DC microgrid fault and network attack distinguishing detection method and system
By adopting a combined detection strategy of centralized unknown input observer and distributed Luenberger observer in DC microgrid, decomposing and distinguishing actuator failures and network attacks, the problem of limited detection effect in the existing technology is solved, and the correctness of the detection effect is achieved Fast and accurate distinction and detection of actuator failures and cyberattacks.
Patent Information
- Application Number
- CN202211228464.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-08
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2042-10-08
AI Technical Summary
The prior art is difficult to simultaneously detect and distinguish actuator failures and network attacks in a DC microgrid, especially when the communication network is heterogeneous with the physically coupled topologically, resulting in the impact of detection effects.
Using network layer attack detection based on centralized unknown input observers and physical layer actuator fault detection based on distributed Luenberger observers, combined with the superposition principle and unknown input-Luenberger dual observer detection strategy, decompose and distinguish between the DC microgrid Physical layer failure and network layer attacks.
It realizes rapid detection and distinction between DC microgrid actuator failures and network attacks, reduces detection complexity, and improves detection accuracy under heterogeneous topology.
Smart Images

Figure CN115473210B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of direct current microgrids, and in particular to a method and system for distinguishing and detecting faults and network attacks in a direct current microgrid. Background Art
[0002] A DC microgrid is a microgrid powered by DC. DC microgrids are booming in remote areas, urban emergency power generation, and large-scale charging facilities for electric vehicles. In the actual use of DC microgrids, due to the long-term wear of the system's own physical components and the influence of the external environment, the physical components are prone to failure. In addition, the DC microgrid's high dependence on communication technology increases the risk of malicious network attacks on the DC microgrid's communication link. Therefore, in real DC microgrid application scenarios, it is not uncommon for network attacks and actuator failures to occur at the same time. The external manifestations of the two situations of DC microgrid's own failure and network attack are similar and difficult to distinguish. If the two can be distinguished, corresponding treatment measures can be taken as soon as possible. At present, the existing technology often only considers a single fault detection or attack detection, and has not yet effectively realized the simultaneous detection and distinction of faults and attacks in DC microgrids; specifically, the existing technology often considers fault or attack detection in a single case where the DC microgrid communication network and physical coupling are isomorphic. In the case where the communication network and physical coupling are topologically heterogeneous, the existing fault or attack detection effect will be affected. Summary of the invention
[0003] The object of the present invention is to provide a method and system for distinguishing and detecting DC microgrid faults and network attacks, so as to solve the problem that the existing DC microgrid actuator faults and network attacks cannot be distinguished.
[0004] In order to solve the above technical problems, the present invention provides a method for distinguishing and detecting DC microgrid faults and network attacks, comprising:
[0005] Decomposing DC microgrid actuator fault detection and DC microgrid network attack detection into network layer attack detection based on centralized unknown input observer and physical layer actuator fault detection based on distributed Luenberger observer, so as to distinguish between DC microgrid physical layer fault and network layer attack; and
[0006] Based on the superposition principle and the unknown input-Luenberger dual observer detection strategy, the DC microgrid actuator fault detection and DC microgrid network attack detection are decomposed, wherein the superposition principle and the unknown input-Luenberger dual observer detection strategy include: based on the analysis of the structural limitations of the unknown input observer for fault detection and the undetectability of the Luenberger observer for covert attacks, the two observers are combined for detection, wherein the limitations of a single observer itself are used to distinguish between faults and attacks.
[0007] Optionally, the DC microgrid fault and network attack distinguishing detection method further includes:
[0008] For the network layer detection strategy, considering that the communication network and physical topology are heterogeneous, and that only the physical coupling information between distributed generation units cannot fully detect the attack, a set of unknown input observers with a centralized architecture is used to obtain the state estimation information of adjacent distributed generation units by augmenting the unknown inputs as well as coupling noise and faults. Then, in view of the detection structure limitations caused by augmenting the unknown inputs, artificial states are introduced, and residuals and corresponding thresholds are designed to effectively detect possible false data injection attacks in the DC microgrid; and
[0009] For the physical layer detection strategy, based on the consideration of process and measurement noise caused by DC microgrid modeling mismatch and external environment, component aging, actuator failure caused by external environment influence, and covert false data injection attack injected into the controller communication channel, the distributed structure Luenberger observer is used through the physical coupling information interaction between the neighbors of DC microgrid distributed generation units to achieve the detection of physical layer failure. ∞ / H - Performance indicators and residual decomposition algorithm are used to detect actuator faults at the physical layer of DC microgrids.
[0010] Optionally, the DC microgrid fault and network attack distinguishing detection method further includes:
[0011] Construct unknown input observer module and Luenberger observer module to detect and distinguish actuator failures and network attacks;
[0012] In the network attack detection module based on unknown input observer, the following unknown input observer is designed as the neighbor estimation state of the local distributed generation unit:
[0013]
[0014] Among them, z j,i (t) is the state of the unknown input observer, Represents the augmented state x j The estimate of (t), represents the unknown state observer output, represents the information received by distributed generation unit i from its neighbors, C i represents the system description matrix, G j , J j , H j is a matrix designed under the following constraints:
[0015]
[0016] The unknown input observer used to estimate the state in conventional attack detection is converted into an unknown input observer group used to estimate the neighbor state under the dual influence of fault attack; based on the improvement of normalizing fault and noise, the residual is made insensitive to fault, so as to accurately detect network attacks under the influence of actuator fault, and at the same time, the improvement of artificial state is introduced to improve the limitation of no redundant detection attack due to the same number of unknown input and output after augmentation;
[0017] In the actuator fault detection module based on Luenberger observer, the following residual decomposition algorithm is designed:
[0018]
[0019] Among them, ∈ i (t) and r i (t) represent the state estimation error and residual of agent i under the influence of network attack and actuator failure, respectively. and r i n (t) represent the state estimation error and residual corresponding to the “nominal” part of agent i, respectively. and r i a (t) represent the state estimation error and residual corresponding to the “attacked” part of agent i, respectively. and r i f (t) represent the state estimation error and residual corresponding to the “faulty” part of agent i, respectively;
[0020] Based on the improvement of residual decomposition, the concealment of Luenberger observer is exploited by false data injection attack, combined with the residual evaluation function based on root mean square and the corresponding threshold, the system is used to detect actuator faults under network attacks.
[0021] Optionally, the DC microgrid fault and network attack distinguishing detection method further includes:
[0022] The DC microgrid system consists of multiple distributed generation units;
[0023] In the initial stage: multiple power generation units are disconnected from each other, and the activated main controller makes the voltage at the common coupling point track the reference voltage to ensure voltage stability;
[0024] In the nominal operation stage: multiple power generation units are connected to each other through power lines and communication channels, both the main controller and the secondary controller are activated, and the voltage balance and current sharing of the DC microgrid system are achieved based on a secure communication channel and a healthy physical layer;
[0025] In the network attack injection phase: a distributed generation unit sends information to a neighboring distributed generation unit through one communication link. If two or more communication links of the same distributed generation unit are attacked by false data injection at the same time, the network attack detection module based on the unknown input observer in each distributed generation unit quickly detects whether the residual curve of the corresponding communication link exceeds the corresponding threshold curve, and issues an attack alarm.
[0026] At the actuator failure stage: If the actuator of the IGBT / MOSFET switch in the distributed generation unit fails based on the network attack injection stage, the actuator fault detection module based on the Luenberger observer in each distributed generation unit quickly detects whether the corresponding residual evaluation function curve exceeds the threshold curve and issues a fault alarm.
[0027] Optionally, the DC microgrid fault and network attack distinguishing detection method further includes:
[0028] Step 1: First, establish a single distributed generation unit model with actuator effectiveness loss failure under the influence of noise as follows:
[0029]
[0030] where x i (t)=[V i (t),I ti (t),ε i ] T Indicates the local state, V i (t) is the common coupling point voltage, I ti (t) is the Buck converter current, εi is the integrator state inside the main controller, and u i (t)=[V ti (t),V r,i ] T Represents the control input, V ti (t) is the main controller input, V r,iis the local reference voltage, y i (t) is the measurable output, represents the physical interconnection between distributed generation units, is the neighbor set of distributed generation unit i, w i (t) = I Li (t) is the external known input, I Li (t) is the load; represent the process noise and measurement noise of the system respectively, Indicates actuator failure; A ii , B i , C i , M i Respectively represent the state matrix, control matrix, observation matrix and external known input matrix, A ij represents the interconnection matrix between distributed generation unit i and distributed generation unit j, E 1i , E 2i are known fault matrices and interference matrices with appropriate dimensions; the specific parameters are given by:
[0031]
[0032] Among them, R ij Indicates line resistance, C ti Represents the parallel capacitance, L ti Filter inductor, R ti Filter resistor;
[0033] Secondly, the following actuator partial failure fault model of IGBT / MOSFET switch in distributed generation unit is given:
[0034] f i (t) = γ i V ti (t),t≥T i f
[0035] Among them, T i f is the moment when the actuator of distributed generation unit i fails, γ i is the failure factor, -1<γ i ≤0, V ti (t) is the main controller input.
[0036] Optionally, the DC microgrid fault and network attack distinguishing detection method further includes:
[0037] Step 2: Define the following false data injection attack model that occurs on the communication link between distributed generation unit controllers:
[0038]
[0039] in, represents the information received by distributed generation unit i from its neighbors, y j (t) represents the output of the augmented system after the artificial state is introduced, Represents a potential false data injection attack The influence of j,i (t) represents the impact of network attacks on the artificial state during communication transmission, represents the false data injected into the communication link (j,i), α j,i (t) is the step function generated by finite superposition, which represents the activation signal of malicious attack;
[0040] Consider the false data injection attack that occurs on the controller communication link. a i A constant bias is cooperatively injected into the communication link where distributed generation unit i receives its neighbor information at all times, so that its impact on the system output is hidden from the Luenberger observer:
[0041]
[0042]
[0043] Among them, T i a is the moment when the network attack is injected into the distributed generation unit i to receive the neighbor information communication link, represents the set of neighbors whose communication links are attacked, represents the physical coupling matrix of the neighbors whose communication links are attacked,
[0044] Optionally, the DC microgrid fault and network attack distinguishing detection method further includes:
[0045] Step 3: Design a network attack detection module based on unknown input observer to detect malicious network attacks on the communication link between distributed generation unit controllers in the presence of faults;
[0046] First, the neighborhood dynamic model of distributed generation unit i is reformulated as follows:
[0047]
[0048] in, represents the normalized uncertainty variable for process noise and actuator faults, represent the process noise and measurement noise of the system respectively, Indicates an actuator failure. represents the unknown input after augmentation, represents the neighbors of distributed generation unit j, contains all the inputs of the neighbors of DG unit i that are unknown to the detector, is the matrix D j The column-full rank matrix of the basis of is based on It is deduced that A jj is the state matrix, C j is the observation matrix, E 1j , E 2j The fault matrix and interference matrix are known to have appropriate dimensions, and the system matrix D j and noise fault coupling matrix The specific description is as follows:
[0049]
[0050] Use state enhancement technology to improve the detection ability of unknown input observers and improve the limitation of the original unknown input observers that there is no redundant detection attack; introduce artificial state variables definition Based on the following relationship:
[0051]
[0052] Among them, x j (t) and y j (t) is the state and output of the original system before the artificial state is introduced, I represents the unit matrix, is the augmented system state, y j (t) represents the output of the system after reconstruction. The following reconstructed system dynamic model is obtained:
[0053]
[0054] in, is the augmented system state, y j (t) represents the output of the system after reconstruction, yes The rearranged transpose, d j,1 (t) = u j (t) The input u is unknown to the local observer of DG unit i but known to the controller of its neighboring DG unit j. j (t) composition, It consists of unknown inputs to the local observer of DG unit i and the controller of its neighboring DG unit j. represents the normalized uncertainty variable for process noise and actuator faults, represent the process noise and measurement noise of the system respectively, Indicates an actuator failure. E 1j , E 2j The fault matrix and interference matrix are known to have appropriate dimensions, and the matrix A jj , D j , C j , N j The specific parameters are as follows:
[0055]
[0056] Design the following centralized unknown input observer group to obtain the state estimation of the neighbors of distributed generation unit i:
[0057]
[0058] Among them, z j,i( t) is the state of the unknown input observer, Represents the augmented state x j The estimate of (t), represents the unknown state observer output, represents the information received by distributed generation unit i from its neighbors, C i represents the observation matrix, G j , J j , H j is a matrix designed under the following constraints:
[0059]
[0060] Define the state estimation error: Define the residual: According to the above unknown input observer matrix, the following dynamic error model under nominal state is designed:
[0061]
[0062] in, represent the process noise and measurement noise of the system respectively, Indicates an actuator failure. Indicates an actuator failure. represents the normalized uncertainty variable for process noise and actuator faults, E 1j , E 2jare known fault and interference matrices with appropriate dimensions, is the coupling matrix of process noise and actuator faults, G j , J j , H j , Θ j It is a matrix designed under appropriate constraints;
[0063] In order to realize the network attack detection on the communication link of the distributed generation unit, the residual threshold is defined according to the dynamic error model under the nominal state as follows:
[0064]
[0065] in σ j >0,Σ j ≥I; are the known positive upper bounds of process noise, measurement noise, and actuator fault, respectively; is a properly defined unknown initial state estimation error the boundaries of;
[0066] In the case of actuator failure in the DC microgrid, when the false data injection attack on the communication link is activated, in the attack detection module based on the unknown input observer, by introducing the residual decomposition algorithm, the error dynamic equation of the system is expressed as the superposition of the "nominal" and "attacked" parts, as shown below:
[0067]
[0068] in and As given in the dynamic error model under the nominal state above, the state estimation error and residual of the “attacked” part are expressed as follows:
[0069]
[0070] in, Represents a potential false data injection attack The influence of j,i (t) represents the impact of network attacks on the artificial state during communication transmission, represents the false data injected into the communication link (j,i), G j , J j , H j is a matrix designed under appropriate constraints, T i a It is the moment when a cyber attack is injected into the communication link;
[0071] According to the dynamic error model under the nominal state designed above and the appropriately defined residual threshold, the following attack detection inequality is designed:
[0072]
[0073] It represents a sufficient condition for attack detection. When this condition is not met, it means that the communication link (j, i) is subjected to a network attack.
[0074] Optionally, the DC microgrid fault and network attack distinguishing detection method further includes:
[0075] Step 4: In order to realize actuator fault detection in the presence of malicious network attacks on the communication link, an actuator fault detection module based on Luenberger observer is designed. First, the following Luenberger observer is designed to obtain the local state estimate of distributed generation unit i:
[0076]
[0077] Among them, A ii , B i , C i , M i Represent the state matrix, control matrix, observation matrix and external known input matrix respectively, u i represents the control input, w i =I Li (t) represents the external known input, I Li (t) is the load, y i represents the local measurable output of distributed generation unit i, L i The proper design of A Li =(A ii +L i C i ) is Hurwitz stable, Indicates the impact of physical interconnection with neighbors and is defined as follows:
[0078]
[0079] Among them, A ij represents the interconnection matrix between distributed generation units i and j. When C j When it is non-strange, Use as x j Neighborhood state estimation of (t) When C j When is singular, the design of the Luenberger observer requires the neighbor state estimate of the unknown input observer in step 4
[0080] Further, based on the obtained local state estimate Define the state estimation error: Define the residual:
[0081] Design the following dynamic error model under nominal conditions:
[0082]
[0083] Among them, C i represents the observation matrix, L i The proper design of A Li =(A ii +L i C i ) is Hurwitz stable, They represent the process noise and measurement noise of the system respectively, and E 1i , E 2i is a known interference matrix with appropriate dimensions;
[0084] According to the dynamic error model under the nominal state, the following time-varying actuator fault detection threshold is defined:
[0085]
[0086] in, is a properly defined unknown initial state estimation error The borders of are the known positive upper bounds of process noise and measurement noise, E 1i , E 2i is a known interference matrix with appropriate dimensions, λ i >0,Λ i ≥I;
[0087] For a given DC microgrid system, attacks and faults are additive. In order to detect actuator faults in the presence of malicious network attacks on the communication link, the following residual decomposition algorithm is introduced to decompose the system residual into three parts: "nominal", "attacked", and "faulty":
[0088]
[0089] in, and It has been given in the dynamic error model under the above nominal state, r i a (t) and r i f(t) The part of the residual that is affected by the attack and the fault, respectively;
[0090] The state estimation error and residual of the “faulty” part are expressed as follows:
[0091]
[0092] Among them, A Li =(A ii +L i C i ) is Hurwitz stable, C i represents the observation matrix, Indicates an actuator failure. is a known fault matrix with appropriate dimensions, T i f is the moment when the actuator of distributed generation unit i fails;
[0093] The state estimation error and residual of the “attacked” part are expressed as follows:
[0094]
[0095] Among them, A Li =(A ii +L i C i ) is Hurwitz stable, C i represents the observation matrix, A ij represents the interconnection matrix between distributed generation unit i and distributed generation unit j, Represents a potential false data injection attack The influence of j,i (t) represents the impact of network attacks on the artificial state during communication transmission, represents the false data injected into the communication link (j,i), T a i It is the moment when the cyber attack is injected into the DG unit i to receive the neighbor information communication link;
[0096] Based on the modeling of the false data injection attack in step 2, the following characteristics of the residual of the "attacked" part of the Luenberger observer are obtained:
[0097] |r i a (t)|=0
[0098] The error dynamic equation of the DC microgrid system affected by both faults and attacks is reformulated as follows:
[0099]
[0100] Among them, ∈ i (t) and r i (t) represent the state estimation error and residual of agent i under the influence of network attack and actuator failure, respectively. and r i n (t) represent the state estimation error and residual corresponding to the “nominal” part of agent i, respectively. and r i f (t) represent the state estimation error and residual corresponding to the affected part of the “fault” of agent i, respectively;
[0101] The following RMS-based error estimation function and corresponding threshold are introduced to achieve timely and accurate fault detection:
[0102]
[0103]
[0104] Based on the definition of the residual evaluation function and the selected threshold, the following detection logic is proposed to detect the occurrence of faults;
[0105]
[0106] Optionally, the DC microgrid fault and network attack distinguishing detection method further includes:
[0107] Step 5: Combining steps 3 and 4, in the task of distinguishing and detecting actuator faults and network attacks in the DC microgrid, the unknown input-Luenberger dual observer detection strategy is implemented. By combining two modules and using different information structures, actuator fault and network attack detection is achieved.
[0108] When there is no fault or attack, the DC microgrid maintains voltage stability, current sharing and voltage balance with the cooperation of the two-stage controller. When an actuator fault occurs, the detection logic based on the Luenberger observer changes from 0 to 1, and the detector issues a fault alarm. When a network attack occurs, the detection logic based on the unknown input observer changes from 0 to 1, and the detector issues an attack alarm. Therefore, in each distributed generation unit, the attack-fault differentiation detector composed of the unknown input-Luenberger dual observer corresponds to the following multi-module structure:
[0109] Module 1 is configured as an unknown input observer group for observing neighbor states. Based on the correct grasp of neighbor model information, the unknown input observer group is used to calculate the estimated state of each neighbor of distributed generation unit i, calculate the state estimation error and residual, and detect network attacks on the communication link between distributed generation unit i and its neighbors by comparing the residual with a predefined threshold.
[0110] Module 2 is configured as a Luenberger observer to observe the local state. Based on the physical coupling information between the systems, the Luenberger observer is used to estimate the local state for the distributed generation unit i, and the state estimation error and the residual evaluation function based on the root mean square are calculated. By comparing the residual evaluation function with the corresponding threshold, it is detected whether there is an actuator fault in the distributed generation unit i.
[0111] The present invention also provides a DC microgrid fault and network attack distinguishing detection system, comprising:
[0112] The dual observer detection strategy module is configured to be based on the superposition principle and the unknown input-Luenberger dual observer detection strategy, which decomposes the DC microgrid actuator fault detection and the DC microgrid network attack detection into the network layer attack detection based on the centralized unknown input observer and the physical layer actuator fault detection based on the distributed Luenberger observer, so as to distinguish the DC microgrid physical layer fault from the network layer attack;
[0113] an unknown input observer module, configured to serve as a network attack detection module, and to perform network layer attack detection based on a centralized unknown input observer; and
[0114] A Luenberger observer module is configured to serve as an actuator fault detection module for performing physical layer actuator fault detection based on a distributed Luenberger observer;
[0115] The detection strategy based on the superposition principle and unknown input-Luenberger dual observer includes: on the basis of analyzing the structural limitations of the unknown input observer for fault detection and the undetectability of the Luenberger observer for covert attacks, the two observers are combined to utilize the original performance advantages of the observers to achieve the detection purpose, and the limitations of a single observer itself become the entry point for differentiated detection of faults and attacks.
[0116] In the DC microgrid fault and network attack distinguishing detection method and system provided by the present invention, a distinguishable detection method for DC microgrid actuator faults and network attacks is proposed, which can adopt different elastic control strategies according to the inherent characteristics of the attack and fault, thereby reducing the complexity of the mitigation strategy.
[0117] The present invention proposes a method for detecting and distinguishing actuator faults and network attacks based on dual observers, aiming to solve the scenario in which a DC microgrid has both actuator faults and network attacks. The proposed method can be sensitive to both actuator faults and network attacks at the same time, thereby achieving rapid detection and differentiation of actuator faults and network attacks.
[0118] The present invention is applicable to a complex situation in which a partial failure of an actuator occurs in a DC microgrid and multiple communication links between controllers of distributed power generation units are simultaneously injected with false data by attackers. BRIEF DESCRIPTION OF THE DRAWINGS
[0119] Figure 1 It is a flow chart of a method for distinguishing and detecting DC microgrid faults and network attacks according to an embodiment of the present invention;
[0120] Figure 2 It is a schematic diagram of the calculation process of a method for distinguishing and detecting DC microgrid faults and network attacks according to an embodiment of the present invention;
[0121] Figure 3 It is a schematic diagram of a DC microgrid system composed of six distributed power generation units according to an embodiment of the present invention;
[0122] Figures 4-5 It is a schematic diagram of part of the detection effect of a method for distinguishing and detecting DC microgrid faults and network attacks according to an embodiment of the present invention. DETAILED DESCRIPTION
[0123] The DC microgrid fault and network attack distinguishing detection method and system proposed by the present invention are further described in detail below in combination with the accompanying drawings and specific embodiments. According to the following description, the advantages and features of the present invention will become clearer. It should be noted that the accompanying drawings are all in a very simplified form and are not in precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.
[0124] The object of the present invention is to provide a method and system for distinguishing and detecting DC microgrid faults and network attacks, so as to solve the problem that the existing DC microgrid actuator faults and network attacks cannot be distinguished.
[0125] To achieve the above-mentioned purpose, the present invention provides a DC microgrid fault and network attack distinguishing detection method and system, including: decomposing DC microgrid actuator fault detection and DC microgrid network attack detection into network layer attack detection based on a centralized unknown input observer and physical layer actuator fault detection based on a distributed Luenberger observer, so as to distinguish the DC microgrid physical layer fault and network layer attack; and decomposing the DC microgrid actuator fault detection and DC microgrid network attack detection based on the superposition principle and the unknown input-Luenberger dual observer detection strategy; the detection strategy based on the superposition principle and the unknown input-Luenberger dual observer includes: on the basis of analyzing the structural limitations of the unknown input observer for fault detection and the undetectability of the Luenberger observer for covert attacks, combining the two observers to achieve the detection purpose by utilizing the original performance advantages of the observers, and making the limitations of the single observer itself become the entry point for the differentiated detection of faults and attacks.
[0126] Figures 1 to 5 The first embodiment of the present invention is proposed, which shows a flow chart of a method for distinguishing between faults and network attacks in a DC microgrid. The method for detecting and distinguishing between actuator faults and network attacks based on dual observers provided by the present invention aims to solve the scenario in which a DC microgrid has both actuator faults and network attacks. The proposed method can be sensitive to both actuator faults and network attacks, thereby achieving rapid detection and distinction between actuator faults and network attacks.
[0127] Based on the superposition principle and the unknown input-Luenberger dual observer detection strategy, the DC microgrid actuator fault and network attack detection problem is decomposed into a network layer attack detection problem based on a centralized unknown input observer and a physical layer actuator fault detection problem based on a distributed Luenberger observer, so as to achieve effective detection and distinction between physical layer faults and network layer attacks.
[0128] For the network layer detection strategy, taking into account the heterogeneity of communication networks and physical topologies, and the inability to fully detect attacks using only the physical coupling information between distributed generation units, a set of unknown input observers with a centralized architecture is used to obtain the state estimation information of adjacent distributed generation units by augmenting the unknown inputs as well as coupling noise and faults. Then, in order to address the limitations of the detection structure caused by the augmented unknown inputs, artificial states are introduced, and the residuals and corresponding thresholds are further designed to effectively detect possible false data injection attacks in DC microgrids.
[0129] For the physical layer detection strategy, based on the consideration of process and measurement noise caused by DC microgrid modeling mismatch and external environment, component aging, actuator failure caused by external environment influence, and covert false data injection attack injected into the controller communication channel, the distributed structure Luenberger observer is used through the physical coupling information interaction between the neighbors of DC microgrid distributed generation units to achieve the detection of physical layer failure. ∞ / H - Performance indicators and residual decomposition algorithms are used to achieve fast and effective detection of actuator faults at the physical layer of DC microgrids.
[0130] Based on the analysis of the structural limitations of the unknown input observer for fault detection and the undetectability of the Luenberger observer for covert attacks, a strategy of combining the two observers is proposed. This strategy can not only utilize the original performance advantages of the observer to achieve the detection purpose but also make the limitations of a single observer itself become the entry point for differentiated detection of faults and attacks.
[0131] like Figure 1 As shown, the specific DC microgrid fault and network attack distinguishing detection method includes five steps. Step 1: First, establish a single distributed generation unit model with actuator effectiveness loss failure under the influence of the following noise:
[0132]
[0133] where x i (t) = [V i (t),I ti (t),ε i ] T Indicates the local state, V i (t) is the voltage at the common coupling point, I ti (t) is the Buck converter current, ε i is the integrator state inside the primary controller, u i (t) = [V ti (t),V r,i ] T Represents the control input, V ti (t) is the main controller input, V r,i is the local reference voltage, y i (t) is the measurable output, represents the physical interconnection between distributed generation units, is the neighbor set of distributed generation unit i, w i (t) = I Li (t) is the external known input, I Li (t) is the load; represent the process noise and measurement noise of the system respectively, Indicates actuator failure; A ii , B i , C i , M i Respectively represent the state matrix, control matrix, observation matrix and external known input matrix, A ij represents the interconnection matrix between distributed generation unit i and distributed generation unit j, E 1i , E 2i are known fault matrices and interference matrices with appropriate dimensions; the specific parameters are given by:
[0134]
[0135] Among them, R ij Indicates line resistance, C ti Represents the parallel capacitance, L ti Filter inductor, R ti Filter resistor;
[0136] Secondly, the following actuator partial effectiveness loss fault model of IGBT / MOSFET switch in distributed generation unit is given:
[0137] f i (t) = γ i V ti (t),t≥T i f
[0138] Where T i f is the moment when the actuator of distributed generation unit i fails, γ i is the failure factor, -1<γ i ≤0, V ti (t) is the main controller input.
[0139] like Figure 2 As shown, step 2: define the following false data injection attack model that occurs on the communication link between distributed generation unit controllers:
[0140]
[0141] in, represents the information received by distributed generation unit i from its neighbors, y j (t) represents the output of the augmented system after the artificial state is introduced, Represents a potential false data injection attack The influence of j,i (t) represents the impact of network attacks on the artificial state during communication transmission, represents the false data injected into the communication link (j,i), α j,i (t) is the step function generated by finite superposition, which represents the activation signal of malicious attack;
[0142] Consider the false data injection attack that occurs on the controller communication link. a i A constant bias is cooperatively injected into the communication link where distributed generation unit i receives its neighbor information at all times, so that its impact on the system output is hidden from the Luenberger observer:
[0143]
[0144]
[0145] Among them, T i a is the moment when the network attack is injected into the distributed generation unit i to receive the neighbor information communication link, represents the set of neighbors whose communication links are attacked, represents the physical coupling matrix of the neighbors whose communication links are attacked,
[0146] Step 3: Design a network attack detection module based on unknown input observer to detect malicious network attacks on the communication link between distributed generation unit controllers in the presence of faults;
[0147] First, the neighborhood dynamic model of distributed generation unit i is reformulated as follows:
[0148]
[0149] in, represents the normalized uncertainty variable for process noise and actuator faults, represent the process noise and measurement noise of the system respectively, Indicates an actuator failure. represents the unknown input after augmentation, represents the neighbors of distributed generation unit j, contains all inputs of the neighbors of DG unit i that are unknown to the detector, is the matrix D j The column-full rank matrix of the basis of is based on It is deduced that A jj is the state matrix, C j is the observation matrix, E 1j , E2j The fault matrix and interference matrix are known to have appropriate dimensions, and the system matrix D j and noise fault coupling matrix The specific description is as follows:
[0150]
[0151] Use state enhancement technology to improve the detection capability based on unknown input observers, improve the limitation of the original unknown input observers that have no redundant detection attacks; introduce artificial state variables definition Based on the following relationship:
[0152]
[0153] Among them, x j (t) and y j (t) is the state and output of the original system before the artificial state is introduced, I represents the unit matrix, is the augmented system state, y j (t) represents the output of the system after reconstruction. The following reconstructed system dynamic model is obtained:
[0154] in, is the augmented system state, y j (t) represents the output of the system after reconstruction, yes The rearranged transpose, d j,1 (t) = u j (t) The input u is unknown to the local observer of DG unit i but known to the controller of its neighboring DG unit j. j (t) composition, It consists of unknown inputs to the local observer of DG unit i and the controller of its neighboring DG unit j. represents the normalized uncertainty variable for process noise and actuator faults, represent the process noise and measurement noise of the system respectively, Indicates an actuator failure. E 1j , E 2j The fault matrix and interference matrix are known to have appropriate dimensions, and the matrix A jj , D j , C j , N j The specific parameters are as follows:
[0155]
[0156] Design the following centralized unknown input observer group to obtain the state estimation of the neighbors of distributed generation unit i:
[0157]
[0158] Among them, z j,i (t) is the state of the unknown input observer, Represents the augmented state x j The estimate of (t), represents the unknown state observer output, represents the information received by distributed generation unit i from its neighbors, C i represents the system description matrix, G j , J j , H j is a matrix designed under the following constraints:
[0159]
[0160] Define the state estimation error: Define the residual: According to the above unknown input observer matrix, the following dynamic error model under nominal state is designed:
[0161]
[0162] in, represent the process noise and measurement noise of the system respectively, Indicates an actuator failure. Indicates an actuator failure. represents the normalized uncertainty variable for process noise and actuator faults, E j1 , E j2 are known fault and interference matrices with appropriate dimensions, is the coupling matrix of process noise and actuator faults, G j , J j , H j , Θ j It is a matrix designed under appropriate constraints;
[0163] In order to realize the network attack detection on the communication link of the distributed generation unit, the residual threshold is defined according to the dynamic error model under the nominal state as follows:
[0164]
[0165] in, σ j >0,Σj ≥I, are known positive upper bounds for process noise, measurement noise, and actuator faults, respectively, is a properly defined unknown initial state estimation error the boundaries of;
[0166] In the case of actuator failure in the DC microgrid, when the false data injection attack on the communication link is activated, in the attack detection module based on the unknown input observer, by introducing the residual decomposition algorithm, the error dynamic equation of the system is expressed as the superposition of the "nominal" and "attacked" parts, as shown below:
[0167]
[0168] in and As given in the dynamic error model under the nominal state above, the state estimation error and residual of the “attacked” part are expressed as follows:
[0169]
[0170] in, Represents a potential false data injection attack The influence of j,i (t) represents the impact of network attacks on the artificial state during communication transmission, represents the false data injected into the communication link (j,i), G j , J j , H j is a matrix designed under appropriate constraints, T i a It is the moment when a cyber attack is injected into the communication link;
[0171] According to the dynamic error model under the nominal state designed above and the appropriately defined residual threshold, the following attack detection inequality is designed:
[0172]
[0173] It represents a sufficient condition for attack detection. When this condition is not met, it means that the communication link (j, i) is subjected to a network attack.
[0174] Step 4: In order to realize actuator fault detection in the presence of malicious network attacks on the communication link, an actuator fault detection module based on Luenberger observer is designed. First, the following Luenberger observer is designed to obtain the local state estimate of distributed generation unit i:
[0175]
[0176] Among them, A ii , B i , C i , M i Represent the state matrix, control matrix, observation matrix and external known input matrix respectively, u i represents the control input, w i =I Li (t) represents the external known input, I Li (t) is the load, y i represents the local measurable output of distributed generation unit i, L i The proper design of A Li =(A ii +L i C i ) is Hurwitz stable, Indicates the impact of physical interconnection with neighbors and is defined as follows:
[0177]
[0178] Among them, A ij represents the interconnection matrix between distributed generation units i and j. When C j When it is non-strange, Use as x j Neighborhood state estimation of (t) When C j When is singular, the design of the Luenberger observer requires the neighbor state estimate of the unknown input observer in step 4
[0179] Further, based on the obtained local state estimate Define the state estimation error: Define the residual:
[0180] Design the following dynamic error model under nominal conditions:
[0181]
[0182] Among them, C i represents the observation matrix, L i The proper design of A Li =(A ii +L i C i ) is Hurwitz stable, They represent the process noise and measurement noise of the system respectively, and E 1i , E2i is a known interference matrix with appropriate dimensions;
[0183] According to the dynamic error model under the nominal state, the following time-varying actuator fault detection threshold is defined:
[0184]
[0185] in, is a properly defined unknown initial state estimation error The borders of are the known positive upper bounds of process noise and measurement noise, E 1i , E 2i is a known interference matrix with appropriate dimensions, λ i >0,Λ i ≥I;
[0186] For a given DC microgrid system, attacks and faults are additive. In order to detect actuator faults in the presence of malicious network attacks on the communication link, the following residual decomposition algorithm is introduced to decompose the system residual into three parts: "nominal", "attacked", and "faulty":
[0187]
[0188] in, and r i n (t) has been given in the dynamic error model under the above nominal state, r i a (t) and r i f (t) The part of the residual that is affected by the attack and the fault, respectively;
[0189] The state estimation error and residual of the “faulty” part are expressed as follows:
[0190]
[0191] Among them, A Li =(A ii +L i C i ) is Hurwitz stable, C i represents the observation matrix, Indicates an actuator failure. is a known fault matrix with appropriate dimensions, T i f is the moment when the actuator of distributed generation unit i fails;
[0192] The state estimation error and residual of the “attacked” part are expressed as follows:
[0193]
[0194] Among them, A Li =(A ii +L i C i ) is Hurwitz stable, C i represents the observation matrix, A ij represents the interconnection matrix between distributed generation unit i and distributed generation unit j, Represents a potential false data injection attack The influence of j,i (t) represents the impact of network attacks on the artificial state during communication transmission, represents the false data injected into the communication link (j,i), T a i It is the moment when the cyber attack is injected into the DG unit i to receive the neighbor information communication link;
[0195] Based on the modeling of the false data injection attack in step 2, the following characteristics of the residual of the "attacked" part of the Luenberger observer are obtained:
[0196] |r i a (t)|=0
[0197] The error dynamic equation of the DC microgrid system affected by both faults and attacks is reformulated as follows:
[0198]
[0199] in, and r i n (t) represent the state estimation error and residual corresponding to the “nominal” part of agent i, respectively. and r i f (t) represent the state estimation error and residual corresponding to the affected part of the “fault” of agent i, respectively;
[0200] The following RMS-based error estimation function and corresponding threshold are introduced to achieve timely and accurate fault detection:
[0201]
[0202]
[0203] Based on the definition of the residual evaluation function and the selected threshold, the following detection logic is proposed to detect the occurrence of faults;
[0204]
[0205] Step 5: Combining steps 3 and 4, in the task of distinguishing and detecting actuator faults and network attacks in the DC microgrid, the unknown input-Luenberger dual observer detection strategy is implemented. By combining two modules and using different information structures, actuator fault and network attack detection is achieved.
[0206] When there is no fault or attack, the DC microgrid maintains voltage stability, current sharing and voltage balance with the cooperation of the two-stage controller. When an actuator fault occurs, the detection logic based on the Luenberger observer changes from 0 to 1, and the detector issues a fault alarm. When a network attack occurs, the detection logic based on the unknown input observer changes from 0 to 1, and the detector issues an attack alarm. Therefore, in each distributed generation unit, the attack-fault differentiation detector composed of the unknown input-Luenberger dual observer corresponds to the following multi-module structure:
[0207] Module 1 is configured as an unknown input observer group for observing neighbor states. Based on the correct grasp of neighbor model information, the unknown input observer group is used to calculate the estimated state of each neighbor of distributed generation unit i, calculate the state estimation error and residual, and detect network attacks on the communication link between distributed generation unit i and its neighbors by comparing the residual with a predefined threshold.
[0208] Module 2 is configured as a Luenberger observer to observe the local state. Based on the physical coupling information between the systems, the Luenberger observer is used to estimate the local state for the distributed generation unit i, and the state estimation error and the residual evaluation function based on the root mean square are calculated. By comparing the residual evaluation function with the corresponding threshold, it is detected whether there is an actuator fault in the distributed generation unit i.
[0209] The present invention constructs two modules, namely, an unknown input observer and a Luenberger observer, to realize the detection and differentiation of actuator failure and network attack. In the attack detection module based on the unknown input observer, the following unknown input observer is designed to estimate the state of the neighbors of the local distributed generation unit:
[0210]
[0211] where z j,i (t) is the state of the unknown input observer, Represents the augmented state x j The estimate of (t), represents the unknown state observer output, represents the information received by distributed generation unit i from its neighbors, C i represents the observation matrix, G j , J j , H j is a matrix designed under appropriate constraints.
[0212] The unknown input observer used to estimate the state in conventional attack detection is converted into an unknown input observer group used to estimate the neighbor state under the dual influence of fault attack in the present invention. The present invention is based on the improvement of normalizing the fault and noise to make the residual insensitive to the fault, thereby achieving accurate detection of network attacks under the influence of actuator faults. At the same time, based on the improvement of introducing artificial states, the limitation of no redundant detection of attacks caused by the same number of unknown inputs and outputs after augmentation is improved.
[0213] The present invention constructs two modules, an unknown input observer and a Luenberger observer, to detect and distinguish actuator faults and network attacks. In the actuator fault detection module based on the Luenberger observer, the following residual decomposition algorithm is designed:
[0214]
[0215] Among them, ∈ i (t) and r i (t) represent the state estimation error and residual of agent i under the influence of network attack and actuator failure, respectively. and r i n (t) represent the state estimation error and residual corresponding to the “nominal” part of agent i, respectively. and r i a (t) represent the state estimation error and residual corresponding to the affected part of agent i “attacked”, respectively. and r i f (t) represent the state estimation error and residual corresponding to the affected part of the “fault” of agent i, respectively.
[0216] Based on the improvement of residual decomposition, the concealment of Luenberger observer is exploited by false data injection attack, combined with the residual evaluation function based on root mean square and the corresponding threshold, the system can accurately detect actuator faults under network attacks.
[0217] The present invention is applicable to the complex situation where a partial failure of an actuator occurs in a DC microgrid and multiple communication links between controllers of distributed generation units are simultaneously injected with false data by attackers. A DC microgrid system consisting of six distributed generation units, such as Figure 3 As shown:
[0218] Initial stage: The six generating units are disconnected from each other, and the activated primary controller makes the voltage at the common coupling point track the reference voltage to ensure voltage stability.
[0219] Nominal operation stage: The six power generation units are connected to each other through power lines and communication channels, both the main controller and the secondary controller are activated, and the consistent control objectives (voltage balance, current sharing) of the DC microgrid system are achieved based on a secure communication channel and a healthy physical layer.
[0220] Network attack injection stage: The communication link from distributed generation unit 2 to distributed generation unit 4 and the communication link from distributed generation unit 3 to distributed generation unit 4 are simultaneously attacked by false data injection. Subsequently, the attack detection module based on the unknown input observer quickly detects that the residual curve of the corresponding link exceeds the corresponding threshold curve and issues an attack alarm.
[0221] Actuator failure stage: Based on the network attack injection stage, the actuator of the IGBT / MOSFET switch in the distributed generation unit 5 has a partial loss of effectiveness failure. Subsequently, the fault detection module based on the Luenberger observer quickly detects that the residual evaluation function curve corresponding to the distributed generation unit 5 exceeds the threshold curve and issues a fault alarm. Some detection effect images are shown below Figures 4-5 As shown, Figure 4 For attack detection effect, Figure 5 This is the fault detection effect.
[0222] In summary, the above embodiments describe in detail different configurations of the DC microgrid fault and network attack distinguishing detection method and system. Of course, the present invention includes but is not limited to the configurations listed in the above embodiments. Any content that is transformed based on the configuration provided in the above embodiments belongs to the scope of protection of the present invention. Those skilled in the art can draw inferences from the contents of the above embodiments.
[0223] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0224] The above description is only a description of the preferred embodiments of the present invention, and is not intended to limit the scope of the present invention. Any changes or modifications made by a person skilled in the art in the field of the present invention based on the above disclosure shall fall within the scope of protection of the claims.
Claims
1. A method for distinguishing and detecting DC microgrid faults and network attacks, characterized in that: include: Decomposing DC microgrid actuator fault detection and DC microgrid network attack detection into network layer attack detection based on centralized unknown input observer and physical layer actuator fault detection based on distributed Luenberger observer, so as to distinguish between DC microgrid physical layer fault and network layer attack; and Based on the superposition principle and the unknown input-Luenberger dual observer detection strategy, the DC microgrid actuator fault detection and the DC microgrid network attack detection are decomposed, wherein the superposition principle and the unknown input-Luenberger dual observer detection strategy include: based on the analysis of the structural limitations of the unknown input observer for fault detection and the undetectability of the Luenberger observer for covert attacks, the two observers are combined for detection, wherein the limitations of a single observer itself are used to distinguish between faults and attacks; The method further comprises: For the network layer detection strategy, considering that the communication network and physical topology are heterogeneous, and that only the physical coupling information between distributed generation units cannot fully detect the attack, a set of unknown input observers with a centralized architecture is used to obtain the state estimation information of adjacent distributed generation units by augmenting the unknown inputs as well as coupling noise and faults. Then, in view of the detection structure limitations caused by augmenting the unknown inputs, artificial states are introduced, and residuals and corresponding thresholds are designed to effectively detect possible false data injection attacks in the DC microgrid; and For the physical layer detection strategy, based on the consideration of process and measurement noise caused by DC microgrid modeling mismatch and external environment, component aging, actuator failure caused by external environment influence, and covert false data injection attack injected into the controller communication channel, the distributed structure Luenberger observer is used through the physical coupling information interaction between the neighbors of DC microgrid distributed generation units to achieve the detection of physical layer failure. ∞ / H - Performance indicators and residual decomposition algorithms to detect actuator faults at the physical layer of DC microgrids; The method further comprises: Construct unknown input observer module and Luenberger observer module to detect and distinguish actuator failures and network attacks; In the network attack detection module based on unknown input observer, the following unknown input observer is designed as the neighbor estimation state of the local distributed generation unit: Among them, z j,i (t) is the state of the unknown input observer, Represents the augmented state x j The estimate of (t), represents the unknown state observer output, represents the information received by distributed generation unit i from its neighbors, C i represents the system description matrix, G j , J j , H j It is a matrix designed under the following constraints; The unknown input observer used to estimate the state in conventional attack detection is converted into an unknown input observer group used to estimate the neighbor state under the dual influence of fault attack; based on the improvement of normalizing fault and noise, the residual is made insensitive to fault, so as to accurately detect network attacks under the influence of actuator fault, and at the same time, the improvement of artificial state is introduced to improve the limitation of no redundant detection attack due to the same number of unknown input and output after augmentation; In the actuator fault detection module based on Luenberger observer, the following residual decomposition algorithm is designed: Among them, ∈ i (t) and r i (t) represent the state estimation error and residual of agent i under the influence of network attack and actuator failure, respectively. and They represent the state estimation error and residual corresponding to the "nominal" part of agent i, respectively. and r i a (t) represent the state estimation error and residual corresponding to the affected part of agent i "attacked", and r i f (t) respectively represent the state estimation error and residual corresponding to the affected part of the "fault" of agent i; Based on the improvement of residual decomposition, the concealment of Luenberger observer is exploited by false data injection attack, combined with the residual evaluation function based on root mean square and the corresponding threshold, the system is used to detect actuator faults under network attacks.
2. The DC microgrid fault and network attack distinguishing detection method according to claim 1, characterized in that: Also includes: The DC microgrid system consists of multiple distributed generation units; In the initial stage: multiple power generation units are disconnected from each other, and the activated main controller makes the voltage at the common coupling point track the reference voltage to ensure voltage stability; In the nominal operation stage: multiple power generation units are connected to each other through power lines and communication channels, both the main controller and the secondary controller are activated, and the voltage balance and current sharing of the DC microgrid system are achieved based on a secure communication channel and a healthy physical layer; In the network attack injection phase: a distributed generation unit sends information to a neighboring distributed generation unit through one communication link. If two or more communication links of the same distributed generation unit are attacked by false data injection at the same time, the network attack detection module based on the unknown input observer in each distributed generation unit quickly detects whether the residual curve of the corresponding communication link exceeds the corresponding threshold curve, and issues an attack alarm. At the actuator failure stage: If the actuator of the IGBT / MOSFET switch in the distributed generation unit fails based on the network attack injection stage, the actuator fault detection module based on the Luenberger observer in each distributed generation unit quickly detects whether the corresponding residual evaluation function curve exceeds the threshold curve and issues a fault alarm.
3. The DC microgrid fault and network attack distinguishing detection method according to claim 1, characterized in that: Also includes: Step 1: First, establish a single distributed generation unit model with actuator effectiveness loss failure under the influence of noise as follows: where x i (t) = [V i (t),I ti (t),ε i ] T Indicates the local state, V i (t) is the common coupling point voltage, I ti (t) is the Buck converter current, ε i is the integrator state inside the main controller, u i (t)=[V ti (t),V r,i ] T Represents the control input, V ti (t) is the main controller input, V r,i is the local reference voltage, y i (t) is the measurable output, represents the physical interconnection between distributed generation units, is the neighbor set of distributed generation unit i, w i (t) = I Li (t) is the external known input, I Li (t) is the load; represent the process noise and measurement noise of the system respectively, Indicates actuator failure; A ii , B i , C i , M i Respectively represent the state matrix, control matrix, observation matrix and external known input matrix, A ij represents the interconnection matrix between distributed generation units i and j, B fi , E 1i , E 2i are known fault matrices and interference matrices with appropriate dimensions; the specific parameters are given by: Among them, R ij Indicates line resistance, C ti Represents the parallel capacitance, L ti Filter inductor, R ti Filter resistor; Secondly, the following actuator partial failure fault model of IGBT / MOSFET switch in distributed generation unit is given: f i (t)=γ i V ti (t),t≥T i f Among them, T i f is the moment when the actuator of distributed generation unit i fails, γ i is the failure factor, -1<γ i ≤0, V ti (t) is the main controller input.
4. The method for distinguishing and detecting DC microgrid faults and network attacks according to claim 3, characterized in that: Also includes: Step 2: Define the following false data injection attack model that occurs on the communication link between distributed generation unit controllers: in, represents the information received by distributed generation unit i from its neighbors, y j (t) represents the output of the augmented system after the artificial state is introduced, Represents a potential false data injection attack The influence of j,i (t) represents the impact of network attacks on the artificial state during communication transmission, represents the false data injected into the communication link (j,i), α j,i (t) is the step function generated by finite superposition, which represents the activation signal of malicious attack; Consider the false data injection attack that occurs on the controller communication link. a i A constant bias is cooperatively injected into the communication link where distributed generation unit i receives its neighbor information at all times, so that its impact on the system output is hidden from the Luenberger observer: Among them, T a i is the moment when the network attack is injected into the distributed generation unit i to receive the neighbor information communication link, represents the set of neighbors whose communication links are attacked, represents the physical coupling matrix of the neighbors whose communication links are attacked, 5. The method for distinguishing and detecting DC microgrid faults and network attacks according to claim 4, characterized in that: Also includes: Step 3: Design a network attack detection module based on unknown input observer to detect malicious network attacks on the communication link between distributed generation unit controllers in the presence of faults; First, the neighborhood dynamic model of distributed generation unit i is reformulated as follows: in, represents the normalized uncertainty variable for process noise and actuator faults, represent the process noise and measurement noise of the system respectively, Indicates an actuator failure. represents the unknown input after augmentation, represents the neighbors of distributed generation unit j, contains all inputs of the neighbors of DG unit i that are unknown to the detector, is the matrix D j The column-full rank matrix of the basis of is based on It is deduced that A jj is the state matrix, C j is the observation matrix, E 1j , E 2j The fault matrix and interference matrix are known to have appropriate dimensions, and the system matrix D j and noise fault coupling matrix The specific description is as follows: Use state enhancement technology to improve the detection ability of unknown input observers and improve the limitation of the original unknown input observers that there is no redundant detection attack; introduce artificial state variables definition Based on the following relationship: Among them, x j (t) and y j (t) is the state and output of the original system before the artificial state is introduced, I represents the unit matrix, is the augmented system state, y j (t) represents the output of the system after reconstruction, and the following reconstructed system dynamic model is obtained: in, is the augmented system state, y j (t) represents the output of the system after reconstruction, yes The rearranged transpose, d j,1 (t) = u j (t) The input u is unknown to the local observer of DG unit i but known to the controller of its neighboring DG unit j. j (t) composition, It consists of unknown inputs to the local observer of DG unit i and the controller of its neighboring DG unit j. represents the normalized uncertainty variable for process noise and actuator faults, represent the process noise and measurement noise of the system respectively, Indicates an actuator failure. E 1j , E 2j The fault matrix and interference matrix are known to have appropriate dimensions, and the matrix A jj , D j , C j , N j The specific parameters are as follows: Design the following centralized unknown input observer group to obtain the state estimation of the neighbors of distributed generation unit i: Among them, z j,i (t) is the state of the unknown input observer, Represents the augmented state x j The estimate of (t), represents the unknown state observer output, represents the information received by distributed generation unit i from its neighbors, C i represents the observation matrix, G j , J j , H j is a matrix designed under the following constraints: Define the state estimation error: Define the residual: According to the above unknown input observer matrix, the following dynamic error model under nominal state is designed: in, represents the normalized uncertainty variable for process noise and actuator faults, represent the process noise and measurement noise of the system respectively, Indicates an actuator failure. is the coupling matrix of process noise and actuator faults, E 1j , E 2j are known fault and interference matrices with appropriate dimensions, C j represents the observation matrix, G j , J j , H j , Θ j It is a matrix designed under appropriate constraints; In order to realize the network attack detection on the communication link of the distributed generation unit, the residual threshold is defined according to the dynamic error model under the nominal state as follows: in, σ j >0,Σ j ≥I, are known positive upper bounds for process noise, measurement noise, and actuator faults, respectively, is a properly defined unknown initial state estimation error the boundaries of; In the case of actuator failure in the DC microgrid, when the false data injection attack on the communication link is activated, in the attack detection module based on the unknown input observer, by introducing the residual decomposition algorithm, the error dynamic equation of the system is expressed as the superposition of the "nominal" and "attacked" parts, as shown below: in and As given in the dynamic error model under the nominal state above, the state estimation error and residual of the "attacked" part are expressed as follows: in, Represents a potential false data injection attack The influence of j,i (t) represents the impact of network attacks on the artificial state during communication transmission, represents the false data injected into the communication link (j,i), G j , J j , H j is a matrix designed under appropriate constraints, T i a It is the moment when a cyber attack is injected into the communication link; According to the dynamic error model under the nominal state designed above and the appropriately defined residual threshold, the following attack detection inequality is designed: It represents a sufficient condition for attack detection. When this condition is not met, it means that the communication link (j, i) is subjected to a network attack.
6. The method for distinguishing and detecting DC microgrid faults and network attacks according to claim 5, characterized in that: Also includes: Step 4: In order to realize actuator fault detection in the presence of malicious network attacks on the communication link, an actuator fault detection module based on Luenberger observer is designed. First, the following Luenberger observer is designed to obtain the local state estimate of distributed generation unit i: Among them, A ii , B i , C i , M i Represent the state matrix, control matrix, observation matrix and external known input matrix respectively, u i represents the control input, w i =I Li (t) represents the external known input, I Li (t) is the load, y i represents the local measurable output of distributed generation unit i, L i The proper design of A Li =(A ii +L i C i ) is Hurwitz stable, Indicates the impact of physical interconnection with neighbors and is defined as follows: Among them, A ij represents the interconnection matrix between distributed generation units i and j. When C j When it is non-strange, Use as x j Neighborhood state estimation of (t) When C j When is singular, the design of the Luenberger observer requires the neighbor state estimate of the unknown input observer in step 4 Further, based on the obtained local state estimate Define the state estimation error: Define the residual: Design the following dynamic error model under nominal conditions: Among them, C i represents the observation matrix, L i The proper design of A Li =(A ii +L i C i ) is Hurwitz stable, They represent the process noise and measurement noise of the system respectively, and E 1i , E 2i is a known interference matrix with appropriate dimensions; According to the dynamic error model under the nominal state, the following time-varying actuator fault detection threshold is defined: in, is a properly defined unknown initial state estimation error The boundary of i >0,Λ i ≥I, are the known positive upper bounds of process noise and measurement noise, E 1i , E 2i is a known interference matrix with appropriate dimensions; For a given DC microgrid system, attacks and faults are additive. In order to detect actuator faults in the presence of malicious network attacks on the communication link, the following residual decomposition algorithm is introduced to decompose the system residual into three parts: "nominal", "attacked", and "faulty": in, and It has been given in the dynamic error model under the above nominal state, r i a (t) and r i f (t) The part of the residual that is affected by the attack and the fault, respectively; The state estimation error and residual of the "faulty" part are expressed as follows: Among them, A Li =(A ii +L i C i ) is Hurwitz stable, C i represents the observation matrix, Indicates actuator failure, B fi is a known fault matrix with appropriate dimensions, T i f is the moment when the actuator of distributed generation unit i fails; The state estimation error and residual of the "attacked" part are expressed as follows: Among them, A Li =(A ii +L i C i ) is Hurwitz stable, C i represents the observation matrix, A ij represents the interconnection matrix between distributed generation unit i and distributed generation unit j, Represents a potential false data injection attack The influence of j,i (t) represents the impact of network attacks on the artificial state during communication transmission, represents the false data injected into the communication link (j,i), T a i It is the moment when the cyber attack is injected into the DG unit i to receive the neighbor information communication link; Based on the modeling of the false data injection attack in step 2, the following characteristics of the residual of the "attacked" part of the Luenberger observer are obtained: |r i a (t)|=0 The error dynamic equation of the DC microgrid system affected by both faults and attacks is reformulated as follows: Among them, ∈ i (t) and r i (t) represent the state estimation error and residual of agent i under the influence of network attack and actuator failure, respectively. and They represent the state estimation error and residual corresponding to the "nominal" part of agent i, respectively. and r i f (t) respectively represent the state estimation error and residual corresponding to the affected part of the "fault" of agent i; The following RMS-based error estimation function and corresponding threshold are introduced to achieve timely and accurate fault detection: Based on the definition of the residual evaluation function and the selected threshold, the following detection logic is proposed to detect the occurrence of faults; 7. The method for distinguishing and detecting DC microgrid faults and network attacks according to claim 6, characterized in that: Also includes: Step 5: Combining steps 3 and 4, in the task of distinguishing and detecting actuator faults and network attacks in the DC microgrid, the unknown input-Luenberger dual observer detection strategy is implemented. By combining two modules and using different information structures, actuator fault and network attack detection is achieved. When there is no fault or attack, the DC microgrid maintains voltage stability, current sharing and voltage balance with the cooperation of the two-stage controller. When an actuator fault occurs, the detection logic based on the Luenberger observer changes from 0 to 1, and the detector issues a fault alarm. When a network attack occurs, the detection logic based on the unknown input observer changes from 0 to 1, and the detector issues an attack alarm. Therefore, in each distributed generation unit, the attack-fault differentiation detector composed of the unknown input-Luenberger dual observer corresponds to the following multi-module structure: Module 1 is configured as an unknown input observer group for observing neighbor states. Based on the correct grasp of neighbor model information, the unknown input observer group is used to calculate the estimated state of each neighbor of distributed generation unit i, calculate the state estimation error and residual, and detect network attacks on the communication link between distributed generation unit i and its neighbors by comparing the residual with a predefined threshold. Module 2 is configured as a Luenberger observer to observe the local state. Based on the physical coupling information between the systems, the Luenberger observer is used to estimate the local state for the distributed generation unit i, and the state estimation error and the residual evaluation function based on the root mean square are calculated. By comparing the residual evaluation function with the corresponding threshold, it is detected whether there is an actuator fault in the distributed generation unit i.
8. A DC microgrid fault and network attack differentiation detection system, which uses the method according to any one of claims 1 to 7, characterized in that: include: The dual observer detection strategy module is configured to be based on the superposition principle and the unknown input-Luenberger dual observer detection strategy, which decomposes the DC microgrid actuator fault detection and the DC microgrid network attack detection into the network layer attack detection based on the centralized unknown input observer and the physical layer actuator fault detection based on the distributed Luenberger observer, so as to distinguish the DC microgrid physical layer fault from the network layer attack; an unknown input observer module, configured to serve as a network attack detection module, and to perform network layer attack detection based on a centralized unknown input observer; and A Luenberger observer module is configured to serve as an actuator fault detection module for performing physical layer actuator fault detection based on a distributed Luenberger observer; The detection strategy based on the superposition principle and unknown input-Luenberger dual observer includes: on the basis of analyzing the structural limitations of the unknown input observer for fault detection and the undetectability of the Luenberger observer for covert attacks, the two observers are combined to utilize the original performance advantages of the observers to achieve the detection purpose, and the limitations of a single observer itself become the entry point for differentiated detection of faults and attacks.
