Distribution network state perception and control method and system based on industrial Internet of Things

Abstract

The present invention provides a distribution network state perception and control method and system based on the Industrial Internet of Things, which relates to the field of distribution network management technology. The method includes collecting data through IoT sensor nodes, constructing a dynamic coupling matrix, configuring chaotic oscillator parameters to amplify disturbance signals, calculating stability indexes to identify weak nodes, establishing antigen-antibody mapping and optimizing antibody populations to obtain fault response strategies, and ultimately matching the optimal control solution and issuing control instructions. This method achieves accurate perception and efficient control of the distribution network's operating state, thereby improving the safety, stability, and fault response capabilities of the distribution network.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network management, and in particular to a distribution network state perception and control method and system based on industrial Internet of Things. Background Art

[0002] With the advancement of smart grid construction, the safe and stable operation of distribution networks, as the terminal link of the power system, is crucial for ensuring power reliability. Traditional distribution network monitoring systems primarily rely on SCADA systems to collect data from key nodes. However, with the influx of new loads such as distributed energy resources and electric vehicles, the operating environment of distribution networks is becoming increasingly complex, placing higher demands on state perception and control. The development of Industrial Internet of Things (IIoT) technology has provided new technical means for distribution network state monitoring. By deploying a large number of sensor nodes, comprehensive perception and real-time monitoring of the distribution network's operating status can be achieved.

[0003] Currently, distribution network state perception and control technologies primarily utilize centralized data collection and processing, transmitting collected data to a master station system via a dedicated communication network for analysis and processing. Furthermore, numerous studies are leveraging artificial intelligence and big data analysis techniques to assess and predict distribution network operating conditions and implement control strategies based on pre-defined rules. Furthermore, fault diagnosis and treatment primarily rely on expert systems and model matching methods to identify fault characteristics and generate treatment solutions.

[0004] However, existing technologies still have shortcomings in practical applications. Traditional centralized data processing architectures are difficult to cope with the highly dynamic characteristics of distribution networks and cannot effectively capture the dynamic coupling relationship between nodes, resulting in insufficient perception of the overall state of the system, especially in the identification of key nodes during the disturbance propagation process. Existing fault diagnosis methods are mostly based on static rules or fixed models and lack adaptive learning capabilities. When faced with new fault modes caused by the connection of new equipment and changes in load characteristics in the distribution network, the diagnostic accuracy is low and it is impossible to respond to system changes in a timely and effective manner. The generation and execution of control strategies lack intelligent coordination mechanisms, and the control scheme cannot be adjusted in real time according to dynamic changes in network topology and node status. This leads to poor control effects in complex fault scenarios and makes it difficult to achieve self-healing and intelligent operation of the distribution network. Summary of the Invention

[0005] The embodiments of the present invention provide a distribution network state perception and control method and system based on industrial Internet of Things, which can solve the problems in the prior art.

[0006] A first aspect of an embodiment of the present invention provides a method for sensing and controlling a distribution network state based on the industrial Internet of Things, comprising:

[0007] The IoT sensor nodes deployed in the distribution network collect data in real time and transmit the data to the edge computing layer based on the IoT communication protocol to obtain a real-time monitoring data set;

[0008] Based on the real-time monitoring data set, the information entropy correlation between nodes is calculated, and a dynamic coupling matrix is ​​constructed. By iteratively calculating the direction and intensity of information flow between nodes, a group emergence effect is formed, and the correlation intensity matrix of the network topology is obtained;

[0009] According to the correlation strength matrix, the chaotic oscillator parameters are configured for each node. The disturbance signal is amplified through the phase synchronization process between the oscillators. The oscillation characteristics of adjacent nodes are compared to obtain the disturbance eigenvector.

[0010] Based on the disturbance eigenvector, the stability index is calculated and the temporal evolution characteristics are determined. Combined with the node density distribution and cascading failure risk assessment, the weak node set in the network is identified.

[0011] For the weak node set, fault characteristics are extracted to establish antigen-antibody mapping. The antibody population is optimized through affinity-based clonal mutation operations, and the memory library is optimized. The fault response strategy set is obtained based on the memory library screening.

[0012] The optimal control solution is matched from the fault response strategy set, and the corresponding control instructions are sent to the controllable devices through the Internet of Things communication protocol.

[0013] In an optional embodiment, based on the real-time monitoring data set, the information entropy correlation between nodes is calculated, and a dynamic coupling matrix is ​​constructed. By iteratively calculating the direction and intensity of information flow between nodes, a group emergence effect is formed, and the obtained network topology correlation intensity matrix includes:

[0014] The real-time monitoring data set is divided into time windows, and the data sequence within the time window is segmented into equal intervals. The statistical frequency is used to obtain the probability distribution. The information entropy of a single node and the mutual information entropy of node pairs are calculated. Based on the maximum mutual information entropy and information entropy, the information entropy correlation is determined.

[0015] Initializing a dynamic coupling matrix using the information entropy correlation, wherein the non-diagonal elements of the dynamic coupling matrix are determined by the information entropy correlation of the corresponding node pairs, the diagonal elements are set to zero, and are updated in each time window;

[0016] Based on the iterative calculation of the dynamic coupling matrix, the state increment of the node is obtained by multiplying the current state of the node with the dynamic coupling matrix. The direction component of the state increment corresponds to the direction of information flow, and the amplitude component corresponds to the flow intensity. The state increment is superimposed on the current state of the node to complete a round of iteration;

[0017] After multiple rounds of iterative calculations, when local synchronization occurs in the node group, a group emergence effect is formed. The final stable state of the node is extracted, and the state similarity of the node pair is calculated and weightedly fused with the dynamic coupling matrix to obtain the correlation strength matrix.

[0018] In an optional embodiment, chaotic oscillator parameters are configured for each node according to the correlation strength matrix, the disturbance signal is amplified through the phase synchronization process between the oscillators, and the oscillation characteristics of adjacent nodes are compared to obtain the disturbance feature vector including:

[0019] Calculating the diagonal elements, maximum element values, and average values ​​of the correlation strength matrix, determining a first vibrator parameter based on the product of the diagonal elements and a preset first adjustment coefficient, determining a second vibrator parameter based on the product of the maximum element value and a preset second adjustment coefficient, and determining a third vibrator parameter based on the product of the average value and a preset third adjustment coefficient;

[0020] Constructing a chaotic oscillator dynamics equation for each sensor according to the first oscillator parameter, the second oscillator parameter, and the third oscillator parameter, and solving the chaotic oscillator dynamics equation to obtain a chaotic oscillator state value;

[0021] According to the element values ​​of the correlation strength matrix, the chaotic oscillator dynamic equation is expanded into a coupling equation between sensors. The chaotic oscillator phase value is obtained by solving the coupling equation through a multi-level phase extraction method to determine the response sensitivity of the chaotic oscillator to the disturbance signal.

[0022] The oscillation difference of the chaotic oscillators of adjacent sensors is calculated by weighting the difference between the chaotic oscillator state value and the chaotic oscillator phase value;

[0023] The oscillation differences between all adjacent sensors are combined in the order of sensor numbers to obtain the disturbance characteristic vector of the distribution network.

[0024] In an optional embodiment, based on the element values ​​of the correlation strength matrix, the chaotic oscillator dynamics equation is expanded into a coupling equation between sensors, the chaotic oscillator phase value is obtained by solving the coupling equation using a multi-level phase extraction method, and determining the response sensitivity of the chaotic oscillator to the disturbance signal includes:

[0025] Substituting the element values ​​of the correlation strength matrix into the chaotic oscillator dynamics equation to establish the coupling equation between sensors;

[0026] performing a Hilbert-Huang transform on the solution of the coupling equation to decompose the sensor response signal into a superposition of multiple eigenmode functions and residual terms;

[0027] Performing a continuous wavelet transform on each of the eigenmode functions, calculating wavelet coefficients according to a preset wavelet basis function, performing a Hilbert transform on each eigenmode function based on the wavelet coefficients, obtaining a real component and an imaginary component of each eigenmode function, calculating an inverse tangent value of the real component and the imaginary component, and superimposing the product of the inverse tangent value of each eigenmode function and the corresponding weight coefficient to obtain a chaotic oscillator phase value of the sensor;

[0028] In a preset observation time window, the absolute value of the derivative of the chaotic oscillator phase value with respect to time is calculated, the absolute value is integrated and normalized in the observation time window, and the response sensitivity of the chaotic oscillator to the disturbance signal is determined.

[0029] In an optional embodiment, based on the disturbance eigenvector, calculating the stability index and determining the time series evolution characteristics, combined with the node density distribution and cascading failure risk assessment, identifying the weak node set in the network includes:

[0030] Calculate the Euclidean norm of the disturbance eigenvector in the time dimension, and calculate the initial stability index of the node based on the logarithmic change rate of the Euclidean norm;

[0031] Record the time series changes of the initial stability index, construct the node stability matrix, and obtain the node fluctuation value by calculating the standard deviation of each node index in the node stability matrix;

[0032] Based on the preset stability threshold, the number of times when the index of each node exceeds the stability threshold is counted, and the number of times is divided by the observation duration to obtain the probability value of node instability;

[0033] The weight coefficients are set for the node instability probability value, node fluctuation value, and the mean of the node index, and the node comprehensive vulnerability value is obtained by weighted summation.

[0034] Based on the node's comprehensive vulnerability value, a node density distribution matrix is ​​constructed. The local density of the node is obtained by calculating the average of the comprehensive vulnerability values ​​of each node and a preset number of adjacent nodes. The density peak node is identified based on the temporal evolution characteristics of the node local density. The attribution coefficients of other nodes are calculated with the density peak node as the center. Nodes with attribution coefficients greater than the preset attribution threshold are divided into a high-risk node subset.

[0035] The stability transfer coefficient between adjacent nodes in the high-risk node subset is calculated, and the node cascading failure risk value is evaluated based on the stability transfer coefficient. The high-risk sensor node subset is screened according to the node cascading failure risk value, and the weak node set is output.

[0036] In an optional embodiment, for a set of weak nodes, fault features are extracted to establish an antigen-antibody mapping, and the antibody population is optimized through affinity-based clonal mutation operations, thereby optimizing the memory library. The fault response strategy set obtained based on the memory library screening includes:

[0037] The fault occurrence time, fault duration, and fault impact range of each node in the weak node set are constructed into a fault feature vector, and the fault feature vector is normalized by maximum and minimum values ​​to obtain a standard fault feature vector;

[0038] Using the standard fault feature vector as an antigen, generating multiple antibodies as an initial antibody population, wherein the antibodies correspond to fault response strategies;

[0039] constructing an affinity function based on the Euclidean distance between the standard fault feature vector and the antibody, wherein the affinity function includes a feature weight coefficient for balancing the importance of different fault features;

[0040] Calculating the affinity value of each antibody in the antibody population according to the affinity function, cloning and amplifying the antibodies according to the affinity value, wherein the number of clones amplified is positively correlated with the affinity value according to a preset positive proportional value, to obtain amplified antibodies;

[0041] Performing a direction-guided Gaussian mutation operation on the amplified antibodies, determining the mutation direction based on neighboring antibody information and historical evolution information, and obtaining mutant antibodies;

[0042] Construct a memory library to store high-affinity antibodies according to a preset affinity threshold. When the affinity value of the variant antibody is greater than the affinity value of the antibody in the memory library, the variant antibody is used to replace it.

[0043] Generate a collection of failure response strategies based on antibodies in the memory bank.

[0044] In an optional embodiment, performing a direction-guided Gaussian mutation operation on the amplified antibodies, determining the mutation direction based on neighboring antibody information and historical evolution information, and obtaining the mutant antibodies includes:

[0045] Performing a mutation operation on the amplified antibodies, and calculating the mutation asynchrony length of the amplified antibodies according to the affinity value of each amplified antibody using an exponential decay function, wherein the mutation asynchrony length is inversely correlated with the affinity value according to a preset inverse proportional value;

[0046] For each amplified antibody, based on the Euclidean distance between the amplified antibody and other amplified antibodies, the amplified antibodies whose distance is less than a preset nearest neighbor threshold are selected to construct a nearest neighbor antibody set;

[0047] Determining a weight coefficient according to the affinity value of each neighboring antibody in the neighboring antibody set, and performing a weighted summation on the product of the difference between the neighboring antibody and the current amplified antibody and the weight coefficient to obtain a local search direction vector;

[0048] The best antibody in each generation of the population is recorded, and the difference between the best antibodies of two adjacent generations and the product of the preset time decay coefficient are weighted and summed to obtain the historical evolution direction vector;

[0049] Linearly combining the local search direction vector and the historical evolution direction vector according to a preset combination weight coefficient to obtain a comprehensive direction vector, calculating the product of the comprehensive direction vector and the variable step length, determining a directional variation component, and adding the directional variation component to the random Gaussian variation component to obtain a total variation;

[0050] The total variation amount is added to the current amplified antibody to obtain a variant antibody.

[0051] A second aspect of an embodiment of the present invention provides a distribution network state perception and control system based on the industrial Internet of Things, including:

[0052] The first unit is used to collect data in real time through IoT sensor nodes deployed in the distribution network, and transmit the data to the edge computing layer based on the IoT communication protocol to obtain a real-time monitoring data set;

[0053] The second unit is used to calculate the information entropy correlation between nodes based on the real-time monitoring data set, construct a dynamic coupling matrix, and iteratively calculate the direction and intensity of information flow between nodes to form a group emergence effect and obtain the correlation intensity matrix of the network topology;

[0054] The third unit is used to configure the chaotic oscillator parameters for each node according to the correlation strength matrix, amplify the disturbance signal through the phase synchronization process between the oscillators, compare the oscillation characteristics of adjacent nodes, and obtain the disturbance eigenvector;

[0055] The fourth unit is used to calculate the stability index and determine the time-series evolution characteristics based on the disturbance eigenvector, and identify the weak node set in the network by combining the node density distribution and cascading failure risk assessment;

[0056] The sixth unit is used to extract fault characteristics from weak node sets, establish antigen-antibody mapping, optimize the antibody population through affinity-based clonal mutation operations, optimize the memory library, and obtain a set of fault response strategies based on the memory library screening;

[0057] The seventh unit is used to match the optimal control solution from the fault response strategy set and send the corresponding control instructions to the controllable devices through the Internet of Things communication protocol.

[0058] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0059] processor;

[0060] a memory for storing processor-executable instructions;

[0061] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0062] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0063] In an embodiment of the present invention, data is collected in real time through IoT sensor nodes, a dynamic coupling matrix is ​​constructed based on information entropy correlation, and chaotic oscillators are used to amplify disturbance signals, thereby achieving accurate perception of the distribution network state and early identification of anomalies, effectively improving the accuracy and real-time performance of distribution network operation status monitoring; an intelligent decision-making mechanism based on antigen-antibody mapping is introduced, antibody populations are optimized through cloning mutation operations, and a memory library is established to screen the optimal response strategy, thereby achieving adaptive processing and intelligent control of distribution network faults and improving the system's ability and efficiency in responding to complex faults; a complete closed-loop system from perception to decision-making to control is constructed, and accurate issuance of control instructions is achieved through the IoT communication protocol, thereby enhancing the self-healing ability and operational reliability of the distribution network, reducing the risk of cascading failures, extending the service life of equipment, optimizing grid resource allocation, and significantly improving the intelligent operation and maintenance level and power supply quality of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 This is a flow chart of a method for sensing and controlling distribution network status based on the Industrial Internet of Things according to an embodiment of the present invention;

[0065] Figure 2 Schematic diagram for performance comparison between the chaotic oscillator method and the traditional method;

[0066] Figure 3 Flowchart of fault response strategy based on artificial immune algorithm. DETAILED DESCRIPTION

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0068] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0069] Figure 1 FIG is a flow chart of a method for sensing and controlling the state of a distribution network based on the industrial Internet of Things according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0070] The IoT sensor nodes deployed in the distribution network collect data in real time and transmit the data to the edge computing layer based on the IoT communication protocol to obtain a real-time monitoring data set;

[0071] Based on the real-time monitoring data set, the information entropy correlation between nodes is calculated, and a dynamic coupling matrix is ​​constructed. By iteratively calculating the direction and intensity of information flow between nodes, a group emergence effect is formed, and the correlation intensity matrix of the network topology is obtained;

[0072] According to the correlation strength matrix, the chaotic oscillator parameters are configured for each node. The disturbance signal is amplified through the phase synchronization process between the oscillators. The oscillation characteristics of adjacent nodes are compared to obtain the disturbance eigenvector.

[0073] Based on the disturbance eigenvector, the stability index is calculated and the temporal evolution characteristics are determined. Combined with the node density distribution and cascading failure risk assessment, the weak node set in the network is identified.

[0074] For the weak node set, fault characteristics are extracted to establish antigen-antibody mapping. The antibody population is optimized through affinity-based clonal mutation operations, and the memory library is optimized. The fault response strategy set is obtained based on the memory library screening.

[0075] The optimal control solution is matched from the fault response strategy set, and the corresponding control instructions are sent to the controllable devices through the Internet of Things communication protocol.

[0076] In an optional embodiment, based on the real-time monitoring data set, the information entropy correlation between nodes is calculated, and a dynamic coupling matrix is ​​constructed. By iteratively calculating the direction and intensity of information flow between nodes, a group emergence effect is formed, and the correlation intensity matrix of the network topology is obtained, including:

[0077] The real-time monitoring data set is divided into time windows, and the data sequence within the time window is segmented into equal intervals. The statistical frequency is used to obtain the probability distribution. The information entropy of a single node and the mutual information entropy of node pairs are calculated. Based on the maximum mutual information entropy and information entropy, the information entropy correlation is determined.

[0078] Initializing a dynamic coupling matrix using the information entropy correlation, wherein the non-diagonal elements of the dynamic coupling matrix are determined by the information entropy correlation of the corresponding node pairs, the diagonal elements are set to zero, and are updated in each time window;

[0079] Based on the iterative calculation of the dynamic coupling matrix, the state increment of the node is obtained by multiplying the current state of the node with the dynamic coupling matrix. The direction component of the state increment corresponds to the direction of information flow, and the amplitude component corresponds to the flow intensity. The state increment is superimposed on the current state of the node to complete a round of iteration;

[0080] After multiple rounds of iterative calculations, when local synchronization occurs in the node group, a group emergence effect is formed. The final stable state of the node is extracted, and the state similarity of the node pair is calculated and weightedly fused with the dynamic coupling matrix to obtain the correlation strength matrix.

[0081] In one specific embodiment, the acquired real-time monitoring data set is divided into time windows. For example, 72 consecutive hours of monitoring data from 10 sensor nodes on an industrial production line are divided into 72 time windows per hour. Within each time window, the data sequence is segmented into equally spaced segments, such as 12 segments of 5 minutes each per hour. Statistics are collected for the data within each segment, and the frequency of occurrence of each value range is calculated to obtain a probability distribution. Specifically, the data range is divided into 10 intervals, and the number of data points in each interval is counted and divided by the total number of data points to obtain a probability value.

[0082] Based on the obtained probability distribution, the information entropy of a single node is calculated. For example, the data distribution probability of node A within a certain time window is [0.12, 0.08, 0.15, 0.09, 0.11, 0.13, 0.07, 0.09, 0.08, 0.08]. The information entropy is calculated as 2.25 by taking the negative sum of the probabilities multiplied by their logarithms. The mutual information entropy of the node pair is also calculated. For example, the joint probability distribution matrix of nodes A and B is a 10×10 matrix, where the element values ​​represent the probability that A and B fall within the corresponding interval. The mutual information entropy calculated by combining the joint probability and the marginal probability is 0.68. The information entropy correlation is calculated based on the information entropy of a single node and the mutual information entropy of the node pair. The information entropy correlation is equal to the mutual information entropy divided by the maximum information entropy of the two nodes involved in the calculation. For example, the information entropy correlation of nodes A and B is 0.68 divided by max(2.25, 2.10), which equals 0.302.

[0083] The calculated information entropy correlation is used to initialize the dynamic coupling matrix. For a 10-node network, a 10×10 matrix is ​​constructed, where the off-diagonal elements Mij represent the information entropy correlation between nodes i and j, and the diagonal elements are set to zero. In the initial dynamic coupling matrix, for example, M12 = 0.302 indicates that the coupling strength between nodes 1 and 2 is 0.302. This matrix is ​​updated within each time window to reflect the dynamic changes in the relationships between nodes. In the second time window, if the information entropy correlation between nodes 1 and 2 reaches 0.325, M12 is updated to 0.325.

[0084] Based on the dynamic coupling matrix, an iterative calculation is performed. Each node's initial state is set to a random value, such as the node state vector [0.45, 0.67, 0.23, 0.89, 0.34, 0.56, 0.78, 0.12, 0.90, 0.43]. In each iteration, the node's state increment is calculated by multiplying the node's current state with the dynamic coupling matrix. For example, for node 1, its state increment is the product of the first row element and the state vector, resulting in a calculated state increment of 0.09. The direction component of the state increment represents the direction of information flow; a positive value indicates that information is flowing into the node; the amplitude component represents the flow intensity; a larger value indicates a stronger flow. The state increment is added to the node's current state, and the new state of node 1 is 0.45 + 0.09 = 0.54, completing one iteration.

[0085] After multiple rounds of iterative calculations, for example, after 100 iterations, the node group begins to show local synchronization, forming a group emergence effect. In the industrial production line example, the state values ​​of the three nodes related to energy consumption converged after the 78th iteration, reaching [0.73, 0.72, 0.74], indicating that these nodes formed a functional subgroup. The final stable states of the nodes were extracted. For example, the state vector after iterative convergence was [0.73, 0.72, 0.74, 0.41, 0.42, 0.40, 0.65, 0.66, 0.67, 0.66]. The state similarity of each pair of nodes was calculated. For example, the state similarity between nodes 1 and 2 was 1 minus the absolute value of the state difference, that is, 1 - |0.73 - 0.72| = 0.99. The state similarity was weightedly fused with the dynamic coupling matrix, with weights of 0.6 and 0.4, to obtain the final correlation strength matrix. For example, the association strength between node 1 and node 2 is 0.6×0.99+0.4×0.302=0.715.

[0086] In this embodiment, the information correlation between nodes is calculated by using information entropy and mutual information entropy, so that the coupling relationship has more physical and statistical significance, and the model's ability to express the correlation of the real system is improved; the coupling matrix dynamically updated within the time window can capture the system state changes in real time, and improve the modeling accuracy of the system's dynamic behavior; the information propagation and co-evolution between nodes are simulated through the state iteration process, local synchronization phenomena are identified, and potential group emergence characteristics are extracted; the state similarity and information entropy correlation are combined, and weighted fusion is used to generate the correlation strength matrix, which comprehensively considers static correlation and dynamic evolution, and improves the stability and accuracy of the final correlation result.

[0087] In an optional embodiment, according to the correlation strength matrix, the chaotic oscillator parameters are configured for each node, the disturbance signal is amplified through the phase synchronization process between the oscillators, and the oscillation characteristics of adjacent nodes are compared to obtain the disturbance feature vector including:

[0088] Calculating the diagonal elements, maximum element values, and average values ​​of the correlation strength matrix, determining a first vibrator parameter based on the product of the diagonal elements and a preset first adjustment coefficient, determining a second vibrator parameter based on the product of the maximum element value and a preset second adjustment coefficient, and determining a third vibrator parameter based on the product of the average value and a preset third adjustment coefficient;

[0089] Constructing a chaotic oscillator dynamics equation for each sensor according to the first oscillator parameter, the second oscillator parameter, and the third oscillator parameter, and solving the chaotic oscillator dynamics equation to obtain a chaotic oscillator state value;

[0090] According to the element values ​​of the correlation strength matrix, the chaotic oscillator dynamic equation is expanded into a coupling equation between sensors. The chaotic oscillator phase value is obtained by solving the coupling equation through a multi-level phase extraction method to determine the response sensitivity of the chaotic oscillator to the disturbance signal.

[0091] The oscillation difference of the chaotic oscillators of adjacent sensors is calculated by weighting the difference between the chaotic oscillator state value and the chaotic oscillator phase value;

[0092] The oscillation differences between all adjacent sensors are combined in the order of sensor numbers to obtain the disturbance characteristic vector of the distribution network.

[0093] In one specific implementation, a distribution network monitoring system consisting of 10 sensor nodes is used as an example. The correlation strength matrix R between these sensors is obtained through historical data analysis. The diagonal elements of matrix R represent the characteristics of the sensors themselves, while the off-diagonal elements represent the correlation strength between the sensors. Calculations show that the average diagonal element value of matrix R is 0.85, the maximum element value is 0.93, and the average value of the entire matrix is ​​0.42.

[0094] For each sensor node, the chaotic oscillator parameters are configured based on the characteristics of the correlation strength matrix. These parameters consist of three components: the first oscillator parameter α is related to the diagonal elements of the correlation strength matrix; the second oscillator parameter β is related to the maximum element value of the correlation strength matrix; and the third oscillator parameter γ is related to the average value of the correlation strength matrix. Specifically, the first adjustment coefficient is set to 1.2, the second adjustment coefficient is set to 0.8, and the third adjustment coefficient is set to 1.5. Therefore, the first oscillator parameter α = 0.85 × 1.2 = 1.02, the second oscillator parameter β = 0.93 × 0.8 = 0.744, and the third oscillator parameter γ = 0.42 × 1.5 = 0.63.

[0095] Based on the above parameters, a chaotic oscillator dynamics equation is constructed for each sensor, containing three state variables x, y, and z, which represent the motion state of the oscillator in three dimensions. By describing the time-varying patterns of these three state variables, the dynamics equation reflects the chaotic system's sensitivity to initial conditions. In practical applications, these equations can be solved using the fourth-order Runge-Kutta method, with a time step of 0.01 and 10,000 iterations. For the first sensor, the initial state is set to x(0)=0.1, y(0)=0, and z(0)=0. The initial states of the remaining sensors are randomly perturbed based on this state, with the perturbation amplitude not exceeding 0.001. Through numerical solution, the state trajectory of the chaotic oscillator corresponding to each sensor is obtained, and the state values ​​of the last 2,000 steps are recorded as the steady-state response.

[0096] Based on the element values ​​of the correlation strength matrix, a coupling equation is constructed between sensors. For sensors i and j, if the corresponding element Rij in the correlation strength matrix is ​​greater than a threshold of 0.3, they are considered to be coupled, and the coupling strength is proportional to Rij. The coupling equation describes how chaotic oscillators corresponding to different sensors influence each other, specifically how they amplify small perturbation signals through phase synchronization. A multi-level phase extraction method is used to solve the coupling equation. The instantaneous phase of each oscillator is calculated using the Hilbert transform. The phase signal is then decomposed into different frequency levels through wavelet decomposition, and the phase synchronization index is synthesized. Calculations show that under normal conditions, the phase synchronization index fluctuates between 0.65 and 0.75. However, when the system is disturbed, the phase synchronization index decreases significantly, reaching a minimum of 0.3.

[0097] By comparing the differences in the chaotic oscillator state and phase values, the oscillation differences between adjacent sensors are calculated. Specifically, for sensors i and j, their Euclidean distance dij and phase difference θij in state space are calculated. Then, a weighted combination of dij and θij is used to obtain the comprehensive difference index Dij. The weights can be adjusted based on the specific application scenario. In this example, the weight for state distance is 0.4, and the weight for phase difference is 0.6. Experimental data shows that during normal operation, the comprehensive difference index Dij for adjacent sensors typically does not exceed 0.25. However, when a disturbance occurs, Dij in the affected area can rise sharply to 0.6 or even higher.

[0098] The comprehensive difference indicators of all adjacent sensor pairs (a total of 45 pairs) are combined in order of sensor number (sorted by the first sensor number first, then by the second sensor number if the numbers are the same, and so on) to form a 45-dimensional vector, which is the disturbance feature vector of the distribution network. In practical applications, this feature vector can be input into a classifier to identify the type of disturbance. Test results show that compared with traditional methods, this method improves the accuracy of identifying disturbance types such as voltage flicker, harmonic pollution, and transient overvoltage by approximately 15%. It is particularly sensitive to small, initial disturbances and can issue early warning signals 200 milliseconds in advance.

[0099] In this embodiment, based on the chaotic oscillator dynamics modeling, it is possible to respond to external disturbance signals with high sensitivity and accurately capture the abnormal change characteristics of the distribution network; the coupled oscillator parameters are adaptively adjusted by driving the correlation strength matrix, so that the model has good adaptability to different structures and disturbance conditions; the chaotic oscillator state and phase difference are comprehensively weighted and calculated to improve the ability to distinguish small disturbances or hidden anomalies; the disturbance characteristic vector is formed by combining the oscillation differences between sensors to fully reflect the propagation and impact range of the disturbance in the spatial dimension.

[0100] In an optional embodiment, based on the element values ​​of the correlation strength matrix, the chaotic oscillator dynamics equation is expanded into a coupling equation between sensors, and the chaotic oscillator phase value is obtained by solving the coupling equation using a multi-level phase extraction method. Determining the response sensitivity of the chaotic oscillator to the disturbance signal includes:

[0101] Substituting the element values ​​of the correlation strength matrix into the chaotic oscillator dynamics equation to establish the coupling equation between sensors;

[0102] performing a Hilbert-Huang transform on the solution of the coupling equation to decompose the sensor response signal into a superposition of multiple eigenmode functions and residual terms;

[0103] Performing a continuous wavelet transform on each of the eigenmode functions, calculating wavelet coefficients according to a preset wavelet basis function, performing a Hilbert transform on each eigenmode function based on the wavelet coefficients, obtaining a real component and an imaginary component of each eigenmode function, calculating an inverse tangent value of the real component and the imaginary component, and superimposing the product of the inverse tangent value of each eigenmode function and the corresponding weight coefficient to obtain a chaotic oscillator phase value of the sensor;

[0104] In a preset observation time window, the absolute value of the derivative of the chaotic oscillator phase value with respect to time is calculated, the absolute value is integrated and normalized in the observation time window, and the response sensitivity of the chaotic oscillator to the disturbance signal is determined.

[0105] In one specific embodiment, it is necessary to obtain a correlation strength matrix between sensors in a sensor network. This correlation strength matrix can be obtained by calculating the correlation, mutual information, or other statistical correlation metrics of sensor data. For example, for a network containing five sensors, a 5×5 correlation strength matrix can be obtained, where the element aij represents the correlation strength between sensor i and sensor j, and its value range is typically [0, 1].

[0106] Substituting the elements of the correlation strength matrix into the chaotic oscillator dynamics equations establishes the coupling equations between sensors. These chaotic oscillator dynamics equations generally employ the Kuramoto model or the Rössler oscillator model. Taking the Kuramoto model as an example, for sensor i, its dynamics equation can be expressed as the rate of change of phase θi related to the natural frequency ωi and the coupling strength aij with other sensors. In practice, each sensor can be considered an oscillator with a natural frequency, and the elements of the correlation strength matrix aij are introduced into the equations as coupling coefficients. For five sensors, the coupling coefficients a12 = 0.85, a23 = 0.72, a34 = 0.63, and a45 = 0.91, etc. Substituting these values ​​into the chaotic oscillator dynamics equations establishes a complete set of coupling equations.

[0107] The Hilbert-Huang transform is performed on the solution to the coupled equations to decompose the sensor response signal into a superposition of multiple intrinsic mode functions and residual terms. The Hilbert-Huang transform is a time-frequency analysis method that can decompose nonlinear, nonstationary signals into multiple intrinsic mode functions. Decomposing the response signal of each sensor typically yields 3-10 intrinsic mode functions and one residual term. For example, the response signal of sensor 1 might be decomposed into five intrinsic mode functions: IMF1, IMF2, IMF3, IMF4, and IMF5, and a residual term, r.

[0108] A continuous wavelet transform is performed on each intrinsic mode function, and the wavelet coefficients are calculated based on the preset wavelet basis functions. Wavelet basis functions can be selected, such as Morlet wavelet, Mexican hat wavelet, or Daubechies wavelet. Taking the Morlet wavelet as an example, it has excellent time-frequency localization characteristics and is suitable for analyzing non-stationary signals. In practice, a continuous wavelet transform is performed on each IMF to obtain wavelet coefficients reflecting different scale characteristics. For example, for IMF1, the scale parameter a can be selected from 1 to 64, and the position parameter b can be selected from 1 to the signal length to calculate the wavelet coefficient matrix W1.

[0109] Based on the wavelet coefficients, each intrinsic mode function is Hilbert transformed to obtain its real and imaginary components. The Hilbert transform extracts the instantaneous phase information of the signal. For IMF1, the Hilbert transform yields an analytical signal consisting of a real component x1(t) and an imaginary component y1(t). The real component is the original eigenmode function, while the imaginary component is the Hilbert transform result of the real component.

[0110] Calculate the inverse tangent of the real and imaginary components to obtain the instantaneous phase of the intrinsic mode function. For IMF1, its instantaneous phase φ1(t) is the inverse tangent of x1(t) and y1(t). Similarly, calculate the instantaneous phases φ2(t) to φ5(t) and φr(t) for IMF2 to IMF5 and the residual term, respectively.

[0111] The chaotic oscillator phase value of the sensor is obtained by superimposing the product of the arctangent value of each eigenmode function and the corresponding weight coefficient. The weight coefficient can be determined based on the energy ratio of each eigenmode function. For example, if the energy ratios of IMF1 to IMF5 are 0.42, 0.28, 0.15, 0.09, and 0.06, respectively, the weight coefficients can be set to w1 = 0.42, w2 = 0.28, w3 = 0.15, w4 = 0.09, and w5 = 0.06. The chaotic oscillator phase value Φ(t) is the sum of the weighted phases.

[0112] Within a preset observation time window, calculate the absolute value of the time derivative of the chaotic oscillator's phase. The phase derivative reflects the rate of phase change, while its absolute value indicates the severity of the phase change. For example, within a 10-second observation window, sample every 0.01 seconds, resulting in 1000 sampling points. Calculate the absolute value of the phase derivative |dΦ(t) / dt| at each sampling point.

[0113] The absolute value of the phase derivative is integrated and normalized within the observation time window to determine the chaotic oscillator's response sensitivity to the perturbation signal. The integral calculation can be performed using the trapezoidal rule or Simpson's rule. For example, the integral value I is obtained by integrating |dΦ(t) / dt| within a 10-second window. This value is then divided by the window length of 10 seconds to obtain the normalized value S. This normalized value S is an indicator of the chaotic oscillator's response sensitivity to the perturbation signal.

[0114] In a test of a five-node sensor network, the maximum element value in the correlation strength matrix was 0.91, and the minimum element value was 0.32. Using the above method, the response sensitivities of each sensor to disturbance signals were calculated to be 0.78, 0.65, 0.83, 0.57, and 0.69, respectively. Sensors with higher response sensitivities have stronger detection capabilities for external disturbances and are therefore preferred for early identification of abnormal events.

[0115] Existing technologies typically use statistical thresholding, cluster analysis, or machine learning to detect anomalies in sensor networks. These methods often rely on static features of sensor data or simple time series analysis, lacking consideration of the dynamics of the sensor network as a whole. Traditional approaches typically treat sensors as independent entities, ignoring the impact of complex coupling relationships between sensors on their dynamic behavior. This results in insufficient detection sensitivity for weak anomaly signals in complex environments. Furthermore, existing technologies often employ linear analysis tools such as Fourier transforms when processing nonlinear and nonstationary signals, making it difficult to effectively capture transient features and nonlinear dynamic characteristics within the signal.

[0116] The method of this embodiment introduces a chaotic oscillator dynamics model, treating the sensor network as a coupled oscillator system. Using a correlation strength matrix, it accurately quantifies the mutual influence between sensors, thereby constructing a network dynamics model that better aligns with actual physical processes. In terms of signal processing, this embodiment decomposes nonlinear, nonstationary signals into multiple intrinsic mode functions using the Hilbert-Huang transform. Combining the continuous wavelet transform with the Hilbert transform, this method accurately extracts the signal's multiscale features and instantaneous phase information. This signal processing method better adapts to the complex signal characteristics of real-world environments and overcomes the limitations of traditional Fourier analysis. The method of this embodiment improves the sensor network's ability to detect weak disturbance signals. Abnormal signals often manifest as weak disturbances in their early stages, making them difficult to detect with traditional methods. By constructing a chaotic oscillator model and exploiting the chaotic system's sensitive dependence on initial conditions, this embodiment amplifies the impact of weak disturbance signals. Furthermore, sophisticated time-frequency analysis methods are used to extract phase dynamics, thereby enhancing detection sensitivity.

[0117] Figure 2This figure shows a performance comparison between the chaotic oscillator method and traditional methods. Experimental verification shows that this method demonstrates significant advantages over traditional statistical thresholding and machine learning methods in detecting weak anomaly signals. This method also significantly improves response time, enabling detection of early signs of anomalies much earlier than traditional methods. This high sensitivity and rapid response capability are invaluable for many safety-critical applications, effectively reducing disaster risks and minimizing economic losses. By calculating the response sensitivity of each sensor, this method provides a scientific basis for optimizing the placement of sensor networks, helping to improve the effectiveness of the entire monitoring system.

[0118] In an optional embodiment, based on the disturbance eigenvector, calculating the stability index and determining the time series evolution characteristics, combined with the node density distribution and cascading failure risk assessment, identifying the weak node set in the network includes:

[0119] Calculate the Euclidean norm of the disturbance eigenvector in the time dimension, and calculate the initial stability index of the node based on the logarithmic change rate of the Euclidean norm;

[0120] Record the time series changes of the initial stability index, construct the node stability matrix, and obtain the node fluctuation value by calculating the standard deviation of each node index in the node stability matrix;

[0121] Based on the preset stability threshold, the number of times when the index of each node exceeds the stability threshold is counted, and the number of times is divided by the observation duration to obtain the probability value of node instability;

[0122] The weight coefficients are set for the node instability probability value, node fluctuation value, and the mean of the node index, and the node comprehensive vulnerability value is obtained by weighted summation.

[0123] Based on the node's comprehensive vulnerability value, a node density distribution matrix is ​​constructed. The local density of the node is obtained by calculating the average of the comprehensive vulnerability values ​​of each node and a preset number of adjacent nodes. The density peak node is identified based on the temporal evolution characteristics of the node local density. The attribution coefficients of other nodes are calculated with the density peak node as the center. Nodes with attribution coefficients greater than the preset attribution threshold are divided into a high-risk node subset.

[0124] The stability transfer coefficient between adjacent nodes in the high-risk node subset is calculated, and the node cascading failure risk value is evaluated based on the stability transfer coefficient. The high-risk sensor node subset is screened according to the node cascading failure risk value, and the weak node set is output.

[0125] In a specific embodiment, the disturbance characteristic vector data of each node in the network is obtained. These data reflect the response characteristics of the node when subjected to external disturbances. For each node, the Euclidean norm of its disturbance characteristic vector in the time dimension is calculated. Specifically, assuming that the disturbance characteristic vector of node i at time t is V_i(t), the Euclidean norm at this moment can be obtained by calculating the square root of the sum of the squares of each component of the vector. By calculating the logarithmic rate of change of the Euclidean norm at adjacent moments, the initial stability index S_i(t) of the node is obtained. For example, for an electric power network containing 100 nodes, the Euclidean norm of a node at 10 consecutive time points are [1.2, 1.5, 1.8, 2.3, 2.4, 2.2, 2.0, 2.5, 3.0, 3.2], and the calculated initial stability index sequence is [0.22, 0.18, 0.25, 0.04, -0.09, -0.10, 0.22, 0.18, 0.07].

[0126] Record the temporal changes in each node's initial stability index and construct a node stability matrix M. The rows of this matrix represent different nodes, and the columns represent different moments. The matrix element M_ij represents the stability index of node i at moment j. By calculating the standard deviation of each node index in the node stability matrix, we obtain the node fluctuation value F_i. For the example node above, its fluctuation value F_i is 0.13.

[0127] Based on a preset stability threshold α (for example, α = 0.2), count the number of moments N_i at which each node's index exceeds the stability threshold. Divide this number of moments by the total observation duration T to obtain the node instability probability P_i = N_i / T. In the above example, if the total observation duration is 9 time units and the node's stability index exceeds 0.2 at three moments, its instability probability is 0.33.

[0128] To comprehensively assess node vulnerability, weight coefficients w1, w2, and w3 are set for the node instability probability, node volatility, and the mean of the node index, respectively (for example, w1 = 0.5, w2 = 0.3, and w3 = 0.2). The weighted summation yields the node's comprehensive vulnerability value, V_i = w1 × P_i + w2 × F_i + w3 × A_i, where A_i is the mean of the node index. For this example node, its comprehensive vulnerability value is 0.5 × 0.33 + 0.3 × 0.13 + 0.2 × 0.11 = 0.21.

[0129] Based on the node's comprehensive vulnerability values, a node density distribution matrix D is constructed. The average of the comprehensive vulnerability values ​​of each node and a preset number of neighboring nodes (e.g., the five nearest neighbors) is calculated to obtain the node's local density ρ_i. By analyzing the temporal evolution of the node's local density, the density peak node is identified. With the density peak node as the center, the attribution coefficient δ_i of other nodes is calculated. The attribution coefficient is calculated based on the difference in comprehensive vulnerability between nodes and the network topological distance, reflecting the node's degree of affiliation with the density peak node. Nodes with an attribution coefficient greater than a preset attribution threshold β (e.g., β = 0.7) are classified into the high-risk node subset H.

[0130] After identifying a subset of high-risk nodes, the stability transfer coefficient γ_ij between adjacent nodes in this subset is calculated. This coefficient represents the degree to which the instability of node i affects adjacent node j. It is calculated based on the combined vulnerability difference and connection strength between nodes. Based on the stability transfer coefficient, the node's cascading failure risk value R_i is assessed. This cascading failure risk value takes into account the node's combined vulnerability and its ability to influence surrounding nodes. The calculation formula is R_i = V_i × ∑γ_ij (where j represents all nodes directly connected to node i).

[0131] The high-risk node subset is screened using the node cascading failure risk value, and nodes with risk values ​​exceeding a preset threshold θ (for example, θ = 0.3) are selected as the final weak node set W. In practical applications, for a communication network containing 1,000 nodes, this method ultimately identified 23 critical weak nodes. These nodes indeed exhibited high failure risks and cascading effects in subsequent network stress tests.

[0132] This method also allows for parameter adjustments based on the characteristics of different networks. For example, for more stable power grid systems, the stability threshold α can be increased to 0.25; for more sensitive financial networks, the attribution threshold β can be lowered to 0.6 to ensure that more potential risk nodes are captured. Furthermore, the weight coefficients w1, w2, and w3 can be optimized based on specific application scenarios to balance the impact of instability probability, volatility, and average stability in the comprehensive assessment.

[0133] In this embodiment, the initial stability index, volatility and instability probability of the node are quantified through time series analysis of disturbance characteristics, and the dynamic stability characteristics of the node are comprehensively characterized; multiple stability indicators are integrated to construct a comprehensive vulnerability value of the node, thereby improving the accuracy and robustness of identifying potential weak nodes; local density analysis and density peak identification methods are used to effectively divide high-risk node subsets and reveal potential vulnerable areas; the stability transfer coefficient is used to analyze the linkage relationship between high-risk nodes, to achieve quantitative assessment of node cascading failure risks, and to assist in the screening of key weak nodes and the formulation of defense strategies.

[0134] In an optional embodiment, for a set of weak nodes, fault features are extracted to establish an antigen-antibody mapping, and the antibody population is optimized through affinity-based clonal mutation operations, and the memory library is optimized. The fault response strategy set obtained based on the memory library screening includes:

[0135] The fault occurrence time, fault duration, and fault impact range of each node in the weak node set are constructed into a fault feature vector, and the fault feature vector is normalized by maximum and minimum values ​​to obtain a standard fault feature vector;

[0136] Using the standard fault feature vector as an antigen, generating multiple antibodies as an initial antibody population, wherein the antibodies correspond to fault response strategies;

[0137] constructing an affinity function based on the Euclidean distance between the standard fault feature vector and the antibody, wherein the affinity function includes a feature weight coefficient for balancing the importance of different fault features;

[0138] Calculating the affinity value of each antibody in the antibody population according to the affinity function, cloning and amplifying the antibodies according to the affinity value, wherein the number of clones amplified is positively correlated with the affinity value according to a preset positive proportional value, to obtain amplified antibodies;

[0139] Performing a direction-guided Gaussian mutation operation on the amplified antibodies, determining the mutation direction based on neighboring antibody information and historical evolution information, and obtaining mutant antibodies;

[0140] Construct a memory library to store high-affinity antibodies according to a preset affinity threshold. When the affinity value of the variant antibody is greater than the affinity value of the antibody in the memory library, the variant antibody is used to replace it.

[0141] Generate a collection of failure response strategies based on antibodies in the memory bank.

[0142] Figure 3This is a flow chart of a fault response strategy based on an artificial immune algorithm. In one specific implementation, node fault features in a set of weak nodes are extracted and standardized. For each weak node, its historical fault occurrence time (calculated in hours of operation), fault duration (in minutes), and fault impact range (number of affected associated nodes) are recorded. For example, for weak node A, its fault feature vector can be represented as [24.5, 45, 8], indicating that the node failed after 24.5 hours of operation, lasted for 45 minutes, and affected 8 associated nodes. Due to the large differences in the feature value ranges of different dimensions, normalization is required, using the maximum and minimum value normalization method. Assuming the fault occurrence time range is [0, 168] hours, the fault duration range is [1, 120] minutes, and the impact range is [1, 50] nodes, the above fault feature vector becomes [0.146, 0.370, 0.143] after normalization. This standardized vector serves as the antigen in the artificial immune system, representing the fault scenario to be addressed.

[0143] Generate an initial population of antibodies, each corresponding to a potential fault response strategy. Antibody encoding uses real-number encoding, and each antibody contains multiple decision variables, such as resource allocation ratio, repair priority, and backup path selection parameters. For example, for a fault response strategy with three decision variables, 50 initial antibodies can be randomly generated, with each antibody's decision variable value ranging from [0, 1]. A specific example antibody is [0.25, 0.68, 0.43], indicating a resource allocation ratio of 0.25, a repair priority of 0.68, and a backup path selection parameter of 0.43.

[0144] To evaluate the adaptability of antibodies (i.e., fault response strategies) to antigens (i.e., fault scenarios), an affinity function is constructed. The affinity function is calculated based on Euclidean distance and introduces feature weight coefficients to balance the importance of different fault features. Assume that the feature weight coefficients are set to [0.5, 0.3, 0.2], indicating that the importance of fault occurrence time, duration, and impact range decreases in sequence. For the standardized fault feature vectors [0.146, 0.370, 0.143] and the antibodies [0.25, 0.68, 0.43], their weighted Euclidean distances are calculated and converted into affinity values. The higher the affinity value, the more suitable the antibody is for the current fault scenario.

[0145] Based on the calculated affinity values, the antibody population is cloned and expanded. Antibodies with higher affinity values ​​will yield more clones. If the cloning coefficient is set to 10, then the number of clones for an antibody with an affinity of 0.8 is 8, and the number of clones for an antibody with an affinity of 0.5 is 5. Assuming that there are 5 antibodies in the initial population with affinities of [0.85, 0.76, 0.62, 0.58, 0.41], the number of antibodies obtained after clonal expansion is [9, 8, 6, 6, 4], for a total of 33 amplified antibodies.

[0146] A direction-guided Gaussian mutation operation is performed on the amplified antibodies to enhance antibody diversity and improve search efficiency. The mutation operation takes into account the information of neighboring antibodies and historical evolution information to determine the mutation direction. Specifically, for each amplified antibody, its distance vector with the high-affinity antibody is calculated, and the mutation direction vector is constructed based on the historical evolution trend. The mutation amplitude adopts an adaptive Gaussian distribution. The mutation amplitude of antibodies with high affinity is small, and the mutation amplitude of antibodies with low affinity is large. For example, for an antibody with an affinity of 0.85 [0.25, 0.68, 0.43], its mutation amplitude is set to 0.05, and after mutation, it may obtain [0.262, 0.671, 0.422]; while for an antibody with an affinity of 0.41, the mutation amplitude may be set to 0.15.

[0147] After mutation, a memory bank is established to store high-affinity antibodies. The memory bank capacity is set to 10, and the affinity threshold is set to 0.7. Initially, the memory bank is empty; as the algorithm iterates, antibodies with an affinity greater than 0.7 are added to the memory bank. When the memory bank is full and a new antibody with a higher affinity emerges, the antibody with the lowest affinity in the memory bank is replaced. For example, if the memory bank already has antibodies with affinities of [0.92, 0.88, 0.85, 0.82, 0.78, 0.76, 0.75, 0.73, 0.72, 0.71], and the affinity of the newly mutated antibody is 0.79, the antibody with an affinity of 0.71 in the memory bank is replaced.

[0148] In this embodiment, by standardizing the failure time, duration, and impact range of weak nodes, a unified measurement of fault feature vectors is formed, thereby enhancing the accuracy and comparability of fault characterization; an immune mechanism is used to generate antibody populations corresponding to a variety of fault response strategies, thereby improving the flexibility of responses when facing different types of faults; an affinity-based clonal amplification and direction-guided Gaussian mutation mechanism is introduced, and the search path is optimized by combining historical and neighborhood information to accelerate the convergence and discovery of high-quality strategies; a memory library is constructed to dynamically retain high-affinity antibodies, forming a knowledge accumulation for fault response, and realizing fast and efficient strategy recommendation and scheduling for new faults.

[0149] In an optional embodiment, performing a direction-guided Gaussian mutation operation on the amplified antibodies, determining the mutation direction based on neighboring antibody information and historical evolution information, and obtaining the mutant antibodies includes:

[0150] Performing a mutation operation on the amplified antibodies, and calculating the mutation asynchrony length of the amplified antibodies according to the affinity value of each amplified antibody using an exponential decay function, wherein the mutation asynchrony length is inversely correlated with the affinity value according to a preset inverse proportional value;

[0151] For each amplified antibody, based on the Euclidean distance between the amplified antibody and other amplified antibodies, the amplified antibodies whose distance is less than a preset nearest neighbor threshold are selected to construct a nearest neighbor antibody set;

[0152] Determining a weight coefficient according to the affinity value of each neighboring antibody in the neighboring antibody set, and performing a weighted summation on the product of the difference between the neighboring antibody and the current amplified antibody and the weight coefficient to obtain a local search direction vector;

[0153] The best antibody in each generation of the population is recorded, and the difference between the best antibodies of two adjacent generations and the product of the preset time decay coefficient are weighted and summed to obtain the historical evolution direction vector;

[0154] Linearly combining the local search direction vector and the historical evolution direction vector according to a preset combination weight coefficient to obtain a comprehensive direction vector, calculating the product of the comprehensive direction vector and the variable step length, determining a directional variation component, and adding the directional variation component to the random Gaussian variation component to obtain a total variation;

[0155] The total variation amount is added to the current amplified antibody to obtain a variant antibody.

[0156] In the Industrial Internet of Things-based distribution network state perception and control system, when performing mutation operations on amplified antibodies, an affinity adaptive mechanism is used to calculate the variable step length. The variable step length is calculated using an exponential decay function, with the affinity value used as the exponential parameter in the formula and multiplied by a preset inverse proportional value. This design ensures that antibodies with high affinity conduct a narrower, more precise search, while antibodies with low affinity conduct a wider, more extensive search. The preset inverse proportional value is typically set between 1.5 and 3.0 and can be adjusted based on the accuracy requirements of distribution network state monitoring. A larger inverse proportional value makes the impact of affinity on the variable step length more significant, making it suitable for scenarios requiring a more precise search; a smaller inverse proportional value makes the variable step length change more gradually, helping to maintain population diversity.

[0157] For each amplified antibody, the similarity between antibodies is measured by calculating the Euclidean distance. The Euclidean distance calculation involves the square root of the sum of the squared differences of the values ​​of each dimension of the antibody feature vector. In the distribution network state perception application, the antibody feature vector usually contains information in multiple dimensions such as voltage level, current magnitude, power factor, harmonic content, and temperature. The selection of the preset nearest neighbor threshold is very critical. If it is too large, the nearest neighbor antibody set will contain too many dissimilar antibodies, and if it is too small, the set may be empty. Usually, the nearest neighbor threshold is set to 0.3 to 0.5 times the average Euclidean distance based on the dimension of the feature vector and the data distribution characteristics. For high-dimensional feature spaces, an adaptive nearest neighbor threshold can be used. The initial threshold is large and gradually decreases as the iteration proceeds to achieve a transition from extensive exploration to fine search.

[0158] After constructing a set of neighbor antibodies, a weight coefficient needs to be assigned to each neighbor antibody in the set. The weight coefficient is calculated using a soft maximization method. This method subtracts the minimum affinity value in the set from the affinity value of each neighbor antibody, and then divides the affinity value by the range of affinity values ​​to obtain the normalized affinity. The normalized affinity is then exponentially calculated and divided by the sum of all exponents to obtain the weight coefficient. This soft maximization method can amplify affinity differences, giving neighbor antibodies with high affinity greater influence. In distribution network status monitoring, affinity is usually defined as the antibody's detection accuracy for abnormal conditions or its match with historical fault patterns. The calculation of the weight coefficient can also introduce a distance factor. That is, the closer the neighbor antibody is to the current antibody, the greater the weight, which can better utilize local information.

[0159] The local search direction vector is calculated by multiplying the difference vector between each neighboring antibody and the current amplified antibody by the corresponding weight coefficient and then accumulating them. This vector represents the possible optimization direction within the local area, pointing to the area with higher affinity. In implementation, attention should be paid to the normalization of the difference vector to avoid the influence of dimensional differences in different dimensions on the direction. In addition, a momentum factor can be set to include the local search direction of the previous generation in the calculation to enhance the continuity and stability of the search. In distribution network status monitoring, this local search strategy can effectively identify subtle change patterns in grid parameters and improve the detection ability of gradually worsening faults.

[0160] The calculation of the historical evolution direction vector involves the accumulation of optimal antibody information from multiple generations. Not only does it record the difference between the optimal antibodies of two adjacent generations, but it can also consider evolutionary trends over longer time spans. This is achieved by introducing a time decay coefficient, which is usually between 0.7 and 0.9. The larger the time decay coefficient, the more historical information is retained; the smaller the coefficient, the more sensitive the algorithm is to recent changes. In distribution network status monitoring, an appropriate time decay coefficient setting can balance the ability to quickly respond to sudden faults and track long-term evolution trends. The historical evolution direction vector can also use a weighted sliding window method to only consider the evolutionary information of the most recent generations, avoiding interference from premature exploration directions on the current search.

[0161] The comprehensive direction vector is a linear combination of the local search direction vector and the historical evolution direction vector. The setting of the combined weight coefficient reflects a balanced consideration of local information and global trends. In the early stages of distribution network state perception, a larger historical evolution weight can be set to promote global exploration; in the later stages, the local search weight is increased to strengthen the detailed search of potential areas. The combined weight coefficient can also be designed to change dynamically. As the number of iterations increases, the local search weight gradually increases and the historical evolution weight gradually decreases, achieving a smooth transition from exploration to development.

[0162] The directional mutation component is the product of the comprehensive direction vector and the variable asynchrony length. This makes the mutation operation not only directional, but also allows the amplitude of the mutation to be adaptively adjusted according to the affinity of the antibody. In order to maintain the diversity and exploration ability of the population, a random Gaussian mutation component is also introduced. The standard deviation of the random Gaussian mutation can be set to a certain proportion of the variable asynchrony length, usually between 0.1 and 0.3. In distribution network state perception, this random component helps to discover new fault modes or abnormal conditions, rather than being limited to the optimization of known modes. The total mutation amount is the weighted sum of the directional mutation component and the random Gaussian mutation component. The weight can be adjusted according to the specific needs of distribution network monitoring to balance deterministic search and random exploration.

[0163] The total mutation amount is added to the current amplified antibody to obtain the mutated antibody. In practice, the dimensional values ​​of the mutated antibody are checked to see if they exceed the reasonable range, and boundary processing is performed if necessary. After the mutation operation is completed, the affinity of the mutated antibody is evaluated to verify the validity of the mutation. In distribution network status monitoring, affinity evaluation typically includes testing the mutated antibody on historical data and verifying its consistency with expert knowledge.

[0164] Existing antibody mutation techniques typically employ random Gaussian or uniform mutations, lacking directional guidance and resulting in inefficient searches within the complex state space of distribution networks. Traditional immune algorithms often experience local optima or slow convergence when dealing with high-dimensional, nonlinear, and multimodal problems. Existing mutation methods often rely on information from a single individual or simple affinity guidance, failing to fully leverage the collective intelligence and historical evolution of the population, making them incapable of addressing the complexity and diversity of distribution network state changes.

[0165] This embodiment introduces an affinity-adaptive step-length mechanism, establishing an inversely proportional relationship between affinity and step-length through an exponential decay function, enabling refined mutation control. It also provides a local search strategy based on a neighboring antibody set, leveraging local spatial information through a weight distribution mechanism. Furthermore, it integrates historical evolutionary direction with local search direction to form a more intelligent directed mutation strategy. Through this adaptive mutation mechanism, the algorithm achieves a balance between global search and local, refined exploration. Through directed mutation that integrates multi-source information, the algorithm can more efficiently search for the optimal feature representation of the distribution network state.

[0166] The distribution network state perception and control system based on the industrial Internet of Things in an embodiment of the present invention includes:

[0167] The first unit is used to collect data in real time through IoT sensor nodes deployed in the distribution network, and transmit the data to the edge computing layer based on the IoT communication protocol to obtain a real-time monitoring data set;

[0168] The second unit is used to calculate the information entropy correlation between nodes based on the real-time monitoring data set, construct a dynamic coupling matrix, and iteratively calculate the direction and intensity of information flow between nodes to form a group emergence effect and obtain the correlation intensity matrix of the network topology;

[0169] The third unit is used to configure the chaotic oscillator parameters for each node according to the correlation strength matrix, amplify the disturbance signal through the phase synchronization process between the oscillators, compare the oscillation characteristics of adjacent nodes, and obtain the disturbance eigenvector;

[0170] The fourth unit is used to calculate the stability index and determine the time-series evolution characteristics based on the disturbance eigenvector, and identify the weak node set in the network by combining the node density distribution and cascading failure risk assessment;

[0171] The sixth unit is used to extract fault characteristics from weak node sets, establish antigen-antibody mapping, optimize the antibody population through affinity-based clonal mutation operations, optimize the memory library, and obtain a set of fault response strategies based on the memory library screening;

[0172] The seventh unit is used to match the optimal control solution from the fault response strategy set and send the corresponding control instructions to the controllable devices through the Internet of Things communication protocol.

[0173] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0174] processor;

[0175] a memory for storing processor-executable instructions;

[0176] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0177] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0178] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0179] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A distribution network state perception and control method based on industrial Internet of Things, characterized by: include: The IoT sensor nodes deployed in the distribution network collect data in real time and transmit the data to the edge computing layer based on the IoT communication protocol to obtain a real-time monitoring data set; Based on the real-time monitoring data set, the information entropy correlation between nodes is calculated, and a dynamic coupling matrix is ​​constructed. By iteratively calculating the direction and intensity of information flow between nodes, a group emergence effect is formed, and the correlation intensity matrix of the network topology is obtained; According to the correlation strength matrix, the chaotic oscillator parameters are configured for each node. The disturbance signal is amplified through the phase synchronization process between the oscillators. The oscillation characteristics of adjacent nodes are compared to obtain the disturbance eigenvector. Based on the disturbance eigenvector, the stability index is calculated and the temporal evolution characteristics are determined. Combined with the node density distribution and cascading failure risk assessment, the weak node set in the network is identified. For the weak node set, fault characteristics are extracted to establish antigen-antibody mapping. The antibody population is optimized through affinity-based clonal mutation operations, and the memory library is optimized. The fault response strategy set is obtained based on the memory library screening. The optimal control solution is matched from the fault response strategy set, and the corresponding control instructions are sent to the controllable devices through the Internet of Things communication protocol.

2. The method according to claim 1, characterized in that Based on the real-time monitoring data set, the information entropy correlation between nodes is calculated, and a dynamic coupling matrix is ​​constructed. By iteratively calculating the direction and intensity of information flow between nodes, a group emergence effect is formed. The resulting correlation intensity matrix of the network topology includes: The real-time monitoring data set is divided into time windows, and the data sequence within the time window is segmented into equal intervals. The statistical frequency is used to obtain the probability distribution. The information entropy of a single node and the mutual information entropy of node pairs are calculated. Based on the maximum mutual information entropy and information entropy, the information entropy correlation is determined. Initializing a dynamic coupling matrix using the information entropy correlation, wherein the non-diagonal elements of the dynamic coupling matrix are determined by the information entropy correlation of the corresponding node pairs, the diagonal elements are set to zero, and are updated in each time window; Based on the iterative calculation of the dynamic coupling matrix, the state increment of the node is obtained by multiplying the current state of the node with the dynamic coupling matrix. The direction component of the state increment corresponds to the direction of information flow, and the amplitude component corresponds to the flow intensity. The state increment is superimposed on the current state of the node to complete a round of iteration; After multiple rounds of iterative calculations, when local synchronization occurs in the node group, a group emergence effect is formed. The final stable state of the node is extracted, and the state similarity of the node pair is calculated and weightedly fused with the dynamic coupling matrix to obtain the correlation strength matrix.

3. The method according to claim 1, characterized in that According to the correlation strength matrix, the chaotic oscillator parameters are configured for each node. The disturbance signal is amplified through the phase synchronization process between the oscillators. The oscillation characteristics of adjacent nodes are compared to obtain the disturbance feature vector including: Calculating the diagonal elements, maximum element values, and average values ​​of the correlation strength matrix, determining a first vibrator parameter based on the product of the diagonal elements and a preset first adjustment coefficient, determining a second vibrator parameter based on the product of the maximum element value and a preset second adjustment coefficient, and determining a third vibrator parameter based on the product of the average value and a preset third adjustment coefficient; Constructing a chaotic oscillator dynamics equation for each sensor according to the first oscillator parameter, the second oscillator parameter, and the third oscillator parameter, and solving the chaotic oscillator dynamics equation to obtain a chaotic oscillator state value; According to the element values ​​of the correlation strength matrix, the chaotic oscillator dynamic equation is expanded into a coupling equation between sensors. The chaotic oscillator phase value is obtained by solving the coupling equation through a multi-level phase extraction method to determine the response sensitivity of the chaotic oscillator to the disturbance signal. The oscillation difference of the chaotic oscillators of adjacent sensors is calculated by weighting the difference between the chaotic oscillator state value and the chaotic oscillator phase value; The oscillation differences between all adjacent sensors are combined in the order of sensor numbers to obtain the disturbance characteristic vector of the distribution network.

4. The method according to claim 3, characterized in that According to the element values ​​of the correlation strength matrix, the chaotic oscillator dynamics equation is expanded into a coupling equation between sensors. The chaotic oscillator phase value is obtained by solving the coupling equation through a multi-level phase extraction method. The response sensitivity of the chaotic oscillator to the disturbance signal is determined by: Substituting the element values ​​of the correlation strength matrix into the chaotic oscillator dynamics equation to establish the coupling equation between sensors; performing a Hilbert-Huang transform on the solution of the coupling equation to decompose the sensor response signal into a superposition of multiple eigenmode functions and residual terms; Performing a continuous wavelet transform on each of the eigenmode functions, calculating wavelet coefficients according to a preset wavelet basis function, performing a Hilbert transform on each eigenmode function based on the wavelet coefficients, obtaining a real component and an imaginary component of each eigenmode function, calculating an inverse tangent value of the real component and the imaginary component, and superimposing the product of the inverse tangent value of each eigenmode function and the corresponding weight coefficient to obtain a chaotic oscillator phase value of the sensor; In a preset observation time window, the absolute value of the derivative of the chaotic oscillator phase value with respect to time is calculated, the absolute value is integrated and normalized in the observation time window, and the response sensitivity of the chaotic oscillator to the disturbance signal is determined.

5. The method according to claim 1, wherein Based on the disturbance eigenvector, the stability index is calculated and the time-series evolution characteristics are determined. Combined with the node density distribution and cascading failure risk assessment, the weak node set in the network is identified, including: Calculate the Euclidean norm of the disturbance eigenvector in the time dimension, and calculate the initial stability index of the node based on the logarithmic change rate of the Euclidean norm; Record the time series changes of the initial stability index, construct the node stability matrix, and obtain the node fluctuation value by calculating the standard deviation of each node index in the node stability matrix; Based on the preset stability threshold, the number of times when the index of each node exceeds the stability threshold is counted, and the number of times is divided by the observation duration to obtain the probability value of node instability; The weight coefficients are set for the node instability probability value, node fluctuation value, and the mean of the node index, and the node comprehensive vulnerability value is obtained by weighted summation. Based on the node's comprehensive vulnerability value, a node density distribution matrix is ​​constructed. The local density of the node is obtained by calculating the average of the comprehensive vulnerability values ​​of each node and a preset number of adjacent nodes. The density peak node is identified based on the temporal evolution characteristics of the node local density. The attribution coefficients of other nodes are calculated with the density peak node as the center. Nodes with attribution coefficients greater than the preset attribution threshold are divided into a high-risk node subset. The stability transfer coefficient between adjacent nodes in the high-risk node subset is calculated, and the node cascading failure risk value is evaluated based on the stability transfer coefficient. The high-risk sensor node subset is screened according to the node cascading failure risk value, and the weak node set is output.

6. The method according to claim 1, wherein For the weak node set, we extract fault characteristics and establish antigen-antibody mapping. We optimize the antibody population and memory library through affinity-based clonal mutation operations. Based on the memory library, we filter out the following fault response strategies: The fault occurrence time, fault duration, and fault impact range of each node in the weak node set are constructed into a fault feature vector, and the fault feature vector is normalized by maximum and minimum values ​​to obtain a standard fault feature vector; Using the standard fault feature vector as an antigen, generating multiple antibodies as an initial antibody population, wherein the antibodies correspond to fault response strategies; constructing an affinity function based on the Euclidean distance between the standard fault feature vector and the antibody, wherein the affinity function includes a feature weight coefficient for balancing the importance of different fault features; Calculating the affinity value of each antibody in the antibody population according to the affinity function, cloning and amplifying the antibodies according to the affinity value, wherein the number of clones amplified is positively correlated with the affinity value according to a preset positive proportional value, to obtain amplified antibodies; Performing a direction-guided Gaussian mutation operation on the amplified antibodies, determining the mutation direction based on neighboring antibody information and historical evolution information, and obtaining mutant antibodies; Construct a memory library to store high-affinity antibodies according to a preset affinity threshold. When the affinity value of the variant antibody is greater than the affinity value of the antibody in the memory library, the variant antibody is used to replace it. Generate a collection of failure response strategies based on antibodies in the memory bank.

7. The method according to claim 6, characterized in that A direction-guided Gaussian mutation operation is performed on the amplified antibodies. The mutation direction is determined based on the neighboring antibody information and historical evolution information. The obtained mutant antibodies include: Performing a mutation operation on the amplified antibodies, and calculating the mutation asynchrony length of the amplified antibodies according to the affinity value of each amplified antibody using an exponential decay function, wherein the mutation asynchrony length is inversely correlated with the affinity value according to a preset inverse proportional value; For each amplified antibody, based on the Euclidean distance between the amplified antibody and other amplified antibodies, the amplified antibodies whose distance is less than a preset nearest neighbor threshold are selected to construct a nearest neighbor antibody set; Determining a weight coefficient according to the affinity value of each neighboring antibody in the neighboring antibody set, and performing a weighted summation on the product of the difference between the neighboring antibody and the current amplified antibody and the weight coefficient to obtain a local search direction vector; The best antibody in each generation of the population is recorded, and the difference between the best antibodies of two adjacent generations and the product of the preset time decay coefficient are weighted and summed to obtain the historical evolution direction vector; Linearly combining the local search direction vector and the historical evolution direction vector according to a preset combination weight coefficient to obtain a comprehensive direction vector, calculating the product of the comprehensive direction vector and the variable step length, determining a directional variation component, and adding the directional variation component to the random Gaussian variation component to obtain a total variation; The total variation amount is added to the current amplified antibody to obtain a variant antibody.

8. A distribution network state perception and control system based on the industrial Internet of Things, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to collect data in real time through IoT sensor nodes deployed in the distribution network, and transmit the data to the edge computing layer based on the IoT communication protocol to obtain a real-time monitoring data set; The second unit is used to calculate the information entropy correlation between nodes based on the real-time monitoring data set, construct a dynamic coupling matrix, and iteratively calculate the direction and intensity of information flow between nodes to form a group emergence effect and obtain the correlation intensity matrix of the network topology; The third unit is used to configure the chaotic oscillator parameters for each node according to the correlation strength matrix, amplify the disturbance signal through the phase synchronization process between the oscillators, compare the oscillation characteristics of adjacent nodes, and obtain the disturbance eigenvector; The fourth unit is used to calculate the stability index and determine the time-series evolution characteristics based on the disturbance eigenvector, and identify the weak node set in the network by combining the node density distribution and cascading failure risk assessment; The sixth unit is used to extract fault characteristics from weak node sets, establish antigen-antibody mapping, optimize the antibody population through affinity-based clonal mutation operations, optimize the memory library, and obtain a set of fault response strategies based on the memory library screening; The seventh unit is used to match the optimal control solution from the fault response strategy set and send the corresponding control instructions to the controllable devices through the Internet of Things communication protocol.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images