Hybrid Predictive Control Method for Island Microgrids Based on Consistency Principle

Abstract

This invention proposes a hybrid predictive control method for island microgrids based on the consistency principle, comprising the following steps: Step S1, dividing the distributed power sources and load clusters, i.e., dividing the island microgrid into n clusters that meet preset conditions according to the spatial distribution of distributed power sources and loads; Step S2, for cluster i containing k distributed power sources and j loads, its internal convergence to an approximate stable point based on the consistency principle; Step S3, solving for the optimal control sequence at time k, and controlling the internal cluster i using model source-load prediction, the model including a reference trajectory cluster Yr generation module, an online optimization module, a control object, a prediction model module, and a prediction correction module; Step S4, repeating steps S2 and S3 at time k+1 until a termination command is received from the control center; This invention can enhance the stable operation capability of island microgrids after unplanned disconnection of submarine cables.

Description

Technical Field

[0001] This invention relates to Background Technology

[0002] Power electronic converters (PECs) are crucial for the reliable integration of distributed renewable energy sources such as wind and solar power into island microgrids, ensuring stable power supply. However, the dispersed and dynamically varying locations of numerous distributed power sources within island microgrids result in highly nonlinear and low-damping dynamic behavior during disturbances such as faults, weakening the support capacity of distributed power sources for the system. Coordinating distributed power sources and loads within the island microgrid and controlling the interaction between them is key to ensuring the reliability and stability of power supply during major disturbances such as faults in submarine interconnection cables.

[0003] Due to the low damping and weak inertia of distributed power sources, when a large disturbance occurs in the island microgrid, such as an unplanned disconnection of the submarine cable connection, the island microgrid will experience severe voltage and frequency fluctuations at extremely high speeds, posing a very high risk of instability. Existing methods mainly rely on decentralized control, where each power source independently navigates the fault. Its advantage lies in low communication requirements, but its control accuracy is poor. To address the shortcomings of decentralized control, the idea of ​​deploying a centralized control center on the island to centrally control the power sources and loads has been proposed. However, the drawback of centralized control is its reliance on a central computing unit for unified scheduling via communication lines. Once the computing center is damaged or the communication lines connecting the various power sources and loads on the island are destroyed, the stable operation efficiency of the island microgrid will drop significantly, even facing a very high risk of collapse. Islands are prone to high temperatures, high humidity, high salinity, and frequent extreme natural disasters, posing a high risk of damage to the computing center and communication lines. Furthermore, centralized control has strict requirements for communication bandwidth, making the initial construction and subsequent operation and maintenance costs considerable.

[0004] Consensus algorithms and model predictive control have gradually become research hotspots in the field of coordinated control of microgrids. Consensus algorithms are a distributed control method that enables coordinated control of distributed power sources across the microgrid through autonomous iterative convergence of local multi-agent systems. Through local information exchange and iterative updates, consensus algorithms can ensure that the output power, voltage, and frequency of each distributed power source converge to a consistent state, thereby achieving load balancing and voltage stability. Consensus algorithms possess strong robustness and adaptability, making them suitable for various complex operating scenarios.

[0005] Model predictive control (MPC) is a model-based optimization control method that effectively addresses the control difficulties caused by the rapid dynamic response of distributed generation in microgrids. By establishing a dynamic model of the island microgrid, MPC predicts the future system state, thereby achieving proactive optimization control of distributed generation. MPC possesses good system performance, multi-objective optimization capabilities, and constraint handling capabilities, making it suitable for handling nonlinear, multivariable, and time-varying characteristics in microgrids.

[0006] Combining consensus algorithms and model predictive control techniques can achieve efficient coordinated control of distributed power sources in microgrids, providing strong support for improving the reliable convergence of island microgrids to a stable operating state after disturbances. However, as the coupling and interaction between power sources, grid, load, and storage deepens, detailed modeling and accurate parameter acquisition of island microgrids become extremely difficult, hindering the integration of consensus algorithms and model predictive control techniques. Summary of the Invention

[0007] This invention proposes a hybrid predictive control method for island microgrids based on the consistency principle, which can enhance the stable operation capability of island microgrids after unplanned disconnection of submarine cables.

[0008] The present invention adopts the following technical solution.

[0009] A hybrid predictive control method for island microgrids based on the consistency principle is used for model-data hybrid collaborative control of island microgrids with a high proportion of renewable energy access. The method is a data-driven reference trajectory generation and model-driven predictive control method, including the following steps.

[0010] Step S1: Division of distributed power sources and load clusters, that is, dividing the island microgrid into n clusters that meet preset conditions according to the spatial distribution of distributed power sources and loads.

[0011] Step S2: For cluster i containing k distributed power sources and j loads, its internal structure converges to an approximately stable point based on the consistency principle.

[0012] Step S3: Solve for the optimal control sequence at time k. Control is performed on cluster i using model source load prediction. The model includes a reference trajectory cluster Yr generation module, an online optimization module, a control object, a prediction model module, and a prediction correction module.

[0013] Step S4: At time k+1, repeat steps S2 and S3 until a termination command is received from the control center.

[0014] In step S1, the preset conditions that the n clusters satisfy are:

[0015] Condition 1: The connections between clusters are loose and sparse, while the internal components of each cluster are closely connected. That is, there are few power line connections between clusters, but the energy exchanges between the internal components of the clusters are frequent and the lines are densely distributed.

[0016] Condition 2: The capacity of the clusters should be basically consistent to ensure the global convergence speed of the island microgrid; let the island microgrid MG be represented by the set C of distributed power sources and loads. MG ,have

[0017] C MG ={C1,C2,…,C i |i=1,2,…n}

[0018] Among them, C i This represents the i-th cluster.

[0019] The specific method of step S2 is as follows: Within each cluster, convergence to an approximately stable point is first achieved based on the consistency principle. For cluster i containing k distributed power sources and j loads, the topology of cluster i is represented by Gi = (K, J, E); where K = {1, 2, ..., k} represents the set of distributed power sources, J = {1, 2, ..., j} represents the set of loads, and E represents a (k+j)×(k+j) matrix of edges. Considering the cluster topology and control commands, cluster i is represented as...

[0020] C i ={G i ,x i ,u i ,w i}

[0021] Where x i It is a vector consisting of the state variables of cluster i, representing the cluster's power, voltage, current, and phase angle; u i It is the internal control variable of cluster i, w i It is an external input control variable for cluster i, used to achieve consistent coordination of jobs between clusters;

[0022] If cluster i is equivalent to a first-order multi-agent system, then...

[0023]

[0024] Where, m∈Γ i , Γ i ={1,2,…,k,…,k+j-1};

[0025] The consistency principle achieves consensus through the following expression:

[0026]

[0027] Among them, ai is an element in the Laplace matrix of cluster i, where the superscripts m and t indicate the connection relationship between nodes m and t;

[0028] When island microgrids experience numerous extreme weather conditions and salt spray corrosion, leading to large errors in line parameters and low communication reliability, a data-driven expression based on the consistency principle is used to ensure consistency.

[0029]

[0030] Where χ represents the corrected state variable output by the data-driven algorithm, and τ represents the error compensation value. The matrix formed by the corrected state variable values ​​and error compensation values ​​within cluster i is χ. i and τ i The two are calculated using the following formula:

[0031]

[0032]

[0033] Where vi, viλ, vi1 are the state variable weight matrices, bi, biλ, bi1 are the state variable bias matrices, νiλ, νi1 are the error compensation weight matrices, and βiλ, βi1, βi are the error compensation bias matrices. The weight matrices and bias matrices are trained and optimized using training algorithms such as gradient descent, and the objective function can be calculated using the squared difference formula.

[0034] In step S3, the reference trajectory cluster Yr consists of the consistency reference control instruction ui, the neighboring cluster consistency reference instruction u, and the actual output Y;

[0035] After receiving the uim calculated from the proposed data-driven consistency expression, cluster i uses it as an adjustment amount and superimposes it with the current local actual output and the consistency reference instructions of neighboring clusters to generate the reference trajectory for the next time step. The expression is:

[0036] Y r =ξu i +ψY i +ζu NER

[0037] Among them, u NER This is a reference instruction for neighboring cluster consistency. The difference between the reference trajectory cluster Yr and the corrected predicted value Yc is input into the online rolling optimization module to solve for the optimal control sequence at time k. The optimal control sequence is obtained by minimizing the objective function, the expression of which is:

[0038]

[0039] The optimization algorithm for solving the optimal control sequence adopts the gradient descent method, and the generated optimal control sequence uses the first control signal of the sequence as input to the controlled object and the prediction model.

[0040] The prediction model is described by mathematical models of each power source and load. The prediction and output error e is obtained by subtracting the prediction model output Ym from the actual output Y of the controlled object. This error, along with the prediction model output Ym, is input into the prediction correction module, where the corrected prediction value Yc is calculated using the Kalman filtering algorithm.

[0041] In step S1, the island microgrid is divided into clusters based on the busbar, so that the island microgrid is divided into multiple clusters according to the spatial distribution and power coupling degree of each micro source.

[0042] The method achieves rapid generation and distribution of consistency targets through a neural network structure, and uses it as the reference trajectory signal input for the model control algorithm of each converter. Its online calculation process is completed by an edge computing unit to reduce communication bandwidth requirements and computing power requirements.

[0043] In step S2, when the island microgrid is subjected to large disturbances, including the disconnection of the submarine connecting cable, local stability is first achieved within the cluster, and then global stability is achieved through communication between clusters. This ensures the timely recovery of the entire island microgrid to near the rated point after the disturbance, and the accuracy of global control is achieved through coordination and fine-tuning between clusters.

[0044] In step S1, the power lines between clusters include at least one lightly loaded line.

[0045] This invention enables system-level stability enhancement for island microgrids, particularly for island microgrid clusters with a high proportion of renewable energy access, where the source, grid, load, and energy storage rapidly and collaboratively enhance the stability of system voltage and frequency after a submarine cable interconnection fault is disconnected. This invention also enhances the stable operation capability of island microgrids after unplanned disconnection of submarine cables.

[0046] In this invention, online computing is performed through edge computing units, which significantly reduces the communication bandwidth requirements and computing power requirements.

[0047] Compared with existing technologies, the advantages of this invention are as follows: This invention uses a hybrid data and model approach to drive source-load control of island microgrids. Compared with traditional methods, this invention adopts a hierarchical control approach, integrating the robustness and intelligence of data-driven methods with the interpretability and security of model-driven methods. This enables the island microgrid to maintain a certain level of stable operation even under large disturbances such as the disconnection of submarine cables, which occur in rapidly changing and harsh environments. Furthermore, by dividing the island microgrid into clusters, local stability is first achieved within each cluster, and then global stability is achieved through inter-cluster communication. This ensures the rapid recovery of the entire island microgrid to near its rated point after a disturbance, while still maintaining precise global control through inter-cluster coordination and fine-tuning. Attached Figure Description

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0049] Appendix Figure 1 This is a schematic diagram of the structure of each component of the present invention;

[0050] Appendix Figure 2 This is a schematic diagram of the system instability simulation waveform when the submarine connecting cable was cut off and the method of this invention was not used in the simulation of a microgrid on an island (where the relative angle deviation is the integral of the difference between each converter cluster and the rated angular frequency with respect to time).

[0051] Appendix Figure 3 This is a schematic diagram of the system stability simulation waveform when the method of this invention is used in the simulation of a microgrid on an island after the submarine connecting cable is cut off (where the relative angle deviation is the integral of the difference between each converter cluster and the rated angular frequency with respect to time). Detailed Implementation

[0052] like Figure 1 As shown, a hybrid predictive control method for island microgrids based on the consistency principle is used for model-data hybrid collaborative control of island microgrids with a high proportion of renewable energy access. The method is a data-driven reference trajectory generation and model-driven predictive control method, including the following steps;

[0053] Step S1: Division of distributed power sources and load clusters, that is, dividing the island microgrid into n clusters that meet preset conditions according to the spatial distribution of distributed power sources and loads.

[0054] Step S2: For cluster i containing k distributed power sources and j loads, its internal structure converges to an approximately stable point based on the consistency principle.

[0055] Step S3: Solve for the optimal control sequence at time k. Control is performed on cluster i using model source load prediction. The model includes a reference trajectory cluster Yr generation module, an online optimization module, a control object, a prediction model module, and a prediction correction module.

[0056] Step S4: At time k+1, repeat steps S2 and S3 until a termination command is received from the control center.

[0057] In step S1, the preset conditions that the n clusters satisfy are:

[0058] Condition 1: The connections between clusters are loose and sparse, while the internal components of each cluster are closely connected. That is, there are few power line connections between clusters, but the energy exchanges between the internal components of the clusters are frequent and the lines are densely distributed.

[0059] Condition 2: The capacity of the clusters should be basically consistent to ensure the global convergence speed of the island microgrid; let the island microgrid MG be represented by the set C of distributed power sources and loads. MG ,have

[0060] C MG ={C1,C2,…,C i |i=1,2,…n}

[0061] Among them, C i This represents the i-th cluster.

[0062] The specific method of step S2 is as follows: Within each cluster, convergence to an approximately stable point is first achieved based on the consistency principle. For cluster i containing k distributed power sources and j loads, the topology of cluster i is represented by Gi = (K, J, E); where K = {1, 2, ..., k} represents the set of distributed power sources, J = {1, 2, ..., j} represents the set of loads, and E represents a (k+j)×(k+j) matrix of edges. Considering the cluster topology and control commands, cluster i is represented as...

[0063] C i ={G i ,x i ,u i ,w i}

[0064] Where x i It is a vector consisting of the state variables of cluster i, representing the cluster's power, voltage, current, and phase angle; u i It is the internal control variable of cluster i, w i It is an external input control variable for cluster i, used to achieve consistent coordination of jobs between clusters;

[0065] If cluster i is equivalent to a first-order multi-agent system, then...

[0066]

[0067] Where, m∈Γ i , Γ i ={1,2,…,k,…,k+j-1};

[0068] The consistency principle achieves consensus through the following expression:

[0069]

[0070] Among them, a i is an element in the Laplace matrix of cluster i, where the superscripts m and t indicate the connection relationship between nodes m and t;

[0071] When island microgrids experience numerous extreme weather conditions and salt spray corrosion, leading to large errors in line parameters and low communication reliability, a data-driven expression based on the consistency principle is used to ensure consistency.

[0072]

[0073] Where χ represents the corrected state variable output by the data-driven algorithm, and τ represents the error compensation value. The matrix formed by the corrected state variable values ​​and error compensation values ​​within cluster i is χ. i and τ i The two are calculated using the following formula:

[0074]

[0075]

[0076] Where vi, viλ, vi1 are the state variable weight matrices, bi, biλ, bi1 are the state variable bias matrices, νiλ, νi1 are the error compensation weight matrices, and βiλ, βi1, βi are the error compensation bias matrices. The weight matrices and bias matrices are trained and optimized using training algorithms such as gradient descent, and the objective function can be calculated using the squared difference formula.

[0077] In step S3, the reference trajectory cluster Yr consists of the consistency reference control instruction ui, the neighboring cluster consistency reference instruction u, and the actual output Y;

[0078] After receiving the uim calculated from the proposed data-driven consistency expression, cluster i uses it as an adjustment amount and superimposes it with the current local actual output and the consistency reference instructions of neighboring clusters to generate the reference trajectory for the next time step. The expression is:

[0079] Y r =ξu i +ψY i+ζu NER

[0080] Among them, u NER This is a reference instruction for neighboring cluster consistency. The difference between the reference trajectory cluster Yr and the corrected predicted value Yc is input into the online rolling optimization module to solve for the optimal control sequence at time k. The optimal control sequence is obtained by minimizing the objective function, the expression of which is:

[0081]

[0082] The optimization algorithm for solving the optimal control sequence adopts the gradient descent method, and the generated optimal control sequence uses the first control signal of the sequence as input to the controlled object and the prediction model.

[0083] The prediction model is described by mathematical models of each power source and load. The prediction and output error e is obtained by subtracting the prediction model output Ym from the actual output Y of the controlled object. This error, along with the prediction model output Ym, is input into the prediction correction module, where the corrected prediction value Yc is calculated using the Kalman filtering algorithm.

[0084] In step S1, the island microgrid is divided into clusters based on the busbar, so that the island microgrid is divided into multiple clusters according to the spatial distribution and power coupling degree of each micro source.

[0085] The method achieves rapid generation and distribution of consistency targets through a neural network structure, and uses it as the reference trajectory signal input for the model control algorithm of each converter. Its online calculation process is completed by an edge computing unit to reduce communication bandwidth requirements and computing power requirements.

[0086] In step S2, when the island microgrid is subjected to large disturbances, including the disconnection of the submarine connecting cable, local stability is first achieved within the cluster, and then global stability is achieved through communication between clusters. This ensures the timely recovery of the entire island microgrid to near the rated point after the disturbance, and the accuracy of global control is achieved through coordination and fine-tuning between clusters.

[0087] In step S1, the power lines between clusters include at least one lightly loaded line.

[0088] like Figure 2 , Figure 3 The comparison shows that before and after using the method of the present invention, the island microgrid can maintain stable operation under the regulation of the proposed method when experiencing the same large disturbance, while the simulation without the proposed method of the present invention shows system instability.

[0089] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0090] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A hybrid predictive control method for island microgrids based on the consistency principle, used for model-data hybrid collaborative control of island microgrids with a high proportion of renewable energy access, characterized by: The method is a data-driven reference trajectory generation and model-driven predictive control method, including the following steps: Step S1: Division of distributed power sources and load clusters, that is, dividing the island microgrid into n clusters that meet preset conditions according to the spatial distribution of distributed power sources and loads. Step S2: For cluster i containing k distributed power sources and j loads, its internal structure converges to an approximately stable point based on the consistency principle. Step S3: Solve for the optimal control sequence at time k. Control is performed on cluster i using model source load prediction. The model includes a reference trajectory cluster Yr generation module, an online optimization module, a control object, a prediction model module, and a prediction correction module. Step S4: At time k+1, repeat steps S2 and S3 until a termination command is received from the control center. In step S1, the preset conditions that the n clusters satisfy are: Condition 1: The connections between clusters are loose and sparse, while the internal components of each cluster are closely connected. That is, there are few power line connections between clusters, but the energy exchanges between the internal components of the clusters are frequent and the lines are densely distributed. Condition 2: The capacity of the clusters should be basically consistent to ensure the global convergence speed of the island microgrid; let the island microgrid MG be represented by the set C of distributed power sources and loads. MG ,have Among them, C i Indicates the i-th cluster; The specific method of step S2 is as follows: Within each cluster, convergence to an approximately stable point is first achieved based on the consistency principle; for cluster i containing k distributed power sources and j loads, the topology of cluster i is represented by Gi=(K,J,E); where K={1,2,…,k} represents the set of distributed power sources, J={1,2,…,j} represents the set of loads, and E represents a matrix where the set of edges is a (k+j)×(k+j) matrix; considering the cluster topology and control commands, cluster i is represented as... Where x i It is a vector consisting of the state variables of cluster i, representing the cluster's power, voltage, current, and phase angle; u i It is the internal control variable of cluster i, w i It is an external input control variable for cluster i, used to achieve consistent coordination of jobs between clusters; If cluster i is equivalent to a first-order multi-agent system, then... Where, m∈Γ i , Γ i ={1,2,…,k,…,k+j-1}; The consistency principle achieves consensus through the following expression: Among them, a i is an element in the Laplace matrix of cluster i, where the superscripts m and t indicate the connection relationship between nodes m and t; When island microgrids experience numerous extreme weather conditions and salt spray corrosion, leading to large errors in line parameters and low communication reliability, a data-driven expression based on the consistency principle is used to ensure consistency. Where χ represents the state variable correction value output by the data-driven algorithm, and τ represents the error compensation value; the matrix formed by the state variable correction values ​​and error compensation values ​​within cluster i is χ. i and τ i Both are obtained through the following formula: Where vi, viλ, vi1 are the state variable weight matrices, bi, biλ, bi1 are the state variable bias matrices, νiλ, νi1 are the error compensation weight matrices, and βiλ, βi1, βi are the error compensation bias matrices; the weight matrices and bias matrices are trained and optimized using the gradient descent training algorithm, and the objective function can be calculated using the squared difference formula. In step S3, the reference trajectory cluster Y r Consistency reference control instruction u i It consists of the neighboring cluster consistency reference instruction u and the actual output Y of the controlled object; Cluster i receives u calculated from the proposed data-driven consistency expression. im Then, this is used as an adjustment amount and superimposed with the current local actual output and the neighboring cluster consistency reference instructions to generate the reference trajectory for the next moment, expressed as follows: Among them, u NER This is the neighbor cluster consensus reference instruction; the difference between the reference trajectory cluster Yr and the corrected prediction value Yc is input into the online rolling optimization module to solve for the optimal control sequence at the current time k. The optimal control sequence is obtained by minimizing the objective function, the expression of which is: The optimization algorithm for solving the optimal control sequence adopts the gradient descent method, and the generated optimal control sequence uses the first control signal of the sequence as input to the controlled object and the prediction model. The prediction model is described by mathematical models of each power source and load. The prediction and output error e is obtained by subtracting the prediction model output Ym from the actual output Y of the controlled object. This error, along with the prediction model output Ym, is input into the prediction correction module, where the corrected prediction value Yc is calculated using the Kalman filtering algorithm.

2. The hybrid predictive control method for island microgrids based on the consistency principle according to claim 1, characterized in that: In step S1, the island microgrid is divided into clusters based on the busbar, so that the island microgrid is divided into multiple clusters according to the spatial distribution and power coupling degree of each micro source.

3. The hybrid predictive control method for island microgrids based on the consistency principle according to claim 2, characterized in that: The method achieves rapid generation and distribution of consistency targets through a neural network structure, and uses it as the reference trajectory signal input for the model control algorithm of each converter. Its online calculation process is completed by an edge computing unit to reduce communication bandwidth requirements and computing power requirements. In step S2, when the island microgrid is subjected to large disturbances, including the disconnection of the submarine connecting cable, local stability is first achieved within the cluster, and then global stability is achieved through communication between clusters. This ensures the timely recovery of the entire island microgrid to near the rated point after the disturbance, and the accuracy of global control is achieved through coordination and fine-tuning between clusters.

4. The hybrid predictive control method for island microgrids based on the consistency principle according to claim 1, characterized in that: In step S1, the power lines between clusters include at least one lightly loaded line.

Images