A parallel inverter control method and system for adaptively adjusting active droop coefficient

Abstract

The application discloses a kind of parallel inverter control methods of adaptive adjustment active droop coefficient, comprising the following steps: step 1: system architecture design parallel inverter network: establish multiple parallel inverters, form distributed power generation system, each inverter is equipped with controller and communication module;Blockchain network establishment: create a private blockchain network, all inverters join blockchain as node, configure blockchain network to support smart contract and distributed ledger function;Step 2: data acquisition and processing internet of things sensor integration: deploy internet of things sensor on each inverter, real-time acquisition voltage, current, power, frequency operating data.The application can monitor the running state of system in real time, detect and isolate fault inverter in time, and adjust the operating parameters of other inverters, ensure that the system can still continue to run stably under fault condition, and quickly recover to the best operating state after fault.

Description

Technical Field

[0001] This invention relates to the technical field of inverter control, and more particularly to a parallel inverter control method and system for adaptively adjusting the active power droop coefficient. Background Technology

[0002] With the rapid development of renewable energy, distributed generation systems have become an important component of modern power systems. Parallel inverters are key power conversion devices in distributed generation systems; they are connected to the grid in parallel to share the power supply task. However, in practical applications, parallel inverter systems face many challenges, such as system stability, uneven load sharing, and self-healing capabilities under fault conditions.

[0003] While traditional droop control methods are widely used in parallel inverter control, their inherent limitations result in low system flexibility in response to load and operating environment changes, making them prone to voltage and frequency instability. Furthermore, traditional control methods often rely on centralized control, making the system highly sensitive to single-point failures and failing to meet the demands of distributed and decentralized control in modern power systems.

[0004] In recent years, with the rapid development of IoT, blockchain, and AI technologies, control methods based on these emerging technologies have provided new approaches to solving the aforementioned problems. For example, by integrating IoT sensors for real-time data acquisition, utilizing blockchain to achieve data transparency and decentralized control, and combining reinforcement learning algorithms to dynamically adjust control parameters, the stability, reliability, and adaptability of the system can be significantly improved. However, how to effectively integrate these emerging technologies and apply them to parallel inverter control systems remains a problem that urgently needs to be solved. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0006] In view of the problems existing in the current parallel inverter control methods and systems that adaptively adjust the active power droop coefficient, this invention is proposed.

[0007] Therefore, the purpose of this invention is to provide a parallel inverter control method and system for adaptively adjusting the active power droop coefficient, which can monitor the operating status of the system in real time, detect and isolate faulty inverters in a timely manner, and adjust the operating parameters of other inverters to ensure that the system can continue to operate stably under fault conditions and quickly recover to the optimal operating state after the fault.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a parallel inverter control method for adaptively adjusting the active power droop coefficient, comprising the following steps:

[0009] Step 1: System Architecture Design

[0010] Parallel inverter network: Establish multiple parallel inverters to form a distributed power generation system, with each inverter equipped with a controller and communication module;

[0011] Blockchain network establishment: Create a private blockchain network, with all inverters joining the blockchain as nodes, and configure the blockchain network to support smart contracts and distributed ledger functions;

[0012] Step 2: Data Acquisition and Processing

[0013] IoT sensor integration: Deploy IoT sensors on each inverter to collect real-time operating data such as voltage, current, power, and frequency. Use edge computing devices to process the sensor data and perform preliminary screening and noise filtering.

[0014] System performance feedback: Establish a data feedback mechanism to send processed data to the controller and the blockchain network;

[0015] Step 3: Reinforcement Learning Model Development

[0016] Model Construction: Construct a reinforcement learning model with the inverter's operating data as input and the adjusted active power droop coefficient as output.

[0017] Reward function design:

[0018] Design a reward function to incentivize the model to maintain voltage and frequency stability while achieving load balancing;

[0019] Model training and optimization:

[0020] The reinforcement learning model is trained using historical operational data and simulation data, and the model parameters are continuously optimized.

[0021] Regularly retrain the model to adapt to new environments and load conditions;

[0022] Step 4: Decentralized Control via Blockchain

[0023] Data on the blockchain:

[0024] The real-time operating data of the inverter is uploaded to the blockchain via an IoT gateway to ensure the transparency and immutability of the data;

[0025] Each inverter node records and broadcasts its own and other nodes' operating data;

[0026] Smart contract execution:

[0027] Deploy smart contracts to enable data sharing and control collaboration among inverters; the smart contracts dynamically adjust the active power droop coefficient of each inverter based on the operating data recorded in the blockchain network and the output of the reinforcement learning model.

[0028] Consensus Algorithm:

[0029] Consensus algorithms are used to ensure the consistency of data and control commands among inverter nodes; consensus algorithms also guarantee the decentralization and fault tolerance of the control system.

[0030] Step 5: Real-time control and adaptive adjustment

[0031] Control command issuance: Based on the output of the smart contract and reinforcement learning model, real-time control commands are generated and issued to each inverter through edge computing devices;

[0032] Adaptive adjustment: Each inverter adaptively adjusts the active power droop coefficient according to the received control commands to ensure stable system operation and load balancing; the inverter provides real-time feedback on the adjustment results and updates the operating data in the blockchain network;

[0033] Step 6: System Monitoring and Maintenance

[0034] Fault detection and isolation: Implement self-healing algorithms, monitor the system's operating status in real time, detect and isolate faulty inverters, and ensure the system's continuous operation and performance optimization under fault conditions.

[0035] As a preferred embodiment of the parallel inverter control method for adaptively adjusting the active power droop coefficient described in this invention, wherein the reinforcement learning model in step 3 includes:

[0036] Status: The current status of the system, including inverter operating data, voltage, current, frequency, and load;

[0037] Action: The system takes action to adjust the active power droop coefficient;

[0038] Reward: Feedback obtained after the system takes an action, used to measure the quality of the action;

[0039] Strategy: The strategy that determines the action a system takes in a given state;

[0040] Value function: evaluates the quality of a state or state-action pair;

[0041] Wherein, state is represented as:

[0042] State S t Represented as a vector, this includes key operating data of the inverter: St =[V t ,I t ,f t ,L t ], V t I represents voltage. t f represents current. t L represents frequency. t Indicates load;

[0043] Action space: Action A t It involves adjusting the value of the active power droop coefficient k: A t =Δk, where Δk is the increment or decrement of the active power droop coefficient;

[0044] Reward function: Reward function R t Designed as follows:

[0045] R t =-(w1·Voltage Deviation+w2·Frequency Deviation+w3·Load Imbalance+w4·Power Loss),

[0046] Where: w1, w2, w3, w4 are weighting coefficients, which are adjusted according to the actual application;

[0047] Voltage Deviation refers to the voltage deviation, defined as the difference between the target voltage and the actual voltage;

[0048] Frequency Deviation refers to the frequency deviation, defined as the difference between the target frequency and the actual frequency.

[0049] Load Imbalance indicates the degree of load imbalance;

[0050] Policy and value function: The DQN algorithm is used as the reinforcement learning algorithm. DQN approximates the value function S(s,a;θ) through a deep neural network, where θ is the parameter of the neural network.

[0051] As a preferred embodiment of the parallel inverter control method for adaptively adjusting the active power droop coefficient described in this invention, the DQN algorithm steps include:

[0052] S1. Initialize the experience replay pool D;

[0053] S2. Initialize the Q-network and the target Q-network with parameters θ and θ', respectively;

[0054] S3. For each round of training:

[0055] Get the current state S t ;

[0056] Choose action A according to the ∈-greedy strategy. t ;

[0057] Perform action A t Obtain the next state S t+1 and reward R t ;

[0058] (S) t A t ,R t ,S t+1 Stored in the experience replay pool D;

[0059] A batch of data (S) was randomly sampled from the experience replay pool. i A i ,R i ,S i+1 );

[0060] Calculate the target Q value: y i =R i +γmax a' Q(S i+1 ,a';θ');

[0061] Calculate the loss function:

[0062] Update the Q-network parameters θ using gradient descent.

[0063] The target Q network parameters θ' = θ are updated periodically.

[0064] As a preferred embodiment of the parallel inverter control method for adaptively adjusting the active power droop coefficient described in this invention, the reinforcement learning model training process uses historical operating data and simulation data for model training, continuously optimizing the neural network parameters θ. In actual operation, the current state S is input. t The optimal action A is obtained from the trained DQN model. t The inverter's control parameters are adjusted in real time.

[0065] As a preferred embodiment of the parallel inverter control method for adaptively adjusting the active power droop coefficient described in this invention, wherein: the consensus algorithm in step 4 adopts practical Byzantine fault tolerance, and the specific algorithm steps include:

[0066] A1. System Model: Assume there are n nodes, of which f nodes may be malicious, requiring n≥3f+1; Practical Byzantine fault tolerance includes three main phases: pre-preparation, preparation, and commit;

[0067] A2, Phase 1: Pre-preparation, the master node sends a pre-preparation message to all backup nodes. The pre-preparation message includes a proposal m to reach consensus and a sequence number s.

[0068] A3. Phase Two: Preparation. After receiving the pre-preparation message, the backup node verifies the correctness of the message and broadcasts the preparation message to all nodes. The preparation message includes the proposal m and the sequence number s.

[0069] A4. Phase 3: Commit. After receiving at least 2f preparation messages, the backup node broadcasts a commit message to all nodes. The commit message includes the proposal m and the sequence number s.

[0070] A5. Reaching Consensus: After all nodes have received at least 2f+1 commit messages, they consider proposal m to have reached consensus and execute the proposal. The message passing formula is as follows:

[0071] Message total =n·Message prepare +n·Message commit

[0072] Among them, Message prepare and Message commit These represent the number of prepared messages and the number of committed messages, respectively.

[0073] As a preferred embodiment of the parallel inverter control method for adaptively adjusting the active power droop coefficient described in this invention, the specific steps of the self-healing algorithm in step 6 are as follows:

[0074] B1. Real-time monitoring of system operation status:

[0075] Collect operating data from each inverter in the system, including voltage, current, frequency, and temperature, and use the data acquisition module to acquire and update the inverter's status information in real time;

[0076] Formula: Let S i (t) is the state vector of the i-th inverter at time t:

[0077] S i (t)=[V i (t),I i (t),f i (t),T i (t)]

[0078] Among them, V i (t) represents the voltage of the i-th inverter at time t; I i (t) represents the current of the i-th inverter at time t; f i (t) represents the frequency of the i-th inverter at time t; Ti (t) represents the temperature of the i-th inverter at time t;

[0079] B2. Fault Detection:

[0080] Define a fault detection threshold and use an anomaly detection algorithm to analyze the inverter's status data in real time to determine whether a fault has occurred.

[0081] Formula: Let the threshold be {V} min V max ,I min ,I max ,f min ,f max ,T max The fault detection conditions are:

[0082]

[0083] B3. Fault Isolation:

[0084] When an inverter fault is detected, immediately isolate the faulty inverter to prevent it from affecting the normal operation of the system; redistribute the load to ensure stable system operation; Formula: Let the faulty inverter be the i-th inverter, and its state be S. i (t), the isolation operation is: Isolate(S i (t)), the formula for redistributing the load is:

[0085]

[0086] Where: L new (t) represents the load of each inverter after redistribution;

[0087] L total (t) represents the total system load;

[0088] L i (t) represents the load of the faulty inverter;

[0089] n represents the total number of inverters in the system.

[0090] As a preferred embodiment of the parallel inverter control method for adaptively adjusting the active power droop coefficient described in this invention, the self-healing algorithm further includes self-healing control, which uses a prediction model and historical data to predict and analyze the system operating state, dynamically adjusts the inverter control parameters, and optimizes system performance.

[0091] The prediction model is: Predict(S) i The adjustment amount for self-healing control is ΔP(t), and the control formula is:

[0092] ΔP(t) = Predict(S(t)) - S(t)

[0093] Update the inverter control parameters based on the adjustment amount:

[0094] P new (t)=P(t)+ΔP(t)

[0095] Where: Predict(S(t)) is the predicted system state;

[0096] ΔP(t) represents the difference between the predicted state and the actual state;

[0097] P(t) represents the current control parameters of the inverter;

[0098] P new (t) represents the updated control parameters.

[0099] As a preferred embodiment of the parallel inverter control method for adaptively adjusting the active power droop coefficient described in this invention, the parallel inverter control system for adaptively adjusting the active power droop coefficient includes a parallel inverter system architecture, a data acquisition and processing subsystem, a reinforcement learning control subsystem, a blockchain decentralized control subsystem, a real-time control and adaptive adjustment subsystem, and a system monitoring and self-healing maintenance subsystem.

[0100] The parallel inverter system architecture includes inverter units and a blockchain network.

[0101] The data acquisition and processing subsystem includes an IoT sensor network and an edge computing module. The IoT sensor network is integrated on each inverter to monitor and acquire voltage, current, power, and frequency data in real time. The edge computing module is responsible for performing preliminary processing on the acquired data, reducing data noise, and providing preliminary system performance feedback. The feedback mechanism transmits the processed data to the controller and blockchain network to provide real-time data support for the reinforcement learning model.

[0102] The reinforcement learning control subsystem includes a reinforcement learning model module: an input port for receiving operating data from the inverter; an output port for providing an adjusted active power droop coefficient for real-time control of the inverter's operation; a reward function for incentivizing the model to maintain stable system voltage and frequency and achieve load balancing; and a model training module for training the model using historical and simulation data and periodically updating model parameters to adapt to new system conditions.

[0103] The blockchain decentralized control subsystem includes a data management module: data on-chain: inverter operating data is uploaded to the blockchain via an IoT gateway to ensure data transparency and immutability; smart contract execution module: smart contracts are executed based on on-chain data and the output of the reinforcement learning model to dynamically adjust the inverter's active power droop coefficient; consensus algorithm module: a consensus algorithm is used to ensure the consistency of data and control commands between inverter nodes.

[0104] The real-time control and adaptive adjustment subsystem includes a control command generation module and an adaptive adjustment module. The control command generation module generates and issues real-time control commands based on the output of the smart contract and the reinforcement learning model. The adaptive adjustment module includes an adjustment module: each inverter adaptively adjusts the active power droop coefficient according to the received control commands to ensure system stability and load balance; and a feedback module: the inverter feeds back the adjusted operating status to the blockchain network to achieve closed-loop control.

[0105] The system monitoring and self-healing maintenance subsystem includes a real-time monitoring module, a fault isolation module, and a maintenance module. The real-time monitoring module implements a self-healing algorithm to monitor and detect faulty inverters in the system in real time. The fault isolation module automatically isolates the faulty inverter when it is detected and adjusts the operation of other inverters to maintain system continuity and optimize performance. The maintenance module generates fault reports and shares them through a blockchain network to facilitate system maintenance and optimization.

[0106] As a preferred embodiment of the parallel inverter control method for adaptively adjusting the active power droop coefficient described in this invention, the inverter unit comprises multiple parallel inverters, each inverter being equipped with the following modules:

[0107] Controller: Responsible for executing control algorithms, adjusting active power droop coefficients, and executing received control commands;

[0108] Communication module: Supports communication with other inverters, ensuring data sharing and control collaboration, and supports blockchain network communication;

[0109] Internet of Things (IoT) sensors include voltage sensors, current sensors, and temperature sensors, used to collect operational data.

[0110] Edge computing devices: process sensor data, perform initial data screening and noise filtering, and transmit the processed data to controllers and blockchain networks.

[0111] As a preferred embodiment of the parallel inverter control method for adaptively adjusting the active power droop coefficient described in this invention, the blockchain network includes:

[0112] Private Blockchain: Establish a private blockchain, with all inverters joining the blockchain network as nodes;

[0113] Blockchain Nodes: Inverters and their controllers act as blockchain nodes, responsible for recording, broadcasting, and sharing operational data;

[0114] Smart contract module: Deployed on the blockchain network, used to perform data sharing and control collaboration.

[0115] The beneficial effects of this invention are:

[0116] 1. This invention constructs a reinforcement learning model to adaptively adjust the active power droop coefficient based on real-time operating data of the inverter (such as voltage, frequency, load, etc.). The reinforcement learning model, through the design of a reasonable reward function, incentivizes the system to achieve uniform load distribution while maintaining stable voltage and frequency, thereby significantly improving the stability and load sharing accuracy of the parallel inverter system.

[0117] 2. This invention implements a self-healing algorithm, which can monitor the operating status of the system in real time, detect and isolate faulty inverters in a timely manner, and adjust the operating parameters of other inverters to ensure that the system can continue to operate stably under fault conditions and quickly recover to the optimal operating state after the fault. Attached Figure Description

[0118] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0119] Figure 1 This is an overall schematic diagram of the parallel inverter control method for adaptively adjusting the active power droop coefficient according to the present invention.

[0120] Figure 2 This diagram illustrates a verification embodiment of the self-healing algorithm based on analog data for the parallel inverter control method of adaptively adjusting the active power droop coefficient according to the present invention.

[0121] Figure 3 This diagram illustrates a simulation experiment verification embodiment of the parallel inverter control method for adaptively adjusting the active power droop coefficient according to the present invention. Detailed Implementation

[0122] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0123] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0124] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0125] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth.

[0126] Reference Figure 1 This paper provides a control method for a parallel inverter that adaptively adjusts the active power droop coefficient, comprising the following steps:

[0127] Step 1: System Architecture Design

[0128] Parallel inverter network: Establish multiple parallel inverters to form a distributed power generation system, with each inverter equipped with a controller and communication module;

[0129] Blockchain network establishment: Create a private blockchain network, with all inverters joining the blockchain as nodes, and configure the blockchain network to support smart contracts and distributed ledger functions;

[0130] Step 2: Data Acquisition and Processing

[0131] IoT sensor integration: Deploy IoT sensors on each inverter to collect real-time operating data such as voltage, current, power, and frequency. Use edge computing devices to process the sensor data and perform preliminary screening and noise filtering.

[0132] System performance feedback: Establish a data feedback mechanism to send processed data to the controller and the blockchain network;

[0133] Step 3: Reinforcement Learning Model Development

[0134] Model Construction: Construct a reinforcement learning model with the inverter's operating data as input and the adjusted active power droop coefficient as output.

[0135] Reward function design:

[0136] Design a reward function to incentivize the model to maintain voltage and frequency stability while achieving load balancing;

[0137] Model training and optimization:

[0138] The reinforcement learning model is trained using historical operational data and simulation data, and the model parameters are continuously optimized.

[0139] Regularly retrain the model to adapt to new environments and load conditions;

[0140] Step 4: Decentralized Control via Blockchain

[0141] Data on the blockchain:

[0142] The real-time operating data of the inverter is uploaded to the blockchain via an IoT gateway to ensure the transparency and immutability of the data;

[0143] Each inverter node records and broadcasts its own and other nodes' operating data;

[0144] Smart contract execution:

[0145] Deploy smart contracts to enable data sharing and control collaboration among inverters; the smart contracts dynamically adjust the active power droop coefficient of each inverter based on the operating data recorded in the blockchain network and the output of the reinforcement learning model.

[0146] Consensus Algorithm:

[0147] Consensus algorithms are used to ensure the consistency of data and control commands among inverter nodes; consensus algorithms also guarantee the decentralization and fault tolerance of the control system.

[0148] Step 5: Real-time control and adaptive adjustment

[0149] Control command issuance: Based on the output of the smart contract and reinforcement learning model, real-time control commands are generated and issued to each inverter through edge computing devices;

[0150] Adaptive adjustment: Each inverter adaptively adjusts the active power droop coefficient according to the received control commands to ensure stable system operation and load balancing; the inverter provides real-time feedback on the adjustment results and updates the operating data in the blockchain network;

[0151] Step 6: System Monitoring and Maintenance

[0152] Fault detection and isolation: Implement self-healing algorithms, monitor the system's operating status in real time, detect and isolate faulty inverters, and ensure the system's continuous operation and performance optimization under fault conditions.

[0153] The reinforcement learning model in step 3 includes:

[0154] Status: The current status of the system, including inverter operating data, voltage, current, frequency, and load;

[0155] Action: The system takes action to adjust the active power droop coefficient;

[0156] Reward: Feedback obtained after the system takes an action, used to measure the quality of the action;

[0157] Strategy: The strategy that determines the action a system takes in a given state;

[0158] Value function: evaluates the quality of a state or state-action pair;

[0159] Wherein, state is represented as:

[0160] State S t Represented as a vector, this includes key operating data of the inverter: S t =[V t ,I t ,f t ,L t ], V t I represents voltage. t f represents current. t L represents frequency. t Indicates load;

[0161] Action space: Action A t It involves adjusting the value of the active power droop coefficient k: A t =Δk, where Δk is the increment or decrement of the active power droop coefficient;

[0162] Reward function: Reward function R t Designed as follows:

[0163] R t =-(w1·Voltage Deviation+w2·Frequency Deviation+w3·Load Imbalance+w4·Power Loss),

[0164] Where: w1, w2, w3, w4 are weighting coefficients, which are adjusted according to the actual application;

[0165] Voltage Deviation refers to the voltage deviation, defined as the difference between the target voltage and the actual voltage;

[0166] Frequency Deviation refers to the frequency deviation, defined as the difference between the target frequency and the actual frequency.

[0167] Load Imbalance indicates the degree of load imbalance;

[0168] Policy and value function: The DQN algorithm is used as the reinforcement learning algorithm. DQN approximates the value function S(s,a;θ) through a deep neural network, where θ is the parameter of the neural network.

[0169] Furthermore, the DQN algorithm steps include:

[0170] S1. Initialize the experience replay pool D;

[0171] S2. Initialize the Q-network and the target Q-network with parameters θ and θ', respectively;

[0172] S3. For each round of training:

[0173] Get the current state S t ;

[0174] Choose action A according to the ∈-greedy strategy. t ;

[0175] Perform action A t Obtain the next state S t+1 and reward R t ;

[0176] (S) t A t ,R t ,S t+1 Stored in the experience replay pool D;

[0177] A batch of data (S) was randomly sampled from the experience replay pool. i A i ,R i ,S i+1 );

[0178] Calculate the target Q value: y i =R i +γmax a' Q(S i+1 ,a';θ');

[0179] Calculate the loss function:

[0180] Update the Q-network parameters θ using gradient descent.

[0181] The target Q network parameters θ' = θ are updated periodically. Specifically, the reinforcement learning model training process uses historical running data and simulation data for model training, continuously optimizing the neural network parameters θ. In actual operation, the current state S is input. t The optimal action A is obtained from the trained DQN model. t The inverter's control parameters are adjusted in real time.

[0182] The consensus algorithm in step 4 employs Practical Byzantine Fault Tolerance, and the specific algorithm steps include:

[0183] A1. System Model: Assume there are n nodes, of which f nodes may be malicious, requiring n≥3f+1; Practical Byzantine fault tolerance includes three main phases: pre-preparation, preparation, and commit;

[0184] A2, Phase 1: Pre-preparation, the master node sends a pre-preparation message to all backup nodes. The pre-preparation message includes a proposal m to reach consensus and a sequence number s.

[0185] A3. Phase Two: Preparation. After receiving the pre-preparation message, the backup node verifies the correctness of the message and broadcasts the preparation message to all nodes. The preparation message includes the proposal m and the sequence number s.

[0186] A4. Phase 3: Commit. After receiving at least 2f preparation messages, the backup node broadcasts a commit message to all nodes. The commit message includes the proposal m and the sequence number s.

[0187] A5. Reaching Consensus: After all nodes have received at least 2f+1 commit messages, they consider proposal m to have reached consensus and execute the proposal. The message passing formula is as follows:

[0188] Message total =n·Message prepare +n·Message commit

[0189] Among them, Message prepare and Message commit These represent the number of prepared messages and the number of committed messages, respectively.

[0190] The specific steps of the self-healing algorithm in step 6 are as follows:

[0191] B1. Real-time monitoring of system operation status:

[0192] Collect operating data from each inverter in the system, including voltage, current, frequency, and temperature, and use the data acquisition module to acquire and update the inverter's status information in real time;

[0193] Formula: Let S i (t) is the state vector of the i-th inverter at time t:

[0194] S i (t)=[V i (t),I i (t),f i (t),T i (t)]

[0195] Among them, V i (t) represents the voltage of the i-th inverter at time t; I i (t) represents the current of the i-th inverter at time t; f i (t) represents the frequency of the i-th inverter at time t; T i (t) represents the temperature of the i-th inverter at time t;

[0196] B2. Fault Detection:

[0197] Define a fault detection threshold and use an anomaly detection algorithm to analyze the inverter's status data in real time to determine whether a fault has occurred.

[0198] Formula: Let the threshold be {V} min V max ,I min ,I max ,f min ,f max ,T max The fault detection conditions are:

[0199]

[0200] B3. Fault Isolation:

[0201] When an inverter fault is detected, immediately isolate the faulty inverter to prevent it from affecting the normal operation of the system; redistribute the load to ensure stable system operation; Formula: Let the faulty inverter be the i-th inverter, and its state be S. i (t), the isolation operation is: Isolate(S i (t)), the formula for redistributing the load is:

[0202]

[0203] Where: L new (t) represents the load of each inverter after redistribution;

[0204] L total (t) represents the total system load;

[0205] L i (t) represents the load of the faulty inverter;

[0206] n represents the total number of inverters in the system.

[0207] Furthermore, the self-healing algorithm also includes self-healing control, which uses predictive models and historical data to predict and analyze the system operating status, dynamically adjusts the inverter's control parameters, and optimizes system performance.

[0208] The prediction model is: Predict(S) i The adjustment amount for self-healing control is ΔP(t), and the control formula is:

[0209] ΔP(t) = Predict(S(t)) - S(t)

[0210] Update the inverter control parameters based on the adjustment amount:

[0211] P new (t)=P(t)+ΔP(t)

[0212] Where: Predict(S(t)) is the predicted system state;

[0213] ΔP(t) represents the difference between the predicted state and the actual state;

[0214] P(t) represents the current control parameters of the inverter;

[0215] P new (t) represents the updated control parameters.

[0216] Specific examples are as follows:

[0217] Assuming the system has three inverters connected in parallel (Node A, Node B, Node C), set the following thresholds:

[0218] Voltage threshold: V min =210V, V max =240V

[0219] Current threshold: I min =0A,I max =10A

[0220] Frequency threshold: f min =49.5Hz, f max =50.5Hz

[0221] Temperature threshold: T max =80℃

[0222] Self-healing process

[0223] 1. Real-time monitoring:

[0224] NodeA state: S A (t) = [230V, 8A, 50Hz, 75℃]

[0225] Node B state: S B (t)=[235V, 7A, 50.2Hz, 78℃]

[0226] Node C state: SC (t) = [220V, 9A, 50Hz, 82℃]

[0227] 2. Fault detection:

[0228] The temperature of Node C exceeds the threshold T max =80℃, fault detected;

[0229] 3. Fault isolation:

[0230] Isolate Node C and perform the operation: Isolate(S C (t))

[0231] Redistribute the load, assuming the total system load is L. total (t) = 24A, the load before the fault was:

[0232] NodeA: 8A; Node B: 7A; Node C: 9A;

[0233] After a failure, the load is redistributed:

[0234]

[0235] NodeA: 7.5A; Node B: 7.5A;

[0236] 4. Self-healing control:

[0237] Predict the system state using a predictive model: Predict(S(t)); dynamically adjust control parameters to optimize system performance;

[0238] Through the above steps, the self-healing algorithm can monitor the operating status of the parallel inverter system in real time, detect and isolate faulty inverters, ensure stable system operation and optimize performance.

[0239] Reference Figure 2 Example 1: Verification of self-healing algorithm based on simulated data

[0240] Simulated data generation

[0241] We generate simulation data for the parallel inverter system, including voltage, current, frequency, and temperature parameters, to generate simulation data and verify the effectiveness of the self-healing algorithm.

[0242] Reference Figure 3 Example 2: Simulation Experiment Verification

[0243] Simulation environment settings

[0244] In a simulation environment, the fault detection and isolation process of a three-parallel inverter system is simulated. A self-healing algorithm is used to isolate the faulty inverter and redistribute the load when a fault occurs.

[0245] Results and Analysis

[0246] 1. Voltage Curves: This section shows the voltage changes of the three inverters before and after the fault occurred. It can be seen that when the fault occurred (marked by the red line), the voltage curve of Node C did not change significantly, because voltage adjustment does not directly affect fault isolation operation.

[0247] 2. Current curve: When a fault occurs, the current of Node C drops rapidly to zero, while the current of Node A and Node B increases accordingly, redistributing the load to ensure the stability and continuous operation of the system.

[0248] 3. Frequency curve: This shows the frequency change of the inverter, indicating the stability of the system frequency during fault isolation.

[0249] 4. Temperature profile: Displays the temperature change of Node C. When a fault is detected and the temperature exceeds the threshold, Node C is isolated to ensure the safe operation of the system.

[0250] The effectiveness and reliability of the self-healing algorithm in parallel inverter systems can be verified through the analysis of the two embodiments and their results.

[0251] The adaptive adjustment of the active power droop coefficient parallel inverter control system includes a parallel inverter system architecture, a data acquisition and processing subsystem, a reinforcement learning control subsystem, a blockchain decentralized control subsystem, a real-time control and adaptive adjustment subsystem, and a system monitoring and self-healing maintenance subsystem.

[0252] Specifically, the parallel inverter system architecture includes inverter units and a blockchain network. The inverter unit includes multiple parallel inverters, and each inverter is equipped with the following modules:

[0253] Controller: Responsible for executing control algorithms, adjusting active power droop coefficients, and executing received control commands;

[0254] Communication module: Supports communication with other inverters, ensuring data sharing and control collaboration, and supports blockchain network communication;

[0255] Internet of Things (IoT) sensors include voltage sensors, current sensors, and temperature sensors, used to collect operational data.

[0256] Edge computing devices: process sensor data, perform preliminary data screening and noise filtering, and transmit the processed data to the controller and blockchain network;

[0257] The blockchain network includes:

[0258] Private Blockchain: Establish a private blockchain, with all inverters joining the blockchain network as nodes;

[0259] Blockchain Nodes: Inverters and their controllers act as blockchain nodes, responsible for recording, broadcasting, and sharing operational data;

[0260] Smart contract module: Deployed on the blockchain network, used to execute data sharing and control collaboration;

[0261] The data acquisition and processing subsystem includes an IoT sensor network and an edge computing module. The IoT sensor network is integrated on each inverter to monitor and acquire voltage, current, power, and frequency data in real time. The edge computing module is responsible for performing preliminary processing on the acquired data, reducing data noise, and providing preliminary system performance feedback. The feedback mechanism transmits the processed data to the controller and blockchain network to provide real-time data support for the reinforcement learning model.

[0262] The reinforcement learning control subsystem includes a reinforcement learning model module: an input port for receiving operating data from the inverter; an output port for providing an adjusted active power droop coefficient for real-time control of the inverter's operation; a reward function for incentivizing the model to maintain stable system voltage and frequency and achieve load balancing; and a model training module for training the model using historical and simulation data and periodically updating model parameters to adapt to new system conditions.

[0263] More specifically, the blockchain decentralized control subsystem includes a data management module: data on-chain: the inverter's operating data is uploaded to the blockchain via an IoT gateway to ensure data transparency and immutability; smart contract execution module: smart contracts are executed based on on-chain data and the output of the reinforcement learning model to dynamically adjust the inverter's active power droop coefficient; consensus algorithm module: a consensus algorithm is used to ensure the consistency of data and control commands between inverter nodes.

[0264] The real-time control and adaptive adjustment subsystem includes a control command generation module and an adaptive adjustment module. The control command generation module generates and issues real-time control commands based on the output of the smart contract and the reinforcement learning model. The adaptive adjustment module includes an adjustment module: each inverter adaptively adjusts the active power droop coefficient according to the received control commands to ensure system stability and load balance; and a feedback module: the inverter feeds back the adjusted operating status to the blockchain network to achieve closed-loop control.

[0265] The system monitoring and self-healing maintenance subsystem includes a real-time monitoring module, a fault isolation module, and a maintenance module. The real-time monitoring module implements a self-healing algorithm to monitor and detect faulty inverters in the system in real time. The fault isolation module automatically isolates the faulty inverter when it is detected and adjusts the operation of other inverters to maintain system continuity and optimize performance. The maintenance module generates fault reports and shares them through a blockchain network to facilitate system maintenance and optimization.

[0266] It should be noted that the above 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for controlling a parallel inverter with adaptive adjustment of active droop coefficient, characterized in that: Includes the following steps: Step 1: System Architecture Design Parallel inverter network: Establish multiple parallel inverters to form a distributed power generation system, with each inverter equipped with a controller and communication module; Blockchain network establishment: Create a private blockchain network, with all inverters joining the blockchain as nodes, and configure the blockchain network to support smart contracts and distributed ledger functions; Step 2: Data Acquisition and Processing IoT sensor integration: Deploy IoT sensors on each inverter to collect real-time operating data such as voltage, current, power, and frequency. Use edge computing devices to process the sensor data and perform preliminary screening and noise filtering. System performance feedback: Establish a data feedback mechanism to send processed data to the controller and the blockchain network; Step 3: Reinforcement Learning Model Development Model Construction: Construct a reinforcement learning model with the inverter's operating data as input and the adjusted active power droop coefficient as output. Reward function design: Design a reward function to incentivize the model to maintain voltage and frequency stability while achieving load balancing; Model training and optimization: The reinforcement learning model is trained using historical operational data and simulation data, and the model parameters are continuously optimized. Regularly retrain the model to adapt to new environments and load conditions; The reinforcement learning models include: Status: The current status of the system, including inverter operating data, voltage, current, frequency, and load; Action: The system takes action to adjust the active power droop coefficient; Reward: Feedback obtained after the system takes an action, used to measure the quality of the action; Strategy: The strategy that determines the action a system takes in a given state; Value function: evaluates the quality of a state or state-action pair; Wherein, state is represented as: State In vector representation, the key operating data of the inverter: , represents the voltage, represents the current, represents the frequency, represents the load; Action space: Action is the value of the active droop coefficient : where is the increment or decrement of the active droop coefficient; Reward function: reward function Designed as: ， wherein: , , , are weight coefficients, adjusted according to the practical application; represents the voltage deviation, defined as the difference between the target voltage and the actual voltage; represents the frequency deviation, defined as the difference between the target frequency and the actual frequency; denotes the degree of load imbalance; Policy and value function: The DQN algorithm is used as the reinforcement learning algorithm, which approximates the value function by a deep neural network where are the parameters of the neural network; Step 4: Decentralized Control via Blockchain Data on the blockchain: The real-time operating data of the inverter is uploaded to the blockchain via an IoT gateway to ensure the transparency and immutability of the data; Each inverter node records and broadcasts its own and other nodes' operating data; Smart contract execution: Deploy smart contracts to enable data sharing and control collaboration among inverters; the smart contracts dynamically adjust the active power droop coefficient of each inverter based on the operating data recorded in the blockchain network and the output of the reinforcement learning model. Consensus Algorithm: Consensus algorithms are used to ensure the consistency of data and control commands among inverter nodes; consensus algorithms also guarantee the decentralization and fault tolerance of the control system. Step 5: Real-time control and adaptive adjustment Control command issuance: Based on the output of the smart contract and reinforcement learning model, real-time control commands are generated and issued to each inverter through edge computing devices; Adaptive adjustment: Each inverter adaptively adjusts the active power droop coefficient according to the received control commands to ensure stable system operation and load balancing; the inverter provides real-time feedback on the adjustment results and updates the operating data in the blockchain network; Step 6: System Monitoring and Maintenance Fault detection and isolation: Implement self-healing algorithms, monitor the system's operating status in real time, detect and isolate faulty inverters, and ensure the system's continuous operation and performance optimization under fault conditions.

2. The control method of parallel inverters with adaptive regulation of active droop coefficient according to claim 1, characterized in that: The DQN algorithm steps include: S1, initialize experience replay pool ; S2, initialization network and target network, parameters are respectively and ; S3. For each round of training: acquiring the current state ; According to - a greedy strategy to select actions ; performing an action , obtaining a next state and a reward ; will be stored in the experience replay pool ;​ randomly sample a batch of data from the experience replay pool ; Computing target Values: ; Compute loss function: ; updating by gradient descent method network parameters ; Periodic update target Network parameters .

3. The parallel inverter control method for adaptively adjusting the active power droop coefficient according to claim 2, characterized in that: The reinforcement learning model training process uses historical operation data and simulation data for model training, constantly optimizing neural network parameters In actual operation, the current state is input into the trained DQN model to obtain the optimal action , and the control parameters of the inverter are adjusted in real time.

4. The control method of parallel inverters with adaptive regulation of active droop coefficient according to claim 1, characterized in that: The consensus algorithm in step 4 employs Practical Byzantine Fault Tolerance, and the specific algorithm steps include: A1, System model: Assume there are nodes, where nodes can be malicious, requiring ; Practical Byzantine Fault Tolerance includes three main phases: pre-prepare, prepare, and commit; A2, phase one: pre-preparation, the master node sends a pre-preparation message to all backup nodes, the pre-preparation message including a proposal to be agreed upon and a sequence number ; A3、Phase two: preparation, after the backup node receives the pre-preparation message, the correctness of the message is verified, and the preparation message is broadcast to all nodes, the preparation message includes the proposal and sequence number ; A4. Phase Three: Commit, backup nodes receive at least 2 After the preparation message is sent, a commit message is broadcast to all nodes. The commit message includes the proposal. and serial number ; A5. Reaching consensus: All nodes receive at least 2 After +1 commit message, the proposal is considered... To reach a consensus and implement the proposal, the messaging formula is as follows: in, and These represent the number of prepared messages and the number of committed messages, respectively.

5. The parallel inverter control method for adaptively adjusting the active power droop coefficient according to claim 1, characterized in that: The specific steps of the self-healing algorithm in step 6 are as follows: B1. Real-time monitoring of system operation status: Collect operating data from each inverter in the system, including voltage, current, frequency, and temperature, and use the data acquisition module to acquire and update the inverter's status information in real time; Formula: Let For the first An inverter in time State vector: in, Indicates the first An inverter in time The voltage; Current; Indicates the first An inverter in time The frequency; temperature; B2. Fault Detection: Define fault detection thresholds, set reasonable thresholds based on the normal operating range of the inverter, and use anomaly detection algorithms to analyze the inverter's status data in real time to determine whether a fault has occurred. Formula: Let the threshold be... The fault detection conditions are: B3. Fault Isolation: When an inverter fault is detected, immediately isolate the faulty inverter to prevent it from affecting the normal operation of the system; redistribute the load to ensure stable system operation; Formula: Let the faulty inverter be the first... An inverter, its state is The isolation procedure is as follows: The formula for redistributing the load is: in: For the load of each inverter after redistribution; Total system load; The load of the faulty inverter; This represents the total number of inverters in the system.

6. The parallel inverter control method for adaptively adjusting the active power droop coefficient according to claim 5, characterized in that: The self-healing algorithm also includes self-healing control, which uses predictive models and historical data to predict and analyze the system operating status, dynamically adjusts the inverter control parameters, and optimizes system performance. The prediction model is: The adjustment amount for self-healing control is The control formula is: Update the inverter control parameters based on the adjustment amount: in: The predicted system state; To determine the difference between the predicted state and the actual state; These are the current control parameters of the inverter; These are the updated control parameters.

7. The parallel inverter control method for adaptively adjusting the active power droop coefficient according to claim 1, characterized in that: The adaptive adjustment of the active power droop coefficient parallel inverter control system includes a parallel inverter system architecture, a data acquisition and processing subsystem, a reinforcement learning control subsystem, a blockchain decentralized control subsystem, a real-time control and adaptive adjustment subsystem, and a system monitoring and self-healing maintenance subsystem. The reinforcement learning control subsystem includes a reinforcement learning model: Status: The current status of the system, including inverter operating data, voltage, current, frequency, and load; Action: The system takes action to adjust the active power droop coefficient; Reward: Feedback obtained after the system takes an action, used to measure the quality of the action; Strategy: The strategy that determines the action a system takes in a given state; Value function: evaluates the quality of a state or state-action pair; Wherein, state is represented as: state Represented as a vector, it includes key operating data of the inverter: , Indicates voltage. Represents current. Indicates frequency, Indicates load; Action space: Action It is to adjust the active power droop coefficient. Value: ,in This refers to the increment or decrement of the active power droop coefficient; Reward function: Reward function Designed as follows: ， in: , , , These are weighting coefficients, which can be adjusted according to the actual application. This represents the voltage deviation, defined as the difference between the target voltage and the actual voltage. Frequency deviation is defined as the difference between the target frequency and the actual frequency. Indicates the degree of load imbalance; Policy and value function: The DQN algorithm is used as the reinforcement learning algorithm. DQN approximates the value function through a deep neural network. ,in These are the parameters of the neural network.

8. The parallel inverter control method for adaptively adjusting the active power droop coefficient according to claim 7, characterized in that: The parallel inverter system architecture includes inverter units and a blockchain network. The inverter unit includes multiple parallel inverters, and each inverter is equipped with the following modules: Controller: Responsible for executing control algorithms, adjusting active power droop coefficients, and executing received control commands; Communication module: Supports communication with other inverters, ensuring data sharing and control collaboration, and supports blockchain network communication; Internet of Things (IoT) sensors include voltage sensors, current sensors, and temperature sensors, used to collect operational data. Edge computing devices: process sensor data, perform preliminary data screening and noise filtering, and transmit the processed data to the controller and blockchain network; The blockchain network includes: Private Blockchain: Establish a private blockchain, with all inverters joining the blockchain network as nodes; Blockchain Nodes: Inverters and their controllers act as blockchain nodes, responsible for recording, broadcasting, and sharing operational data; Smart contract module: Deployed on the blockchain network, used to execute data sharing and control collaboration; The data acquisition and processing subsystem includes an IoT sensor network and an edge computing module. The IoT sensor network is integrated on each inverter to monitor and acquire voltage, current, power, and frequency data in real time. The edge computing module is responsible for performing preliminary processing on the acquired data, reducing data noise, and providing preliminary system performance feedback. The feedback mechanism transmits the processed data to the controller and blockchain network to provide real-time data support for the reinforcement learning model. The reinforcement learning control subsystem includes a reinforcement learning model module: an input port for receiving operating data from the inverter; an output port for providing an adjusted active power droop coefficient for real-time control of the inverter's operation; a reward function for incentivizing the model to maintain stable system voltage and frequency and achieve load balancing; and a model training module for training the model using historical and simulation data and periodically updating model parameters to adapt to new system conditions.

9. The parallel inverter control method for adaptively adjusting the active power droop coefficient according to claim 8, characterized in that: The blockchain decentralized control subsystem includes a data management module: data on-chain: inverter operating data is uploaded to the blockchain via an IoT gateway to ensure data transparency and immutability; smart contract execution module: smart contracts are executed based on on-chain data and the output of the reinforcement learning model to dynamically adjust the inverter's active power droop coefficient; consensus algorithm module: a consensus algorithm is used to ensure the consistency of data and control commands between inverter nodes. The real-time control and adaptive adjustment subsystem includes a control command generation module and an adaptive adjustment module. The control command generation module generates and issues real-time control commands based on the output of the smart contract and the reinforcement learning model. The adaptive adjustment module includes an adjustment module: each inverter adaptively adjusts the active power droop coefficient according to the received control commands to ensure system stability and load balance. Feedback module: The inverter feeds back the adjusted operating status to the blockchain network to achieve closed-loop control; The system monitoring and self-healing maintenance subsystem includes a real-time monitoring module, a fault isolation module, and a maintenance module. The real-time monitoring module implements a self-healing algorithm to monitor and detect faulty inverters in the system in real time; the fault isolation module automatically isolates the faulty inverter when a faulty inverter is detected and adjusts the operation of other inverters to maintain system continuity and optimize performance; the maintenance module generates fault reports and shares them through a blockchain network to facilitate system maintenance and optimization.

Images