Photovoltaic cluster flexible supporting method and system based on power line carrier

Abstract

The invention provides a photovoltaic cluster flexible support method and system based on power line carrier, and the method comprises the steps: obtaining a power distribution network vulnerability thermodynamic diagram and a power distribution network stability thermodynamic diagram, carrying out the fault sensing and early warning of a power distribution network through combining a power distribution network early warning and sensing mechanism, and generating a fault early warning and sensing result; acquiring multi-source data of the power distribution network, generating a power distribution network state data set, and performing fault diagnosis on the fault early warning sensing result and the power distribution network state data set based on a space-time diagram convolutional network; a fault diagnosis result is obtained, the fault diagnosis result is analyzed and processed based on a Byzantine fault-tolerant consensus and a VCG auction mechanism, and an initial fault buffering scheme is generated; and performing collaborative optimization on the initial fault buffering scheme based on a multi-agent near-end strategy optimization algorithm and a bat-particle swarm algorithm to generate an optimal fault buffering scheme, and performing flexible support on the fault power distribution network in combination with an adaptive virtual impedance algorithm, thereby realizing rapid, accurate and flexible support of the photovoltaic cluster on the power distribution network fault.

Description

Technical Field

[0001] This invention relates to the field of power line carrier technology, and more specifically, to a flexible support method and system for photovoltaic clusters based on power line carrier. Background Technology

[0002] With the widespread application of photovoltaic power generation, photovoltaic inverters, as the core interface for interaction between distributed power sources and the power grid, have an increasingly prominent impact on system stability due to their control strategies. The reactive power and voltage control of traditional distribution networks mainly rely on centralized voltage regulation equipment in main substations and reactive power compensation strategies for inverters based on local measurements. However, in high-proportion photovoltaic scenarios, voltage transient problems caused by faults exhibit multi-source and cross-regional characteristics, and traditional methods often struggle to achieve rapid, accurate, and coordinated flexible support.

[0003] On the one hand, inverter autonomous control strategies based on local measurements often only consider local node information and lack a global perception of the overall network operation status, which often leads to conflicts in multi-node regulation and thus triggers oscillation risks. On the other hand, traditional methods often have certain limitations in terms of accurate quantification of fault responsibility and practical constraints such as communication channel quality, resulting in insufficient adaptability of control strategies in actual heterogeneous network environments, thus leading to low fault recovery capability and operational resilience of distribution networks with high photovoltaic coverage.

[0004] Therefore, there is an urgent need to develop a collaborative control method that can integrate global perception, dynamic weight consensus and adaptive optimization, so as to enable photovoltaic clusters to provide rapid, accurate and flexible support for power distribution network faults while ensuring communication reliability. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a flexible support method and system for photovoltaic clusters based on power line carrier communication, thereby solving the technical problems in the prior art, such as the lack of global perception of the entire network's operational status and the low fault recovery capability and operational resilience of distribution networks with a high proportion of photovoltaic power.

[0006] The purpose and effectiveness of the flexible support method and system for photovoltaic clusters based on power line carrier communication of the present invention are achieved by the following specific technical means:

[0007] A flexible support method for photovoltaic clusters based on power line carrier communication includes:

[0008] S1: Obtain the distribution network vulnerability heat map and distribution network stability heat map, and combine them with the distribution network early warning and perception mechanism to perform distribution network fault perception and early warning, and generate fault early warning and perception results;

[0009] S2: Acquire multi-source data of the distribution network, generate a distribution network status dataset, and perform fault diagnosis based on the fault early warning perception results and the distribution network status dataset using a spatiotemporal graph convolutional network.

[0010] S3: Obtain fault diagnosis results, analyze and process the fault diagnosis results based on Byzantine fault-tolerant consensus and VCG auction mechanism, and generate an initial fault buffer scheme;

[0011] S4: The initial fault buffering scheme is jointly optimized based on the multi-agent proximal policy optimization algorithm and the bat-particle swarm algorithm to generate the optimal fault buffering scheme.

[0012] S5: Each photovoltaic node in the photovoltaic cluster executes the optimal fault buffering scheme and combines it with the adaptive virtual impedance algorithm to provide flexible support for the faulty distribution network.

[0013] As a further aspect of the present invention, the fault diagnosis results are obtained, and the fault diagnosis results are analyzed and processed based on Byzantine fault-tolerant consensus and VCG auction mechanism to generate an initial fault buffer scheme, including:

[0014] The fault diagnosis results are analyzed and processed based on the Byzantine fault-tolerant consensus method and fuzzy reasoning to generate a fault buffer auction list, which includes at least a fault buffer target list, buffer boundary constraints and fault transfer prediction.

[0015] Based on the VCG auction mechanism and the aforementioned fault buffer auction list, the photovoltaic cluster is auctioned to generate the optimal winning combination. Then, based on cooperative game theory, the optimal winning combination is negotiated and negotiated to generate an initial fault buffer scheme.

[0016] As a further aspect of the present invention, the fault diagnosis results are analyzed and processed based on the Byzantine fault-tolerant consensus method and fuzzy reasoning to generate a fault buffer auction list, including:

[0017] Each photovoltaic node in the photovoltaic cluster acquires local electrical transient data and performs electrical transient data prediction based on a pre-trained lightweight LSTM network to generate predicted electrical transient data.

[0018] Based on the fault source contribution matrix contained in the fault diagnosis results, a basic responsibility weight is generated for each photovoltaic node. The basic responsibility weight is used to quantify the response priority of the photovoltaic node to the fault source.

[0019] The transient electrical quantity data is fuzzed and defuzzified to generate a severity factor. The transient electrical quantity data is then processed according to a preset urgency function to generate a prediction urgency factor.

[0020] Based on a preset adaptive weighting mechanism, the basic responsibility weight, severity factor, and predicted urgency factor are adaptively weighted and fused to generate dynamic responsibility weights. Combined with the Byzantine fault-tolerant consensus method, a fault buffer consensus is reached to generate a fault buffer auction list.

[0021] As a further aspect of the present invention, the method further includes:

[0022] Based on the fault buffer auction list, the buffer demand corresponding to the fault diagnosis results is decomposed into composable buffer task packages. Each photovoltaic node responds to the combination of buffer task packages based on local actual cost data, wherein the actual cost data includes at least power generation cost data, equipment operation and maintenance cost data, and grid security risk cost data.

[0023] Based on a preset objective function with the goal of minimizing the total actual cost data, the photovoltaic nodes of the bidding response are combined and screened in combination with preset constraint rules to generate the optimal winning combination with the minimum total cost. The preset constraint rules include at least the constraints of precise allocation of buffer task packages, line power flow compliance, node voltage safety, and output power capacity.

[0024] Based on the Nash bargaining model, the bid prices of each winning node in the optimal winning bid combination are negotiated and negotiated to generate an initial fault buffer scheme.

[0025] As a further aspect of the present invention, the initial fault buffering scheme is collaboratively optimized based on a multi-agent proximal policy optimization algorithm and a bat-particle swarm optimization algorithm to generate an optimal fault buffering scheme, including:

[0026] Taking the initial fault buffer scheme as input, the multi-agent proximal policy optimization algorithm is used in a pre-built standard digital twin environment to generate the optimization adjustment direction of the initial fault buffer scheme. Combined with the bat-particle swarm algorithm, the distributed scheme is fine-tuned to generate the optimal fault buffer scheme.

[0027] The optimal fault buffering scheme is distributed to the photovoltaic cluster. Real-time distribution network status dataset of the photovoltaic cluster executing the optimal fault buffering scheme within a predetermined execution cycle is collected. After a predetermined execution cycle ends, the real-time distribution network status dataset and the corresponding optimal fault buffering scheme are re-input into the standard digital twin environment to perform feedback optimization of the optimal fault buffering scheme, generate a new optimal fault buffering scheme, and distribute it to the photovoltaic cluster for implementation.

[0028] Repeat the feedback optimization operation until the distribution network status is stable and the optimization benefit is lower than the preset threshold or the distribution network fault is eliminated, then exit the loop.

[0029] As a further aspect of the present invention, the method further includes:

[0030] The scheme optimization model is constructed based on the multi-agent near-end policy optimization algorithm. The scheme optimization model includes a policy sub-network and a value sub-network. The policy sub-network is used to generate optimization policies for the schemes in the input model based on the real-time distribution network status dataset, and the value sub-network is used to evaluate the expected value of the schemes after implementing the optimization policies.

[0031] Obtain the deviation between the actual effect and the expected result after implementing the optimal fault buffering scheme. Construct a real-time power grid state vector based on the deviation value and the real-time power distribution network state dataset. Input the real-time power grid state vector and the optimal fault buffering scheme into the scheme optimization model for scheme optimization.

[0032] The corresponding parameters of the strategy subnetwork and the value subnetwork are updated based on the real-time power grid state vector. The strategy subnetwork generates an optimization strategy. The optimal fault buffer scheme is tuned based on the optimization strategy using the bat-particle swarm optimization algorithm. During the tuning process, the value of the scheme is evaluated based on the value subnetwork to generate an optimal fault buffer scheme that is more in line with the current distribution network state.

[0033] As a further aspect of the present invention, each photovoltaic node in the photovoltaic cluster executes the aforementioned optimal fault buffering scheme, and combines it with an adaptive virtual impedance algorithm to provide flexible support for the faulty distribution network, including:

[0034] The optimal fault buffering scheme is decomposed to generate a buffer instruction set input to each photovoltaic node. Each photovoltaic node analyzes and processes the buffer instruction set based on an adaptive virtual impedance algorithm to generate virtual impedance data. The virtual impedance data is used to smoothly execute the buffer instruction set and includes at least virtual resistance and virtual inductance.

[0035] Based on the virtual impedance data, a smoothing auxiliary instruction set is generated. Each photovoltaic node simultaneously executes the buffer instruction set and the corresponding smoothing auxiliary instruction set to provide flexible support for the faulty distribution network.

[0036] As a further aspect of the present invention, a distribution network vulnerability heatmap and a distribution network stability heatmap are obtained, and a distribution network fault detection and early warning mechanism is combined to generate a fault early warning detection result, including:

[0037] Obtain the current distribution network topology diagram and analyze the intermediary centrality of each structure in the network topology diagram. The larger the intermediary centrality value, the higher the vulnerability of the corresponding structure. Generate a distribution network vulnerability heat map based on the intermediary centrality of each structure.

[0038] Periodically inject micro-disturbance pulses with amplitudes not exceeding the rated power and frequencies within the rated frequency of the power grid into the distribution network, and collect voltage and frequency response data of all nodes in the distribution network. Analyze the data using the recursive least squares method to construct a distribution network stability heatmap to describe the dynamic impedance and stability margin of the distribution network. When the stability margin is lower than a set threshold, trigger a distribution network fault detection early warning flag and generate a fault early warning detection result. The fault early warning detection result includes at least the suspected fault point and the corresponding intermediate centrality and stability margin.

[0039] As a further aspect of the present invention, multi-source data of the distribution network is acquired to generate a distribution network status dataset. Fault diagnosis is then performed based on the fault early warning perception results and the distribution network status dataset using a spatiotemporal graph convolutional network, including:

[0040] When a fault occurs in the distribution network, data is collected from the distribution network to obtain transient electrical quantity data and communication channel status data, and a distribution network status dataset is generated. The transient electrical quantity data includes at least the three-phase voltage phasors and three-phase current phasors of each node in the distribution network, and the communication channel status data includes at least the received signal strength indication, signal-to-noise ratio, channel delay, and bit error rate.

[0041] Based on the independent component analysis method and the fault early warning perception results, the transient data of electrical quantities are analyzed and processed to generate a fault source contribution matrix. The fault source contribution matrix is ​​used to quantify the percentage contribution of each fault source to the drop in transient data of electrical quantities at different nodes in the distribution network.

[0042] Obtain the network topology diagram of the distribution network, input the network topology diagram, the distribution network status dataset, and the fault source contribution matrix into a spatiotemporal graph convolutional network for fault diagnosis, and generate fault diagnosis results. The fault diagnosis results include at least the fault location, fault type, fault severity, predicted impact range, and fault source contribution matrix.

[0043] This invention also discloses a flexible support system for photovoltaic clusters based on power line carrier communication, comprising:

[0044] The data acquisition module is used to acquire the distribution network vulnerability heat map and the distribution network stability heat map, collect multi-source data of the distribution network, and generate a distribution network status dataset.

[0045] The fault early warning module is used to perform fault perception and early warning of the distribution network according to the distribution network early warning and perception mechanism, and generate fault early warning and perception results.

[0046] The fault diagnosis module is used to perform fault diagnosis based on the fault early warning perception results and the power distribution network status dataset using the spatiotemporal graph convolutional network, and generate fault diagnosis results.

[0047] The scheme generation module is used to analyze and process the fault diagnosis results based on the Byzantine fault tolerance consensus and VCG auction mechanism to generate an initial fault buffer scheme.

[0048] The scheme optimization module is used to collaboratively optimize the initial fault buffer scheme based on the multi-agent proximal policy optimization algorithm and the bat-particle swarm algorithm to generate the optimal fault buffer scheme.

[0049] A buffer execution module is used to provide flexible support for the faulty distribution network based on an adaptive virtual impedance algorithm and an optimal fault buffering scheme.

[0050] Based on the above aspects, the embodiments of this application realize the acquisition of distribution network vulnerability heat maps and distribution network stability heat maps, and combine them with the distribution network early warning perception mechanism to perform distribution network fault perception and early warning, generate fault early warning perception results, acquire multi-source data of the distribution network, generate distribution network status dataset, and perform fault diagnosis on the fault early warning perception results and distribution network status dataset based on spatiotemporal graph convolutional network. By performing distribution network fault perception and early warning, and combining it with spatiotemporal graph convolutional network, the responsibility of each node and the fault propagation path are clarified from the global perspective of the distribution network, providing a physical basis for the subsequent generation of corresponding scheme strategies.

[0051] The fault diagnosis results are obtained and analyzed based on the Byzantine fault-tolerant consensus and VCG auction mechanism to generate an initial fault buffer scheme. Through the weighted Byzantine fault-tolerant consensus mechanism, nodes quickly exchange information and reach consensus locally. At the same time, in the weighted Byzantine fault-tolerant consensus mechanism, the dynamic voting weight of the nodes is deeply integrated with their responsibility weight and real-time communication channel status, so that the decision-making power for generating the initial fault buffer scheme is tilted towards the nodes with greater responsibility and better communication, thereby improving the efficiency of reaching consensus and the reliability of the initial fault buffer scheme.

[0052] The initial fault buffering scheme is collaboratively optimized using a multi-agent proximal policy optimization algorithm and a bat-particle swarm optimization algorithm to generate the optimal fault buffering scheme. By combining the multi-agent proximal policy optimization algorithm and the bat-particle swarm optimization algorithm to optimize the initial fault buffering scheme, it is ensured that the actions of each node serve the global optimal goal, thereby avoiding the risk of oscillation caused by local decision conflicts. At the same time, the optimal fault buffering scheme is continuously optimized in an iterative loop, so that the optimal fault buffering scheme always fits the current fault state, thereby improving the reliability of the optimal fault buffering scheme.

[0053] Each photovoltaic node in the photovoltaic cluster executes the optimal fault buffering scheme and combines it with an adaptive virtual impedance algorithm to provide flexible support for the faulty distribution network. The photovoltaic inverter directly executes the optimal instruction corresponding to the optimal fault buffering scheme through virtual impedance technology, thereby achieving flexible support for the faulty distribution network. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the execution flow of a photovoltaic cluster flexible support method based on power line carrier provided in an embodiment of the present invention;

[0055] Figure 2 This is a schematic diagram of a flexible support system for a photovoltaic cluster based on power line carrier provided in an embodiment of the present invention. Detailed Implementation

[0056] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the technical solutions of the present invention, but should not be used to limit the scope of protection of the present invention.

[0057] Implementation, for example, attached Figure 1 , Figure 2 As shown, this embodiment of the invention provides a flexible support method for photovoltaic clusters based on power line carrier, applicable to power line carrier, and includes the following steps:

[0058] Step S1: Obtain the distribution network vulnerability heat map and the distribution network stability heat map, and combine them with the distribution network early warning and perception mechanism to perform distribution network fault perception and early warning, and generate fault early warning and perception results.

[0059] In this embodiment, step S1 includes:

[0060] Step S11: Obtain the heat map of distribution network vulnerability and the heat map of distribution network stability.

[0061] In this embodiment, step S11 includes:

[0062] Step S11-1: Generate a heat map of the vulnerability of the power distribution network.

[0063] Specifically, the network topology diagram of the current distribution network is obtained, and the intermediary centrality of each structure in the network topology diagram is analyzed. The larger the intermediary centrality value, the higher the vulnerability of the corresponding structure. A distribution network vulnerability heat map is generated based on the intermediary centrality of each structure.

[0064] In one possible embodiment, real-time power flow data of each node in the distribution network is collected. The 3σ criterion is used to remove data points that deviate from the mean by three times the standard deviation, and linear interpolation is used to fill in the missing data to generate a real-time power flow dataset. The real-time power flow dataset includes at least active power, reactive power, voltage amplitude, and voltage phase angle. Feature extraction is performed on the network topology diagram of the current distribution network to obtain a power grid topology dataset. The power grid topology dataset includes at least node numbers, such as Bus1 to Bus50, line connection relationships, such as line Line1 connecting two nodes, Bus1 and Bus2, and line parameters, such as resistance and reactance, and transformer turns ratio.

[0065] Based on power flow datasets and power grid topology datasets, power transmission paths in the network topology diagram are defined. The proportion of power transmission paths passing through a node within a fixed time interval is used as the betweenness centrality value of that node. Specifically, for any two nodes i and j, if the active power of i is transmitted to j through line L, then L is the power transmission path between node i and node j. For each node or line, the proportion of the number of power transmission paths passing through that node or line within 10 minutes is counted to the total number of power transmission paths. For example, if there are 1000 power transmission paths in the distribution network topology diagram, and 300 of them pass through line Line5 within 10 minutes, then the betweenness centrality value of Line5 is 0.3.

[0066] Based on the network topology diagram of the distribution network, the betweenness centrality values ​​of the nodes and lines in the network topology diagram are mapped to different colors in the heat map according to a predetermined rule to generate a distribution network vulnerability heat map. For example, nodes or lines with betweenness centrality values ​​in [0.8, 1.0] are mapped to dark red, indicating that the current node or line has extremely high vulnerability; nodes or lines with betweenness centrality values ​​in [0.5, 0.8) are mapped to orange, indicating that the current node or line has high vulnerability; [0.2, 0.5) are mapped to yellow, indicating that the current node or line has medium vulnerability; and (0, 0.2) are mapped to green, indicating that the current node or line has low vulnerability.

[0067] Step S11-2: Generate a heat map of distribution network stability.

[0068] Specifically, micro-disturbance pulses with amplitudes not exceeding the rated power and frequencies within the rated frequency of the power grid are randomly injected into the distribution network. Voltage and frequency response data of all nodes in the distribution network are collected and analyzed using the recursive least squares method to construct a distribution network stability heatmap that describes the dynamic impedance and stability margin of the distribution network. At the same time, micro-disturbance pulses are periodically injected into the distribution network to achieve dynamic updates of the distribution network stability heatmap.

[0069] In one possible embodiment, a small number of inverters are randomly selected periodically from all photovoltaic inverters in the distribution network as disturbance injection sources. For example, active power pulses are injected into the distribution network. The amplitude of the active power pulses is less than 0.5% of the rated power of the inverter. That is, assuming that the rated power of an inverter is 500kW, the disturbance amplitude corresponding to the active power pulse generated by the inverter cannot exceed 2.5kW. At the same time, the duration of the injected disturbance is kept to be 1 to 2 power frequency cycles, and the time of disturbance injection is set not to be during the peak load period of the current distribution network. In addition, the selected photovoltaic inverters are distributed in different feeder areas of the distribution network to avoid the disturbance being too concentrated.

[0070] Simultaneously with the disturbance injection, all photovoltaic inverters deployed at each node of the distribution network collect the instantaneous waveform of the voltage at the corresponding grid connection point and simultaneously collect the corresponding communication channel data to generate the original wide-area response dataset. The collected original wide-area response dataset is then time-stamped and filtered preprocessed to generate a new wide-area response dataset. The least squares method is then used for analysis to obtain the dynamic impedance values ​​between each node of the distribution network. Based on the dynamic impedance values, a heat map of distribution network stability is generated.

[0071] Understandably, the distribution network stability heatmap is used to quantify the stability margin of the distribution network. Specifically, dynamic impedance values ​​at specific frequencies are selected from the distribution network stability heatmap, for example, impedance values ​​within the range of [45, 65] Hz and [95, 105] Hz. The output impedance of the photovoltaic inverter at the specific frequency is matched with the output impedance of the photovoltaic inverter at the pre-established photovoltaic inverter output impedance database. The dynamic impedance value and the output impedance are analyzed and processed using the Nyquist stability criterion to generate the corresponding phase margin and gain margin. Based on the phase margin and gain margin, a normalized number is generated to quantify the stability of the distribution network. For example, using a threshold-based normalization method, assuming the lower limit of the phase margin is set to 30 degrees and the upper limit to 120 degrees, and the lower limit of the gain margin is set to 3 dB and the upper limit to 20 dB, if the phase margin does not exceed 30 degrees, its normalization exponent is 0; if it is not lower than 120 degrees, the exponent is 1; values ​​between the two are calculated according to a linear ratio, that is, if the phase margin is 39 degrees, the corresponding normalization value is 0.1. Similarly, the gain margin is normalized according to the above rules, and finally the normalized phase margin and gain margin are weighted and fused according to a predetermined weight ratio to generate a stable margin.

[0072] Step S12: Based on the distribution network early warning and perception mechanism, perform distribution network fault perception and early warning, and generate fault early warning and perception results.

[0073] Specifically, when the stability margin is lower than a set threshold, a distribution network fault detection early warning flag is triggered, and a fault early warning detection result is generated. The fault early warning detection result includes at least the suspected fault point and the corresponding betweenness centrality and stability margin.

[0074] In one possible embodiment, the stability margin data of the distribution network stability heatmap is periodically read and compared with a preset warning threshold to determine whether a warning is triggered. Specifically, the interval where the stability margin value is below 0.6 is defined as the high-level warning interval, and a high-level warning is triggered when the stability margin data is in the high-level warning interval; the interval where the stability margin value is in [0.6, 0.8) is defined as the medium-level warning interval, and a medium-level warning is triggered when the stability margin data is in the medium-level warning interval; the interval where the stability margin value is in [0.8, 0.9) is defined as the low-level warning interval, and a low-level warning is triggered when the stability margin data is in the low-level warning interval; when the stability margin value exceeds 0.9, no warning is triggered.

[0075] If a node in the distribution network stability heatmap triggers a high-level or medium-level warning, and that node is dark red or orange in the distribution network vulnerability heatmap, the warning level is upgraded by one level. For example, if a node has a stability margin of 0.7, it first triggers a medium-level warning. At the same time, if the node's intermediation centrality value in the distribution network vulnerability heatmap is found to be 0.68, which is located in the orange area, the medium-level warning is upgraded to a high-level warning. For nodes that trigger a high-level warning, their corresponding node number, intermediation centrality, and stability margin are obtained and output as fault warning perception results.

[0076] Step S2: Obtain multi-source data of the distribution network, generate a distribution network status dataset, and perform fault diagnosis based on the fault early warning perception results and the distribution network status dataset using a spatiotemporal graph convolutional network.

[0077] In this embodiment, step S2 includes:

[0078] Step S21: Obtain multi-source data of the distribution network and generate a distribution network status dataset.

[0079] Specifically, when a fault occurs in the distribution network, data is collected from the distribution network to obtain transient electrical quantity data and communication channel status data, generating a distribution network status dataset. The transient electrical quantity data includes at least the three-phase voltage phasors and three-phase current phasors of each node in the distribution network, and the communication channel status data includes at least the received signal strength indication, signal-to-noise ratio, channel delay, and bit error rate. The transient electrical quantity data is analyzed and processed based on independent component analysis and fault early warning perception results to generate a fault source contribution matrix. The fault source contribution matrix is ​​used to quantify the percentage contribution of each fault source to the drop in transient electrical quantity data at different nodes in the distribution network.

[0080] In one possible embodiment, transient electrical quantity data and communication channel status data are acquired when a fault occurs in the distribution network. The acquired data undergoes spatiotemporal alignment, noise filtering, and data standardization. Data standardization can be achieved using max-min normalization to generate a distribution network status dataset. Independent component analysis (ICA) is used to analyze the fault sources in the transient electrical quantity data within the distribution network status dataset, generating a fault source contribution matrix. Specifically, using the three-phase voltage phasors within the transient electrical quantity data as input, ICA is used to separate fault sources, generating a set of potential fault sources. The authenticity of the potential fault source set is verified by combining the corresponding three-phase current phasors, generating a fault source distribution matrix. The contribution matrix is ​​obtained by inverse matrix calculation of the fault source distribution matrix. Based on the fault source distribution matrix and the contribution matrix, a fault source contribution matrix is ​​constructed. The elements in the fault source contribution matrix describe the impact of the fault source on each node of the distribution network. For example, a ij The element in the i-th row and j-th column of the fault source contribution matrix is ​​used to describe the influence coefficient of fault source j on the voltage observation of node i, and a ij The larger the absolute value, the greater the impact of fault source j on node i. Since the element values ​​in the matrix are all normalized values, it can also express the proportion of responsibility that each fault source should bear for the voltage drop of a certain node, or the responsibility weight of node i to fault source j.

[0081] It should be noted that, assuming a potential fault source F1 is isolated from the distribution network through analysis of three-phase voltage phasors, the existence of this potential fault source is only determined from a mathematical analysis perspective. It is necessary to verify whether the potential fault source F1 exists in the physical environment. If the existence of the potential fault source F1 in the physical environment is verified, then the existence of the fault source can be confirmed. The verification of the physical environment can be achieved by analyzing the three-phase current phasors. For example, based on the network topology diagram of the distribution network, the line section with the closest electrical distance to the potential fault source F1 is located, and the transient waveform data of the three-phase current during the fault period of this line section is extracted. By calculating the Pearson correlation coefficient between the transient waveform data of the three-phase current and the potential fault source F1 in the time domain or frequency domain, and checking whether it exceeds a preset significance threshold, if it exceeds the preset significance threshold and the corresponding waveform shape is consistent, then it is proven that the potential fault source F1 has physical authenticity, thereby realizing the authenticity verification of the potential fault source.

[0082] Step S22: Fault diagnosis is performed based on spatiotemporal graph convolutional networks.

[0083] Specifically, the network topology diagram of the distribution network is obtained, and the network topology diagram, the distribution network status dataset, and the fault source contribution matrix are input into the spatiotemporal graph convolutional network for fault diagnosis to generate fault diagnosis results. The fault diagnosis results include at least the fault location, fault type, fault severity, predicted impact range, and fault source contribution matrix.

[0084] In one possible embodiment, each bus or grid connection point in the distribution network is defined as a graph node in a spatiotemporal graph structure based on the network topology diagram. Edges between nodes are defined according to the network topology connections, such as lines and transformers. The edge weights are determined by the electrical parameters of the lines, for example, using the reciprocal of the corresponding impedance value as the edge weight. Based on the distribution network state dataset and the fault source contribution matrix, corresponding attribute values ​​are injected into each graph node. These attribute values ​​include at least electrical quantity time-series features, responsibility weight features, and auxiliary features. The electrical quantity time-series features are the voltage and current transient waveform sequences of the node within a fixed time window before and after a fault. The responsibility weight features are... The percentage contribution of a node to different fault sources is extracted from the fault source contribution matrix. The auxiliary features include at least the node's reference voltage data and the corresponding power generation equipment types of adjacent nodes with edge connections. A spatiotemporal graph convolutional network is constructed, and historical fault cases and their corresponding data are input into the spatiotemporal graph convolutional network for pre-training. This allows the spatiotemporal graph convolutional network to learn local spatiotemporal patterns and fault interaction features. The local spatiotemporal patterns represent how faults propagate through lines in the distribution network. By learning the local spatiotemporal patterns, the fault source localization capability of the spatiotemporal graph convolutional network is trained. The fault interaction features include the type and severity of the fault.

[0085] The network topology diagram, distribution network status dataset, and fault source contribution matrix are input into a spatiotemporal graph convolutional network for fault diagnosis. The fault location, fault type, fault severity, and impact range are output and encapsulated as fault diagnosis results. For example, the Softmax activation function is used to output the fault probability of each node, and the node with the highest probability is the fault location. The fault types include at least five types: single-phase grounding, two-phase short circuit, three-phase short circuit, two-phase grounding short circuit, and voltage drop. The Sigmoid activation function is used to output a normalized value, and the larger the value, the more severe the fault. The Sigmoid activation function is also used to obtain the probability of each node being affected by the fault source.

[0086] Step S3: Obtain the fault diagnosis results, analyze and process the fault diagnosis results based on Byzantine fault-tolerant consensus and VCG auction mechanism, and generate an initial fault buffer scheme.

[0087] Specifically, the fault diagnosis results are analyzed and processed based on the Byzantine fault-tolerant consensus method and fuzzy reasoning to generate a fault buffer auction list. The fault buffer auction list includes at least a fault buffer target list, buffer boundary constraints, and fault transfer prediction. The photovoltaic cluster is auctioned based on the VCG auction mechanism and the fault buffer auction list to generate the optimal winning combination. The optimal winning combination is then negotiated and bargained based on cooperative game theory to generate an initial fault buffer scheme.

[0088] In this embodiment, step S3 includes:

[0089] Step S31: Analyze and process the fault diagnosis results based on the Byzantine fault-tolerant consensus method and fuzzy reasoning to generate a fault buffer auction list.

[0090] Specifically, each photovoltaic node in the photovoltaic cluster acquires local electrical transient data and predicts the electrical transient data based on a pre-trained lightweight LSTM network to generate predicted electrical transient data. Based on the fault source contribution matrix included in the fault diagnosis results, basic responsibility weights are generated for each photovoltaic node, which are used to quantify the response priority of the photovoltaic node to the fault source. The electrical transient data is fuzzified and defuzzified to generate a severity factor. The predicted electrical transient data is processed according to a preset urgency function to generate a predicted urgency factor. Based on a preset adaptive weighting mechanism, the basic responsibility weights, severity factor, and predicted urgency factor are adaptively weighted and fused to generate dynamic responsibility weights. Combined with the Byzantine fault-tolerant consensus method, a fault buffer consensus is reached, generating a fault buffer auction list.

[0091] In this embodiment, step S31 includes:

[0092] Step S31-1: Generate dynamic responsibility weights.

[0093] In one possible implementation, a basic responsibility weight is generated based on the values ​​of each element in the fault source contribution matrix. For example, suppose an element a in the fault source contribution matrix... ij =0.8, which indicates that the voltage drop at node i is affected by fault source j. That is, the voltage drop at node i is caused by a fault in the distribution network, and fault source j is responsible for 80% of the responsibility. Similarly, since node i is most affected by fault source j, node i also has the greatest impact on fault source j. Therefore, the basic responsibility weight of node i can be set to 0.8 to represent the basic degree of influence of node i on fault source j.

[0094] Based on transient electrical quantity data, the corresponding per-unit values ​​are calculated. These per-unit values ​​are then converted into corresponding severity factors through fuzzification and defuzzification operations. For example, taking voltage sag as an example, the per-unit value of the voltage sag before and after a fault is 0.7. Assuming the predefined fuzzification rule defines voltage sag as having three fuzziness levels: slight, moderate, and severe, and setting the fuzziness level to slight, the output level is low, with corresponding input and output peak values ​​of 0.02 and 0.2, respectively. If the fuzziness level is moderate, the output level is medium, with corresponding input and output peak values ​​of 0.07, respectively. 5 and 0.5; if the fuzziness level is severe, the output level is high, and the corresponding input and output peak values ​​are 0.15 and 0.8, respectively; set the membership function as a trigonometric function, assuming the per-unit value of the voltage drop is 0.7, activate the medium and severe fuzziness levels respectively, and the corresponding membership degrees are 0.8 and 0.2, respectively. Use the centroid method for defuzzification, and generate the corresponding severity factor value of (0.8*0.5+0.2*0.8) / (0.8+0.2)=0.56, where 0.5 and 0.8 are the output peak values ​​corresponding to the medium and severe fuzziness levels, respectively.

[0095] Each photovoltaic node in the photovoltaic cluster acquires local electrical transient data and performs electrical transient data prediction based on a pre-trained lightweight LSTM network, generating predicted electrical transient data. This predicted electrical transient data includes at least the predicted minimum voltage value, the critical time to reach the minimum value, and the prediction confidence level. The prediction confidence level can be quantitatively obtained by analyzing the smoothness of the generated prediction sequence. The predicted electrical transient data is then processed using a preset urgency function, which can be expressed as follows: ,in To predict the urgency factor, Let i be the prediction confidence level for node i. The time decay factor, To determine the critical time for node i to reach the predicted minimum voltage value, a corresponding prediction urgency factor is generated. For example, if the predicted minimum voltage value is 0.82 pu, the critical time to reach the minimum value is 120 ms, and the prediction confidence level is 0.95, then the corresponding prediction urgency factor value is approximately 0.95 * 0.091 ≈ 0.086. Assuming that the weight allocation generated by the preset adaptive weight allocation mechanism is based on the ratio of basic responsibility weight: severity factor: prediction urgency factor = 3:2:5, then the corresponding dynamic responsibility weight is 0.8 * 0.3 + 0.56 * 0.2 + 0.086 * 0.5 = 0.395.

[0096] Step S31-2: Generate a fault buffer auction list.

[0097] In one possible embodiment, an improved weighted practical Byzantine fault-tolerant consensus mechanism is used to generate the fault buffer auction list. Specifically, the node with the highest dynamic responsibility weight in the current distribution network is designated as the master node, and the remaining nodes are defined as slave nodes. The master node actively collects key state information of the global distribution network. The key state information includes at least the transient electrical quantity data, predicted transient electrical quantity data, and the dynamic responsibility weight corresponding to each node. Based on the key state information, a consensus proposal is generated. For example, a list of key nodes that need to be prioritized and their target voltage recovery values ​​are specified. The master node generates a corresponding sequence number, view number, and digital signature for the consensus proposal. The sequence number, view number, and digital signature are used by the slave nodes to verify the legitimacy. The consensus proposal and the corresponding sequence number, view number, and digital signature are encapsulated into a pre-prepared message and broadcast to all slave nodes.

[0098] After receiving the pre-preparation message broadcast by the master node, each slave node verifies the validity of the pre-preparation message. Specifically, the slave node determines the validity of the sequence number, view number, and digital signature of the pre-preparation message, and determines whether the data of the consensus proposal conforms to the preset power grid physical constraints, such as whether the voltage target value is within the safe range and whether the proposal generation logic is consistent with the current fault situation. When the pre-preparation message passes the validity verification, it is determined to be a valid pre-preparation message, and the slave node attaches its own dynamic responsibility weight to the valid pre-preparation message for broadcast.

[0099] Each node continuously receives legitimate preparation messages and accumulates their weights. Only when the accumulated weight exceeds two-thirds of the total dynamic responsibility weight of the whole, each node generates a commit message, broadcasts the commit message, and performs weighted accumulation and threshold judgment. When the preset weight threshold is reached, consensus is determined to be reached, and a fault buffer auction list is generated based on the consensus proposal that has reached consensus.

[0100] For example, suppose there are four photovoltaic nodes in the distribution network, Node1, Node2, Node3, and Node4. To reach a consensus and identify the fault source j as the most critical fault point, the dynamic responsibility weights corresponding to the four photovoltaic nodes are obtained as W1=0.4, W2=0.3, W3=0.2, and W4=0.1, respectively. The total global dynamic responsibility weight is 1.0, and the consensus weight threshold is set to 2 / 3 of the total weight, which means a weight sum of approximately 0.67 is required.

[0101] The consensus process begins in the preparation phase. After the master node broadcasts its pre-preparation message, each node verifies it and then begins broadcasting its own preparation message. Assuming that Node1, Node2, and Node3 are communicating without failure, while Node4 is experiencing a communication failure, preventing it from broadcasting messages, each node monitors and records preparation messages from other nodes after broadcasting its own. Taking Node2 as an example, Node2 first verifies the legitimacy of each received pre-preparation message. After successful verification, Node2 begins to accumulate the weights of legitimate preparation messages. That is, after receiving legitimate preparation messages from Node1 and Node3, the total weight of the legitimate preparation messages collected by Node2 is its own weight + Node1's weight + Node3's weight = 0.9. Since 0.9 exceeds the preset weight threshold of 0.67, Node2 meets the commit condition and can enter the commit phase.

[0102] During the commit phase, Node2 broadcasts a commit message containing its weight of 0.3. Similarly, Node1 and Node3 perform the same process, receiving valid preparation messages broadcast by each node and calculating that the total weight exceeds 0.67. At this point, Node1, Node2, and Node3 all enter the commit phase and broadcast commit messages. Node1, Node2, and Node3 then begin collecting commit messages. Node2 receives valid commit messages from Node1 and Node3 and adds its own weight again. Its own weight + Node1's weight + Node3's weight = 0.9, which again exceeds the predetermined weight threshold of 0.67. Therefore, Node2 confirms consensus locally. If Node1 and Node3 also undergo the same weighting and accumulation process and independently confirm consensus, then Node1, Node2, Node3, and Node4 ultimately reach a consensus that "fault source j is the most critical fault point."

[0103] Step S32: Generate an initial fault buffer scheme based on the VCG auction mechanism and the fault buffer auction list.

[0104] Specifically, based on the fault buffer auction list, the buffer requirements corresponding to the fault diagnosis results are decomposed into composable buffer task packages. Each photovoltaic node responds to the combination of buffer task packages based on its local actual cost data, where the actual cost data includes at least power generation cost data, equipment operation and maintenance cost data, and grid security risk cost data. According to a preset objective function, with the goal of minimizing the total actual cost data, the photovoltaic nodes responding to the bids are combined and screened in combination with preset constraint rules to generate the optimal winning combination with the minimum total cost. The preset constraint rules include at least the constraint of accurate allocation of buffer task packages, line power flow compliance constraint, node voltage safety constraint, and output power capacity constraint. Based on the Nash bargaining model, the bid prices of each winning node in the optimal winning combination are negotiated and negotiated to generate an initial fault buffer scheme.

[0105] In one possible implementation, after the Byzantine fault-tolerant consensus is reached and the fault buffer auction list is output, VCG combined auction and secondary screening based on Nash bargaining are executed sequentially to generate an initial fault buffer scheme. Specifically, in the VCG auction stage, the total fault buffer demand confirmed in the fault buffer auction list is first decomposed into multiple task packages that allow combined bidding. Each photovoltaic node, as a rational bidder, bids for individual or combined task packages based on its local marginal cost function. At this time, the auctioneer solves a mixed integer programming problem with preset constraints, and the objective is to minimize the social total while satisfying line capacity and node voltage safety. The project uses a pre-defined objective function to generate the optimal winning bid combination that minimizes the total social cost. The total social cost is represented by the sum of the marginal costs declared by all winning photovoltaic nodes when performing their assigned fault buffer tasks. It measures the total economic cost incurred by the entire photovoltaic cluster in completing a specific reactive power support task. Based on the optimal winning bid combination, a second screening of Nash bargaining is initiated. All winning photovoltaic nodes, as bargaining participants, propose to fine-tune the amount of tasks assigned to them while keeping the total fault buffer demand unchanged. The Pareto solution of the initial scheme corresponding to the optimal winning bid combination is solved by maximizing the Nash product, thereby generating the initial fault buffer scheme.

[0106] For example, suppose the optimal winning combination selected after a VCG auction is for PV nodes A, B, and C to provide 30 MVar, 50 MVar, and 20 MVar of reactive power output, respectively. However, it is found that PV node C's load factor has reached 85%, nearing its limit, and PV node B has high communication link latency. Therefore, the initial plan needs optimization. Nash bargaining is initiated, with PV nodes A, B, and C acting as negotiators. PV node B proposes first, stating that due to its poor communication link quality, it hopes to slightly adjust its output from 50 MVar to 48 MVar, which will slightly reduce its economic profit but significantly improve communication reliability. PV node A then responds, stating that its inverter load is lighter and its communication is good, and it is willing to provide an additional 2 MVar of reactive power output. The output power was adjusted from 30 MVar to 32 MVar. Although this change slightly increased its losses, it significantly improved its utility in terms of load balancing. Photovoltaic node C maintained its output power at 20 MVar. Each photovoltaic node evaluated the fine-tuned scheme based on a multi-attribute utility function. The multi-attribute utility function quantified the overall satisfaction of economic profit, equipment load balancing, and communication reliability. The Nash product of this fine-tuned scheme was calculated based on the multi-attribute utility function. It was found that the Nash product of the new scheme was higher than that of the initial scheme, indicating that this adjustment achieved Pareto improvement in overall satisfaction without harming the interests of any party. Through continuous fine-tuning iterations until the Nash product converged, the corresponding fine-tuned scheme was output as the initial fault buffer scheme.

[0107] Understandably, a multi-attribute utility function can be represented as: ,in For Nash product, For economic profit, For equipment load balancing, For communication reliability, These are the weighting coefficients for economic profit, equipment load balancing, and communication reliability, respectively. Economic profit can be calculated as (VCG winning bid price - actual cost) / preset maximum possible profit. Equipment load balancing can be calculated as 1 - the absolute value of the output adjustment required by the current candidate scheme relative to the initial VCG scheme / the maximum reactive power output of the node. Communication reliability can be calculated as real-time signal-to-noise ratio / maximum signal-to-noise ratio. For example, assuming the VCG winning bid price for photovoltaic node A is 30 MVar, the actual cost is 15 MVar, and after scheme fine-tuning, it becomes 32 MVar. Var, with the preset maximum possible profit set at 25MVar, maximum reactive power output at 50MVar, real-time signal-to-noise ratio at 18dB, and maximum signal-to-noise ratio at 30dB, and the weighting coefficients for economic profit, equipment load balancing, and communication reliability at 2:5:3, then the corresponding economic profit is 0.6, equipment load balancing is 0.96, and communication reliability is 0.6. At this point, the Nash product corresponding to photovoltaic node A after fine-tuning is 0.6*0.2+0.96*0.5+0.6*0.3=0.78 (this is just an example of the calculation process of the Nash product; the specific calculation should be combined with the actual situation).

[0108] Step S4: Based on the multi-agent proximal policy optimization algorithm and the bat-particle swarm optimization algorithm, the initial fault buffering scheme is jointly optimized to generate the optimal fault buffering scheme.

[0109] Specifically, taking the initial fault buffering scheme as input, a multi-agent near-end strategy optimization algorithm is used in a pre-constructed standard digital twin environment to generate the optimization adjustment direction of the initial fault buffering scheme. Combined with the bat-particle swarm optimization algorithm, distributed scheme tuning is performed to generate the optimal fault buffering scheme. The optimal fault buffering scheme is then distributed to the photovoltaic cluster. Real-time distribution network status datasets of the photovoltaic cluster executing the optimal fault buffering scheme are collected within a predetermined execution cycle. After a predetermined execution cycle ends, the real-time distribution network status datasets and the corresponding optimal fault buffering scheme are re-input into the standard digital twin environment to perform feedback optimization of the optimal fault buffering scheme, generating a new optimal fault buffering scheme and distributing it to the photovoltaic cluster for implementation. The feedback optimization operation is repeated until the distribution network status is stable and the optimization benefit is lower than a preset threshold or the distribution network fault is eliminated, at which point the loop exits.

[0110] Understandably, the scheme optimization model constructed based on the multi-agent near-end policy optimization algorithm includes a policy sub-network and a value sub-network. The policy sub-network is used to generate optimization strategies for schemes in the input model based on the real-time distribution network state dataset, and the value sub-network is used to evaluate the expected value of the scheme after implementing the optimization strategy.

[0111] Obtain the deviation between the actual effect and the expected result after implementing the optimal fault buffering scheme. Construct a real-time power grid state vector based on the deviation value and the real-time power distribution network state dataset. Input the real-time power grid state vector and the optimal fault buffering scheme into the scheme optimization model for scheme optimization.

[0112] The corresponding parameters of the strategy subnetwork and the value subnetwork are updated based on the real-time power grid state vector. The strategy subnetwork generates an optimization strategy. The optimal fault buffer scheme is tuned based on the optimization strategy using the bat-particle swarm optimization algorithm. During the tuning process, the value of the scheme is evaluated based on the value subnetwork to generate an optimal fault buffer scheme that is more in line with the current distribution network state.

[0113] In one possible embodiment, the scheme of the input scheme optimization model is mapped to particles based on the genetic algorithm. The position of a particle is an N-dimensional vector, where N is the number of photovoltaic nodes involved in the scheme of the input scheme optimization model. Each element in the vector corresponds to the reactive power output setting value of a photovoltaic node. For example, if the scheme of the input scheme optimization model involves three photovoltaic nodes A, B, and C, then the position of a particle can be represented as [Q1, Q2, Q3], where Q1, Q2, and Q3 represent the reactive power output values ​​of the three photovoltaic nodes A, B, and C, respectively. The entire particle swarm consists of multiple particles, and the population size is set to 50 to 100 particles. The initial position of the particle is set to the optimal fault buffer scheme or the initial fault buffer scheme of the previous control cycle. Each particle is assigned an initial velocity based on preset constraints, and the search space is strictly limited to the interval defined by the optimization strategy generated by the strategy sub-network. For example, if the optimization strategy defines the output interval of photovoltaic node A as [20kVar, 50kVar], then the position of all particles in the corresponding dimension of photovoltaic node A can only change within this range.

[0114] Then, an iterative optimization loop is entered. In each iteration, the standard particle swarm optimization mode is run first: each particle updates its flight speed and direction according to its own historical best position and the current global best position of the population, according to the standard velocity update formula. The inertia weight adopts a linear decreasing strategy, decreasing linearly from 0.9 to 0.3. After the particle position is updated, its fitness value is calculated according to the fitness function, which can be expressed as fitness F = -(m * sum of squares of voltage deviation + n * line overload penalty), where m and n are weight coefficients.

[0115] The algorithm continuously monitors the diversity of the population, for example, by calculating the average distance between particles or assessing whether the historical best solution has not been updated. When a preset trigger condition is met, the algorithm switches to the bat algorithm mode, treating each particle as a bat and adjusting its flight speed according to the pulse frequency, where the pulse frequency value is randomly generated within a preset range. The bats perform local random walks, generating new solutions around the current best solution. The generation and acceptance of new solutions follow the loudness and pulse emission rate mechanism of the bat algorithm. The above iterative process continues until a predetermined convergence condition is met and the iteration ends. If the optimal fitness value changes less than a preset threshold or the maximum number of iterations is reached in multiple consecutive iterations, the algorithm finally outputs the set of reactive power output instructions represented by the globally optimal particle. This set of reactive power output instructions is the latest optimal fault buffer scheme.

[0116] Step S5: Each photovoltaic node in the photovoltaic cluster executes the optimal fault buffering scheme and combines it with the adaptive virtual impedance algorithm to provide flexible support for the faulty distribution network.

[0117] Specifically, the optimal fault buffering scheme is decomposed to generate a buffer instruction set input to each photovoltaic node. Each photovoltaic node analyzes and processes the buffer instruction set based on an adaptive virtual impedance algorithm to generate virtual impedance data. The virtual impedance data is used to smoothly execute the buffer instruction set and includes at least virtual resistance and virtual inductance. A smoothing auxiliary instruction set is generated based on the virtual impedance data. Each photovoltaic node simultaneously executes the buffer instruction set and the corresponding smoothing auxiliary instruction set to provide flexible support for the faulty distribution network.

[0118] Understandably, when the local controller of the photovoltaic inverter receives the buffered instruction set, it does not directly send it to the PWM modulator, but instead inputs it to a pre-trained virtual impedance model. This virtual impedance model obtains the corresponding virtual resistance and virtual inductance components, which are dynamically adjusted by the instantaneous rate of change of the buffered instruction set. This instantaneous rate of change represents the change in the instruction value per unit time. For example, assuming a sampling period of 0.1 milliseconds, a current sampling point instruction value of 10 amps, and a previous sampling point instruction value of 2 amps, then the instantaneous rate of change is (10... A-2A) / 0.1ms=80A / ms. Specifically, when a large instantaneous rate of change of the command is detected, the current surge is suppressed by increasing the virtual resistance component, and the power change is smoothed by increasing the virtual inductance component. Conversely, when the instantaneous rate of change of the command is small, the virtual resistance component and the virtual inductance component are reduced. The buffer command set is convolved based on the virtual resistance component and the virtual inductance component to generate a voltage compensation amount. This compensation amount is then superimposed with a preset standard voltage to generate a voltage modulation signal that acts on the PWM modulator, thereby achieving flexible support for the faulty distribution network.

[0119] This invention provides a flexible support system for photovoltaic clusters based on power line carrier communication, applicable to power line carrier communication, comprising:

[0120] The data acquisition module is used to acquire heat maps of distribution network vulnerability and distribution network stability, collect multi-source data of distribution network, and generate distribution network status dataset.

[0121] The fault early warning module is used to perform fault perception and early warning of the distribution network according to the distribution network early warning perception mechanism, and generate fault early warning perception results.

[0122] The fault diagnosis module is used to perform fault diagnosis based on the fault early warning perception results and the power distribution network status dataset using a spatiotemporal graph convolutional network, and generate fault diagnosis results.

[0123] The scheme generation module is used to analyze and process the fault diagnosis results based on the Byzantine fault tolerance consensus and VCG auction mechanism to generate an initial fault buffer scheme.

[0124] The scheme optimization module is used to collaboratively optimize the initial fault buffer scheme based on the multi-agent proximal policy optimization algorithm and the bat-particle swarm algorithm to generate the optimal fault buffer scheme.

[0125] A buffer execution module is used to provide flexible support for the faulty distribution network based on an adaptive virtual impedance algorithm and an optimal fault buffering scheme.

[0126] The specific usage and function of this embodiment are as follows:

[0127] The system acquires heatmaps of distribution network vulnerability and stability, and combines these with a distribution network early warning and perception mechanism to perform fault detection and early warning, generating fault early warning and perception results. It also acquires multi-source data from the distribution network to generate a distribution network status dataset. Based on a spatiotemporal graph convolutional network, it performs fault diagnosis on the fault early warning and perception results and the distribution network status dataset, obtaining fault diagnosis results. By performing distribution network fault detection and early warning, and combining this with a spatiotemporal graph convolutional network, the system clarifies the responsibilities of each node and the fault propagation path from a global perspective of the distribution network, providing a data foundation for subsequent steps.

[0128] The fault diagnosis results are analyzed and processed based on Byzantine fault-tolerant consensus and VCG auction mechanism to generate an initial fault buffer scheme. Through the weighted Byzantine fault-tolerant consensus mechanism, nodes can quickly exchange information and reach consensus locally. At the same time, in the weighted Byzantine fault-tolerant consensus mechanism, the dynamic voting weight of nodes is deeply integrated with their responsibility weight and real-time communication channel status, so that the decision-making power for generating the initial fault buffer scheme is tilted towards nodes with greater responsibility and better communication, thereby improving the reliability of the initial fault buffer scheme.

[0129] The initial fault buffering scheme is collaboratively optimized using a multi-agent proximal policy optimization algorithm and a bat-particle swarm optimization algorithm to generate the optimal fault buffering scheme. By combining the multi-agent proximal policy optimization algorithm and the bat-particle swarm optimization algorithm to optimize the initial fault buffering scheme, it is ensured that the actions of each node serve the global optimal goal, thereby avoiding the risk of oscillation caused by local decision conflicts. At the same time, the optimal fault buffering scheme is continuously optimized in an iterative loop, so that the optimal fault buffering scheme always fits the current fault state, thereby improving the reliability of the optimal fault buffering scheme.

[0130] Each photovoltaic node in the photovoltaic cluster executes the optimal fault buffering scheme and combines it with an adaptive virtual impedance algorithm to provide flexible support for the faulty distribution network. The photovoltaic inverter directly executes the optimal instruction corresponding to the optimal fault buffering scheme through virtual impedance technology, thereby achieving flexible support for the faulty distribution network.

[0131] Furthermore, embodiments of the present invention also provide an electronic device, the electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the methods described above.

[0132] The following is a detailed introduction to the various components of the electronic device:

[0133] In this context, the processor is the control center of the electronic device. It can be a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs) or one or more field-programmable gate arrays (FPGAs).

[0134] The processor can perform various functions of an electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0135] The memory is used to store the software program that executes the solution of the present invention, and the execution is controlled by the processor. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0136] The memory can be a real-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only (CD-ROM), or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. The memory can be integrated with the processor or exist independently and coupled to the processor through an interface circuit of an electronic device; this embodiment of the invention does not specifically limit this.

[0137] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via limited means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0138] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0139] It should be understood that, in the embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0140] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A flexible support method for photovoltaic clusters based on power line carrier communication, characterized in that, The method includes: S1: Obtain the distribution network vulnerability heat map and distribution network stability heat map, and combine them with the distribution network early warning and perception mechanism to perform distribution network fault perception and early warning, and generate fault early warning and perception results; S2: Acquire multi-source data of the distribution network, generate a distribution network status dataset, and perform fault diagnosis based on the fault early warning perception results and the distribution network status dataset using a spatiotemporal graph convolutional network. S3: Obtain fault diagnosis results, analyze and process the fault diagnosis results based on Byzantine fault-tolerant consensus and VCG auction mechanism, and generate an initial fault buffer scheme; S4: The initial fault buffering scheme is jointly optimized based on the multi-agent proximal policy optimization algorithm and the bat-particle swarm algorithm to generate the optimal fault buffering scheme. S5: Each photovoltaic node in the photovoltaic cluster executes the optimal fault buffering scheme and combines it with the adaptive virtual impedance algorithm to provide flexible support for the faulty distribution network.

2. The flexible support method for photovoltaic clusters based on power line carrier as described in claim 1, characterized in that, Step S3 includes: The fault diagnosis results are analyzed and processed based on the Byzantine fault-tolerant consensus method and fuzzy reasoning to generate a fault buffer auction list, which includes at least a fault buffer target list, buffer boundary constraints and fault transfer prediction. Based on the VCG auction mechanism and the aforementioned fault buffer auction list, the photovoltaic cluster is auctioned to generate the optimal winning combination. Then, based on cooperative game theory, the optimal winning combination is negotiated and negotiated to generate an initial fault buffer scheme.

3. The flexible support method for photovoltaic clusters based on power line carrier according to claim 2, characterized in that, Based on the Byzantine fault-tolerant consensus method and fuzzy reasoning, the fault diagnosis results are analyzed and processed to generate a fault buffer auction list, including: Each photovoltaic node in the photovoltaic cluster acquires local electrical transient data and performs electrical transient data prediction based on a pre-trained lightweight LSTM network to generate predicted electrical transient data. Based on the fault source contribution matrix contained in the fault diagnosis results, a basic responsibility weight is generated for each photovoltaic node. The basic responsibility weight is used to quantify the response priority of the photovoltaic node to the fault source. The transient electrical quantity data is fuzzed and defuzzified to generate a severity factor. The transient electrical quantity data is then processed according to a preset urgency function to generate a prediction urgency factor. Based on a preset adaptive weighting mechanism, the basic responsibility weight, severity factor, and predicted urgency factor are adaptively weighted and fused to generate dynamic responsibility weights. Combined with the Byzantine fault-tolerant consensus method, a fault buffer consensus is reached to generate a fault buffer auction list.

4. The flexible support method for photovoltaic clusters based on power line carrier as described in claim 2, characterized in that, Step S3 further includes: Based on the fault buffer auction list, the buffer demand corresponding to the fault diagnosis results is decomposed into composable buffer task packages. Each photovoltaic node responds to the combination of buffer task packages based on local actual cost data, wherein the actual cost data includes at least power generation cost data, equipment operation and maintenance cost data, and grid security risk cost data. Based on a preset objective function with the goal of minimizing the total actual cost data, the photovoltaic nodes of the bidding response are combined and screened in combination with preset constraint rules to generate the optimal winning combination with the minimum total cost. The preset constraint rules include at least the constraints of precise allocation of buffer task packages, line power flow compliance, node voltage safety, and output power capacity. Based on the Nash bargaining model, the bid prices of each winning node in the optimal winning bid combination are negotiated and negotiated to generate an initial fault buffer scheme.

5. A flexible support method for photovoltaic clusters based on power line carrier communication according to claim 1, characterized in that, Step S4 includes: Taking the initial fault buffer scheme as input, the multi-agent proximal policy optimization algorithm is used in a pre-built standard digital twin environment to generate the optimization adjustment direction of the initial fault buffer scheme. Combined with the bat-particle swarm algorithm, the distributed scheme is fine-tuned to generate the optimal fault buffer scheme. The optimal fault buffering scheme is distributed to the photovoltaic cluster. Real-time distribution network status dataset of the photovoltaic cluster executing the optimal fault buffering scheme within a predetermined execution cycle is collected. After a predetermined execution cycle ends, the real-time distribution network status dataset and the corresponding optimal fault buffering scheme are re-input into the standard digital twin environment to perform feedback optimization of the optimal fault buffering scheme, generate a new optimal fault buffering scheme, and distribute it to the photovoltaic cluster for implementation. Repeat the feedback optimization operation until the distribution network status is stable and the optimization benefit is lower than the preset threshold or the distribution network fault is eliminated, then exit the loop.

6. A flexible support method for photovoltaic clusters based on power line carrier communication according to claim 5, characterized in that, Step S4 further includes: The scheme optimization model is constructed based on the multi-agent near-end policy optimization algorithm. The scheme optimization model includes a policy sub-network and a value sub-network. The policy sub-network is used to generate optimization policies for the schemes in the input model based on the real-time distribution network status dataset, and the value sub-network is used to evaluate the expected value of the schemes after implementing the optimization policies. Obtain the deviation between the actual effect and the expected result after implementing the optimal fault buffering scheme. Construct a real-time power grid state vector based on the deviation value and the real-time power distribution network state dataset. Input the real-time power grid state vector and the optimal fault buffering scheme into the scheme optimization model for scheme optimization. The corresponding parameters of the strategy subnetwork and the value subnetwork are updated based on the real-time power grid state vector. The strategy subnetwork generates an optimization strategy. The optimal fault buffer scheme is tuned based on the optimization strategy using the bat-particle swarm optimization algorithm. During the tuning process, the value of the scheme is evaluated based on the value subnetwork to generate an optimal fault buffer scheme that is more in line with the current distribution network state.

7. A flexible support method for photovoltaic clusters based on power line carrier communication according to claim 1, characterized in that, Step S5 includes: The optimal fault buffering scheme is decomposed to generate a buffer instruction set input to each photovoltaic node. Each photovoltaic node analyzes and processes the buffer instruction set based on an adaptive virtual impedance algorithm to generate virtual impedance data. The virtual impedance data is used to smoothly execute the buffer instruction set and includes at least virtual resistance and virtual inductance. Based on the virtual impedance data, a smoothing auxiliary instruction set is generated. Each photovoltaic node simultaneously executes the buffer instruction set and the corresponding smoothing auxiliary instruction set to provide flexible support for the faulty distribution network.

8. A flexible support method for photovoltaic clusters based on power line carrier communication according to claim 1, characterized in that, Step S1 includes: Obtain the current distribution network topology diagram and analyze the intermediary centrality of each structure in the network topology diagram. The larger the intermediary centrality value, the higher the vulnerability of the corresponding structure. Generate a distribution network vulnerability heat map based on the intermediary centrality of each structure. Periodically inject micro-disturbance pulses with amplitudes not exceeding the rated power and frequencies within the rated frequency of the power grid into the distribution network, and collect voltage and frequency response data of all nodes in the distribution network. Analyze the data using the recursive least squares method to construct a distribution network stability heatmap to describe the dynamic impedance and stability margin of the distribution network. When the stability margin is lower than a set threshold, trigger a distribution network fault detection early warning flag and generate a fault early warning detection result. The fault early warning detection result includes at least the suspected fault point and the corresponding intermediate centrality and stability margin.

9. A flexible support method for photovoltaic clusters based on power line carrier communication according to claim 1, characterized in that, Step S2 includes: When a fault occurs in the distribution network, data is collected from the distribution network to obtain transient electrical quantity data and communication channel status data, and a distribution network status dataset is generated. The transient electrical quantity data includes at least the three-phase voltage phasors and three-phase current phasors of each node in the distribution network, and the communication channel status data includes at least the received signal strength indication, signal-to-noise ratio, channel delay, and bit error rate. Based on the independent component analysis method and the fault early warning perception results, the transient data of electrical quantities are analyzed and processed to generate a fault source contribution matrix. The fault source contribution matrix is ​​used to quantify the percentage contribution of each fault source to the drop in transient data of electrical quantities at different nodes in the distribution network. Obtain the network topology diagram of the distribution network, input the network topology diagram, the distribution network status dataset, and the fault source contribution matrix into a spatiotemporal graph convolutional network for fault diagnosis, and generate fault diagnosis results. The fault diagnosis results include at least the fault location, fault type, fault severity, predicted impact range, and fault source contribution matrix.

10. A flexible support system for photovoltaic clusters based on power line carrier communication, used to implement the method described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire the distribution network vulnerability heat map and the distribution network stability heat map, collect multi-source data of the distribution network, and generate a distribution network status dataset. The fault early warning module is used to perform fault perception and early warning of the distribution network according to the distribution network early warning and perception mechanism, and generate fault early warning and perception results. The fault diagnosis module is used to perform fault diagnosis based on the fault early warning perception results and the power distribution network status dataset using the spatiotemporal graph convolutional network, and generate fault diagnosis results. The scheme generation module is used to analyze and process the fault diagnosis results based on the Byzantine fault tolerance consensus and VCG auction mechanism to generate an initial fault buffer scheme. The scheme optimization module is used to collaboratively optimize the initial fault buffer scheme based on the multi-agent proximal policy optimization algorithm and the bat-particle swarm algorithm to generate the optimal fault buffer scheme. A buffer execution module is used to provide flexible support for the faulty distribution network based on an adaptive virtual impedance algorithm and an optimal fault buffering scheme.

Images