal strength, signal delay and packet loss rate, and the preset communication condition includes signal delay threshold, signal delay weight and signal strength weight. As Figure 3 shown, in actual application scenarios, the communication card is dialed by the wireless communication module, and after the communication card is normally networked and the card ip is obtained, the communication quality related characteristic parameters are continuously monitored, and the monitored data is sent to the communication processing module, and the communication quality is evaluated by the communication processing module, and a comprehensive data set is established, which includes multiple characteristic parameters such as signal strength, delay, and packet loss rate. Based on the comprehensive data set, the channel communication quality is evaluated.

[0099] The established communication quality comprehensive data set is: wherein: Q is the communication quality comprehensive evaluation value, which is used to represent the current communication quality evaluation situation, and its calculation formula is: t is the current time, that is, the timestamp, used to represent the current statistical time information; R is the current signal strength, which represents the current received signal strength instantaneous value, which is obtained from the 4G / 5G communication module through AT instructions. R is related to the network type. Under the 4G network, R is (RSSI, Received Signal Strength Indication), and under the 5G network, R is (SSRSRP, Synchronization Reference Signal Received Power), that is: Under the 4G / 5G network, R is: R bar Average signal strength, the calculation method is , which represents the average value of the received signal strength in a period of T time, where T can be selected according to the actual situation. The smaller T is, the more sensitive R bar represents the average signal strength change, T is generally 1 minute; k R Signal strength weight, which is 0%~100%, generally 50%, used to adjust the weight of the signal strength feature in the communication quality comprehensive data set. The weight can be adjusted according to the signal coverage of the used card, the signal strength stability, and the surrounding frequency band interference. T Delay is the current signal delay, which can be obtained by monitoring ICMP packets, T Delaythr is the signal delay threshold, in milliseconds, T thr is 200ms; k T is the signal delay weight, which is 0%~100%, generally 50%, which is used to adjust the weight of the signal delay feature in the communication quality comprehensive data set. The weight can be adjusted according to the stability of the operator network of the used card. L PLR Packet loss rate, which can be obtained by monitoring ICMP packets, and its size is 0%~100%.

[0100] k PLR is the packet loss rate coefficient, which has only two states of 0 and 1. When the packet loss rate L PLR is equal to 0%, that is, there is no packet loss, k PLR is 1. If the packet loss rate L PLR is not 0%, that is, greater than 0% and less than or equal to 100%, it means that there is packet loss, and the communication appears to be blocked or disconnected, then k PLR is 0.

[0101] The communication quality comprehensive data set includes communication quality observable characteristic parameters, signal strength related parameters mainly used for evaluating signal strength, and signal delay and packet loss rate related parameters used for evaluating network performance.

[0102] Through the above method, the communication quality comprehensive data set is established, and the communication quality of the two channels is continuously evaluated, which can form minute-level, hour-level and 24-hour-level communication quality comprehensive data set records for users to check, facilitate to find communication problems and provide data support for communication optimization.

[0103] The method further comprises: constructing a Lyapunov function V(x) = xTQx / 2, where Q is a positive definite matrix. Verify that dV / dt < 0.

[0104] Lyapunov function is a mathematical tool used to analyze the stability of dynamic systems. It is usually defined as a scalar function V(x), where x is the state vector of the system. The function has the following characteristics: Non-negativity, V(x) = 0 at equilibrium points, and V(x) > 0 at other points.

[0105] Non-increasing, the time derivative of V along the system trajectory , that is, the energy of the system does not increase. If , then the system is asymptotically stable.

[0106] Lyapunov method is based on Lyapunov stability theorem, which points out that if there is a function that satisfies the above conditions, then the system is stable at the equilibrium point. This method does not need to solve the differential equation of the system, but constructs an auxiliary function to judge the stability of the system.

[0107] The method further comprises: issuing the modified instruction to the group inverter to realize AGC / AVC closed-loop regulation.

[0108] The dispatching center (such as DMS master station) issues instructions to the photovoltaic power station, which usually includes target values such as active power, reactive power, power factor, etc. These instructions are transmitted to the control terminal of the photovoltaic power station through the communication network (such as 5G, Ethernet, wireless communication, etc.).

[0109] The control terminal of the photovoltaic power station (such as AGC fusion terminal) receives and analyzes these instructions. According to the running state, capacity, efficiency and other parameters of the on-site inverter, the instructions are decomposed into the adjustment amount of each inverter.

[0110] The control terminal sends specific adjustment instructions to each inverter according to the decomposed instructions to adjust the active power, reactive power or start-stop state of the inverter, so as to realize accurate control of the output of the photovoltaic power station.

[0111] The system forms a closed-loop control by monitoring the operation state and adjustment effect of the inverters in real time. If part of the inverters fail to complete the instructions, the system will automatically redistribute the tasks to ensure the completion of the instructions.

[0112] AGC (Automatic Generation Control) is responsible for adjusting the active power of the photovoltaic power station to meet the peak shaving demand of grid dispatching; AVC (Automatic Voltage Control) is responsible for adjusting the reactive power to maintain the stability of the grid voltage.

[0113] Communication and data interaction: The entire process relies on communication networks and data interaction, including data acquisition, instruction transmission, state feedback, etc., to ensure the stability and reliability of the system.

[0114] Through the above process, the distributed photovoltaic power station can realize the closed-loop regulation of AGC and AVC, improving the stability and flexibility of the grid.

[0115] In this paper, the meaning of the parameters is summarized: Z top is the grid impedance matrix, D ek is the electrical distance of inverter k, D thr is the distance threshold, Z is the group label matrix, P / Q is the power matrix, S v is the voltage sensitivity matrix, κ is the power attenuation coefficient, λ is the sensitivity weight, T is the integration window, γ is the gradient correction factor, R i is the dynamic weight coefficient, P t is the master station instruction, Θ is the channel capacity, β is the fading factor, μ is the average packet loss rate, σ is the standard deviation, LPLR is the packet loss rate observation value, η is the compensation gain, τ is the delay attenuation constant, is the ideal power, x is the system state variable.

[0116] The second aspect of the present application provides a distributed photovoltaic regulation device based on feature grouping and network access, which comprises: a processing unit for executing the above method, a parallel computing unit is built in to accelerate the matrix operation containing Z ik ; a dual-mode communication module for integrating 4G / 5G modules and wired communication modules, supporting automatic switching of primary and backup channels; a dynamic memory for storing sparse matrix Z and voltage sensitivity matrix S v .

[0117] The device further comprises: an overvoltage protection circuit for locking reactive power injection when the node voltage >1.1p.u.

[0118] The parallel computing unit is built in the processing unit to accelerate the matrix operation containing Zik matrix operation.

[0119] The communication implementation method of the embodiment of the application provides the following two communication interface modules, supports simultaneous access of wireless networks and wired networks, and provides a communication processing module to realize automatic switching of the network according to master station channel selection, network communication conditions and user access point adjustment.

[0120] The wireless communication module: provides two standard interfaces 4G / 5G communication modules, including a 4G public network module, a 4G power special network module, a 4G public and special network integrated module and a 5G module; the module can realize a variety of network standard SIM card dual-card dual standby, is used for wireless communication, and establishes a communication link with the master station, and simultaneously continuously monitors and obtains wireless communication quality related data.

[0121] The wired communication module: the device can provide two standard interfaces, is used for wired communication, supports EPON and dispatching data network interfaces, and simultaneously continuously monitors and obtains communication quality related data.

[0122] The two communication interface modules can provide wired and wireless fusion network, wireless master and backup channel, wireless double channel, wired master and backup channel, wired double channel and the like access mode selection. It is suitable for the communication mode of the public wireless network, special network wireless network, EPON and dispatching data network of the current distributed photovoltaic station.

[0123] The communication processing module: according to the preset communication configuration and double channel mode, a corresponding communication processing mode is selected, and automatic switching of communication is realized. For example, if the wired or wireless communication works in one master and one backup, if the wireless communication is interrupted or communication interference, the wired communication can be automatically switched, and vice versa. If the wired or wireless communication works in double channels, if the wireless communication is interrupted or the wired communication is interrupted, the communication module sends a channel interruption alarm but does not need to switch the communication. For example, if the wireless master and backup channel or the wired master and backup channel communication is selected: when the main channel communication is interrupted or communication interference, the backup channel can be automatically switched, and the main channel communication can be automatically switched (or manually switched) back to the main channel. For example, if the wireless double channel or wired double channel communication is selected: when one channel communication is interrupted or communication interference, the channel with communication interruption sends alarm information, and the other channel still works normally.

[0124] Communication monitoring and recording: the communication processing module monitors the communication quality according to the working mode of the two channels, and performs evaluation, if the main link appears wireless signal strength deterioration communication interruption, wired communication interruption, alarm, recording can be performed, minute level, hour level, 24 hour level communication quality comprehensive data set record can be formed; the operation and maintenance personnel can refer to analyze the problem, so as to guarantee the communication quality between the distributed photovoltaic and the master station.

[0125] The processing unit includes a dedicated integral operation unit, whose hardware architecture completes the integral operation of Z ik Element integral item calculation: .

[0126] The dual-mode communication module integrates a 4G / 5G wireless module and a wired module, and supports automatic switching of a primary and a backup channel.

[0127] The dynamic memory stores a sparse matrix Z and a voltage sensitivity matrix S v Sparse matrices are commonly used for efficient storage and calculation in parallel computing.

[0128] In the overvoltage protection circuit, when the node voltage > 1.1 p.u, the reactive power injection is locked out. Overvoltage protection circuits are commonly used in power systems to prevent overvoltage damage to equipment.

[0129] The embodiments provide a distributed photovoltaic wireless primary and backup channel access, a wireless dual-channel access method, and a wireless and wired fusion network access.

[0130] In the communication process of selecting the wireless primary and backup channel access method, the communication quality of the two links is continuously monitored and evaluated, and minute-level, hour-level, and 24-hour-level communication quality comprehensive data set records are formed for users to operate and optimize the network. At the same time, if the communication quality of the primary link deteriorates and reaches a certain threshold value, the condition for reselecting the primary and backup links is met, and one of the following three conditions is met: 1) Communication quality comprehensive data set The current signal strength R deteriorates to reach the threshold condition, that is, the signal strength R is lower than the average signal strength value and reaches a predetermined threshold value, which is generally 20% lower than the average signal strength value To prevent accidental signal strength R fluctuations within a short period of time, when the signal strength R is lower than the average signal strength value and reaches the threshold value, the signal strength R is collected for five times at an interval of 1 second, and if three of the five times have a signal strength R lower than the average signal strength value and reach the threshold value, the condition for reselecting the primary and backup channels is met; 2) Communication delay deterioration reaches the threshold condition, that is, the number of packets with a delay greater than the delay threshold value in a single delay monitoring reaches a certain value, which is generally 3 packets with a delay greater than 200 ms in a group of 5 monitoring packets with an interval of 1 second, then the condition for reselecting the primary and backup channels is met; 3) Communication error rate deterioration reaches the threshold condition, that is, there is a packet loss, and the error rate monitoring shows that the error rate is not 0%, then the condition for reselecting the primary and backup channels is met.

[0131] If any of the above conditions is met, i.e., the condition for reselecting the main and standby channels is reached, the comprehensive evaluation value Q of the link quality of the two channels is compared. If the current main link Q is less than the current standby link Q after comparison, the current standby link is selected as the main link, and the data packet is processed. Otherwise, the current main link remains unchanged. At the same time of reselecting the main and standby links, an alarm is given and recorded.

[0132] If the dual-channel access mode is selected, the device processes the data packet through the communication processing module to realize the communication of the two channels with the master station at the same time, i.e., one channel corresponds to one master station, and the other channel corresponds to the other master station. One is used for communication with the distributed photovoltaic collection control terminal under the first channel, and the other is used for communication with the distributed photovoltaic collection control terminal under the second channel. Further, during the communication process of the dual-channel access mode, the communication quality of the two links is continuously monitored and evaluated, and the comprehensive data set record of the communication quality at the minute, hour, and 24-hour levels can be formed, and the communication quality deterioration alarm and record can be formed for the operation and maintenance personnel to check and analyze the problem, optimize the network, and help improve the communication stability of the distributed photovoltaic collection data uploading and remote control.

[0133] If the wireless and wired fusion network communication is selected, it works in dual-channel mode, which is similar to the wireless dual-channel access mode. Further, during the communication process of the dual-channel access mode, the communication quality of the two links is continuously monitored and evaluated, and the comprehensive data set record of the communication quality at the minute, hour, and 24-hour levels can be formed, and the communication quality deterioration alarm and record can be formed. When it works in the main and standby channel communication, the communication quality of the wireless link is continuously monitored and evaluated, and the comprehensive data set record of the communication quality at the minute, hour, and 24-hour levels can be formed. At the same time, if the main link appears to have poor communication quality in the monitoring, and reaches a certain threshold, the condition for reselecting the main and standby links is reached.

[0134] The above modules can be realized by a combination of software and hardware, and each module can be embedded in or independent of the processor in the electronic device in hardware form, or stored in the memory in the electronic device in software form, so that the processor can call and execute the operations of each module.

[0135] Based on the above embodiments, an electronic device is also provided, which can be a standalone terminal device. Referring to Figure 4For the internal structure diagram of the embodiment electronic device, the electronic device includes: a processor, a memory, a system bus, an I / O interface, a communication interface, etc., wherein the processor implements various distributed photovoltaic wireless network access and communication quality evaluation methods as described in the embodiments when executing the program, the communication interface includes at least two Ethernet interfaces, two interfaces connected with wireless modules, and a serial port, wherein the Ethernet interface provides external connection for external communication, supports 10 / 100Mbps rate adaptation, and cross direct connection adaptation; the serial port can provide external connection through a predetermined interface for local debugging of the device.

[0136] Based on the above embodiments, a computer readable storage medium is also provided, which includes: a read-only memory, a random access memory, a flash memory, a mobile hard disk, and the like readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the distributed photovoltaic regulation method based on feature grouping and network access as described in the embodiments.

[0137] The above embodiments only express the implementation of the present application, and the description is more specific and detailed. However, the present application is not limited to the above specific implementation. The above specific implementation is only illustrative, not restrictive. Those skilled in the art can make many forms under the inspiration of the present application without departing from the scope of the present application and the protection scope of the claims.

Claims

1. A distributed photovoltaic control method based on feature clustering and network access, characterized in that, include: Step S101: Based on the characteristics of the power grid and the inverter, dynamically group all inverters in the entire station to generate a sparse label matrix; Step S103: Collect the real-time active power and reactive power of each inverter, construct the power feature matrix, and simultaneously obtain the grid node voltage sensitivity matrix. Step S105: Based on the sparse label matrix and voltage sensitivity matrix, obtain the group power limit; Step S107: Obtain dynamic weighting coefficients based on the voltage sensitivity matrix; Step S109: Dynamically switch channels based on communication quality assessment values ​​by supporting dual-mode channels that support both wireless and wired connections. Step S1011: Receive the power command issued by the master station through the switched channel, and allocate the group power in combination with the dynamic weighting coefficient and the group power limit.

2. The distributed photovoltaic control method based on feature clustering and network access according to claim 1, characterized in that, Step S101 specifically includes: Based on the grid topology impedance matrix and inverter electrical distance, an improved spectral clustering algorithm is used to divide all inverters in the entire station into multiple dynamic groups, and a sparse label matrix is ​​generated based on the division results.

3. The distributed photovoltaic control method based on feature clustering and network access according to claim 2, characterized in that, Step S103 specifically includes: collecting the real-time active power and reactive power of the inverter, constructing the corresponding power feature matrix, and synchronously acquiring the grid node voltage sensitivity matrix.

4. The distributed photovoltaic control method based on feature clustering and network access according to claim 3, characterized in that, Step S105 specifically includes: dynamically calculating the sparse label matrix, power characteristic matrix, and voltage sensitivity matrix using a correction function containing Hermitian polynomial expansion to determine the group power limit.

5. The distributed photovoltaic control method based on feature clustering and network access according to claim 4, characterized in that, Step S107 specifically includes: jointly generating dynamic weighting coefficients by combining the integral value of the historical power of the group determined based on the sparse label matrix and the power feature matrix with the exponential decay function of the voltage sensitivity gradient.

6. The distributed photovoltaic control method based on feature clustering and network access according to claim 5, characterized in that, The specific steps of step S101 include: receiving a power command issued by the master station, allocating group power according to the dynamic weighting coefficient, and when the allocated group power exceeds the group power limit, adjusting the group power allocation value according to the priority ratio of voltage sensitivity, and injecting compensation power to suppress delayed oscillation.

7. The distributed photovoltaic control method based on feature clustering and network access according to claim 2, characterized in that, The improved spectral clustering algorithm divides all inverters in the entire station into multiple dynamic groups, specifically including: A weight matrix is ​​constructed using the current harmonic eigenvectors of the inverter. Elements of the weight matrix W , where I k Let I be the current harmonic characteristic vector of inverter k. j Let δ be the current harmonic characteristic vector of inverter j, k represent the current inverter to be calculated, j represent other inverters compared with inverter k, and δ be the bandwidth parameter of the Gaussian kernel. The corresponding degree matrix is ​​determined based on the weight matrix, and the Laplacian matrix L=DW is generated, where D is the degree matrix; Eigenvalue decomposition is performed on the Laplacian matrix L to determine the embedding vector of the inverter; Based on the inverter's embedding vector and the electrical distance constraint between inverters within the group, the cluster center is dynamically adjusted using Mahalanobis distance.

8. The distributed photovoltaic control method based on feature clustering and network access according to claim 6, characterized in that, The compensation power satisfies the following relationship: ； In the formula, ΔP is the compensation power, and P it For group power, t delay The value is obtained by calculating the delay parameter from the communication quality dataset, where η is the compensation gain. τ is the delay decay constant.

9. The distributed photovoltaic control method based on feature clustering and network access according to claim 1, characterized in that, In step S109, the determination of the communication quality assessment value includes: real-time monitoring of characteristic parameters related to channel communication quality and establishing a comprehensive communication quality dataset; determining the average signal strength and packet loss rate coefficient based on the comprehensive communication quality dataset; and obtaining the channel communication quality based on the average signal strength and packet loss rate coefficient, combined with preset communication conditions; wherein the characteristic parameters include signal strength, signal delay, and packet loss rate.

10. A distributed photovoltaic control device based on feature clustering and network access, characterized in that, The device includes: The processing unit is used to execute any one of the methods of claims 1-9, and has a built-in parallel computing unit to accelerate matrix operations containing sparsely labeled matrices; Dual-mode communication module, which is used to integrate wireless and wired dual-mode channels, as well as the switching between primary and backup channels; Dynamic memory, which is used to store sparse matrices and voltage sensitivity matrices.

Images