A method for coordinated regulation of fuse action based on an artificial intelligence system

Abstract

The application relates to the field of artificial intelligence system control and discloses a fuse action cooperative regulation method based on an artificial intelligence system, which comprises the following steps: acquiring an initial state topology matrix of a controlled network and generating a pre-compiled action sequence set, collecting physical parameter mutation points asynchronously arrived at an energy isolation terminal, stripping a hardware absolute time stamp parameter, generating a discrete causal time sequence chain reflecting energy evolution logic according to physical sequence, comparing the discrete causal time sequence chain with a standard time sequence chain and issuing a cooperative instruction when matching succeeds, and the application realizes time decoupling of high-complexity logical deduction and transient action response, uses logical state order preservation to hedge feature distortion caused by physical sampling clock jitter, and improves decision certainty and cooperative control robustness of the artificial intelligence system under non-ideal asynchronous communication conditions.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence system control technology, and in particular relates to a method for coordinated control of fuse action based on an artificial intelligence system. Background Technology

[0002] Currently, in DC distribution networks and large-scale energy storage systems with high-frequency topology reconfiguration characteristics, controlled nodes are responsible for blocking the spread of fault energy. Conventional solutions use artificial intelligence models to aggregate operating condition data, calculate multi-dimensional feature distances, and match action thresholds to achieve logical optimization of multi-level nodes. This approach has global coordination capabilities in steady-state operation environments, improving the protection efficiency of each action node. However, existing artificial intelligence control logic has an implicit dependence on the absolute synchronization of sampling channel data. Under millisecond-level transient fault conditions, due to electromagnetic interference and communication channel congestion, the data stream reported by each node to the artificial intelligence control unit has unpredictable clock offsets, data packet jitter, and out-of-order arrival characteristics. This asynchronous delay causes the real-time acquired multi-dimensional data matrix to be misaligned on the time axis, resulting in timing logic conflicts in the control decision-making process.

[0003] To address the data misalignment problem, conventional approaches primarily involve adding underlying synchronization devices or implementing forced data alignment. However, in transient processes with high evolution speeds, such linear improvement paths have significant engineering limitations. The synchronization alignment process incurs substantial computational waiting overhead, causing the decision-making response cycle of the AI ​​system to lag behind the physical evolution of the thermal collapse at the fault point. Furthermore, physical-level alignment operations cannot eliminate logical causal order deviations caused by communication randomness, leaving the AI ​​model vulnerable to recognition failure. For instance, Chinese invention patent application CN120934007A discloses a collaborative control method and system for energy storage systems used in grid frequency regulation. This method identifies timing misalignment events through data monitoring devices and performs task redistribution and communication timing compensation based on output power data deviations. This technical solution belongs to a closed-loop correction mechanism based on feedback regulation, relying on real-time quantification of power loss. In the rapid evolution of DC system faults, the computational overhead is high, the sampling period is delayed, and the generated compensation amount cannot keep up with the physical thermal collapse speed. It cannot eliminate the distortion in model feature calculations caused by non-aligned asynchronous data streams. When high-dimensional topology fluctuations and random time delay jitter occur, logical recognition blind spots are generated, leading to the failure of collaborative control logic.

[0004] Therefore, the technical problem this invention aims to solve is how to extract discrete causal features from non-aligned operating condition data, construct time-series invariant matching logic with asynchronous tolerance in an artificial intelligence system, and achieve precise coordinated triggering of multi-level controlled nodes under transient faults. Summary of the Invention

[0005] This invention aims to solve the problem of collaborative protection failure caused by feature calculation deviations and action timing contradictions in artificial intelligence systems under asynchronous data flow environments.

[0006] In this technical solution, a method for coordinated control of fuse operation based on an artificial intelligence system includes the following steps:

[0007] Step S1: Obtain the steady-state characteristics of the controlled topology network and generate an initial state topology matrix containing the adjacency relationships between controlled energy-isolated terminals;

[0008] Step S2: Input the historical physical parameter sequence and initial state topology matrix of the controlled energy isolation terminal into the pre-trained spatiotemporal trajectory prediction model, and output a pre-compiled action sequence set. The pre-compiled action sequence set includes a predicted topology distortion sub-matrix representing the state evolution path and a cooperative action timing instruction table bound to the predicted topology distortion sub-matrix.

[0009] Step S3: Collect the transient physical state parameter sequence of each controlled energy isolation terminal in the controlled topology network that arrives asynchronously, and extract the first arrival point of the mutation where the absolute value of the physical parameter derivative of each controlled energy isolation terminal exceeds the preset noise threshold.

[0010] Step S4: Remove the absolute timestamp of the underlying hardware carried by the first mutation point, and generate a discrete causal time sequence chain that reflects the physical energy evolution logic based on the physical order in which each mutation first point arrives at the logic processing unit.

[0011] Step S5: Compare the order of each controlled energy isolation terminal in the discrete causal time-series chain with the order of the standard time-series chain in the pre-compiled action sequence set;

[0012] Step S6: When it is determined that the order of the discrete causal timing chain is consistent with the specific standard timing chain, the bound cooperative action timing instruction table is retrieved, and an action trigger signal is sent to the corresponding controlled energy isolation terminal according to the preset time offset parameter.

[0013] Preferably, the initial state topology matrix includes the logical topological distance between each controlled energy isolation terminal, load constraint parameters, and topological invariant features characterizing the adjacency relationship of nodes.

[0014] Preferably, after step S6, a dynamic bias compensation step is also included: Step S71, obtaining the actual response time sequence from the controlled energy isolation terminal receiving the action trigger signal to the completion of the feedback physical action; Step S72, calculating the residual sequence between the actual response time sequence and the predetermined reference response time; Step S73, determining that when five consecutive sample values ​​of the residual sequence maintain the same polarity, extracting the arithmetic mean of the residual sequence as the correction bias; Step S74, updating the time bias parameter of the corresponding controlled energy isolation terminal in the coordinated action timing instruction table using the correction bias.

[0015] Preferably, before step S5, a feature dimensionality reduction and filtering step is included: step S41, obtaining the peak interval parameter and attenuation slope parameter of each node in the transient physical state parameter sequence; step S42, eliminating data points where the attenuation slope parameter is greater than a first preset threshold and the peak interval parameter is less than a second preset threshold, and generating a purification condition vector; step S43, extracting the feature components of the purification condition vector and each predicted topological distortion submatrix, and calculating the correlation feature distance between the purification condition vector and the feature components.

[0016] Preferably, it also includes a topology reduction and degradation step: step S81, determining whether the distance of each associated feature is greater than the preset activation threshold in the pre-compiled action sequence set; step S82, stopping the comparison of discrete causal time-series chains when the distance of each associated feature is greater than the activation threshold within a continuous 3ms time window; step S83, monitoring the absolute value of the transient physical parameters of a single controlled energy isolation terminal, and sending an action trigger signal to the controlled energy isolation terminal when it exceeds the preset safety extreme value threshold.

[0017] Preferably, the method employs a spatiotemporal graph convolutional network, which uses temporal convolutional layers to capture the temporal variation characteristics of transient physical state parameter sequences and uses spatial convolutional layers to capture the topological spatial coupling characteristics between each controlled energy isolation terminal.

[0018] Preferably, in step S2, a graph neural network is used to model the physical properties of each controlled energy isolation terminal, mapping the physical entity to a controlled energy isolation terminal in the logical mapping space.

[0019] Preferably, the action trigger signal in step S6 includes a precise break time for each controlled energy isolation terminal participating in the collaboration, and the break time is allocated based on the logical topological distance between each controlled energy isolation terminal in the controlled topology network.

[0020] Preferably, in step S5, the order-preserving determination is performed using a node state transition function: if any directly adjacent node pair in the controlled topology network... State transition time stamps satisfy logical relationships If so, it is determined that the discrete causal time-series chain matches the standard time-series chain.

[0021] Compared with existing technologies, the present invention provides a method for coordinated control of fuse operation based on an artificial intelligence system, which has the following advantages:

[0022] 1. In the coordinated control of fuse action in an artificial intelligence system, by pre-compiling action sequences under different fault evolution paths during the steady-state operation phase, the high-complexity logic deduction and millisecond-level transient action response are decoupled in the time dimension. This reconstruction of control logic changes the delayed processing mode of recalculating after encountering a fault in the traditional control system. After receiving the physical characteristics of the topological boundary, the control unit only needs to extract the corresponding coordinated action instructions through low information entropy logic chain matching. This offsets the inherent logical delay generated by the reasoning process of the artificial intelligence model at the physical level and ensures the deterministic timing of the action of multi-node actuators under high-speed fault transients.

[0023] 2. The method of this invention replaces the forced alignment of absolute timestamps by constructing discrete causal time-series chains, solving the control blindness problem caused by communication jitter or data disorder in complex power or energy networks. This technical path of matching using time-series invariants formed by the order of arrival of node features no longer relies on the absolute synchronization of the physical sampling clock of the entire network. This enables the control system to accurately identify the topological diffusion path of physical energy evolution under the objective condition that there is asymmetric time delay in the data stream, fundamentally avoiding the distortion of model feature calculation caused by data axis displacement and the resulting failure of cooperative action.

[0024] 3. The method of this invention improves the adaptive robustness of the control system in a dynamically reconstructed topology environment by integrating a collaborative mechanism of spatiotemporal trajectory prediction and asynchronous triggering of feature distance. The pre-compiled action sequence set covers multiple predicted topological distortion sub-matrices, enabling the system to dynamically match the action instruction table that best fits the current physical evolution trend based on the temporal feature vector of the transient operating condition derivative sequence. This logical mapping method based on topological invariants enhances the control strategy's ability to capture high-dimensional topological fluctuations without changing the underlying hardware perception density, and achieves accurate fitting of the globally optimal collaborative scheme with the rapid physical evolution process. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the logical execution of the fuse action coordinated control method of the present invention.

[0026] Figure 2 This is a functional composition diagram of the fuse action coordination and control system of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0028] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, bottom, transverse, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0029] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0030] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0031] A method for coordinated control of fuse operation based on an artificial intelligence system includes the following steps:

[0032] Step S1: Obtain the steady-state characteristics of the controlled topology network and generate an initial state topology matrix containing the adjacency relationships between controlled energy-isolated terminals;

[0033] Step S2: Input the historical physical parameter sequence and initial state topology matrix of the controlled energy isolation terminal into the pre-trained spatiotemporal trajectory prediction model, and output a pre-compiled action sequence set. The pre-compiled action sequence set includes a predicted topology distortion sub-matrix representing the state evolution path and a cooperative action timing instruction table bound to the predicted topology distortion sub-matrix.

[0034] Step S3: Collect the transient physical state parameter sequence of each controlled energy isolation terminal in the controlled topology network that arrives asynchronously, and extract the first arrival point of the mutation where the absolute value of the physical parameter derivative of each controlled energy isolation terminal exceeds the preset noise threshold.

[0035] Step S4: Remove the absolute timestamp of the underlying hardware carried by the first mutation point, and generate a discrete causal time sequence chain that reflects the physical energy evolution logic based on the physical order in which each mutation first point arrives at the logic processing unit.

[0036] Step S5: Compare the order of each controlled energy isolation terminal in the discrete causal time-series chain with the order of the standard time-series chain in the pre-compiled action sequence set;

[0037] Step S6: When it is determined that the order of the discrete causal timing chain is consistent with the specific standard timing chain, the bound cooperative action timing instruction table is retrieved, and an action trigger signal is sent to the corresponding controlled energy isolation terminal according to the preset time offset parameter.

[0038] Preferably, the initial state topology matrix includes the logical topological distance between each controlled energy isolation terminal, load constraint parameters, and topological invariant features characterizing the adjacency relationship of nodes.

[0039] Preferably, after step S6, a dynamic bias compensation step is also included: Step S71, obtaining the actual response time sequence from the controlled energy isolation terminal receiving the action trigger signal to the completion of the feedback physical action; Step S72, calculating the residual sequence between the actual response time sequence and the predetermined reference response time; Step S73, determining that when five consecutive sample values ​​of the residual sequence maintain the same polarity, extracting the arithmetic mean of the residual sequence as the correction bias; Step S74, updating the time bias parameter of the corresponding controlled energy isolation terminal in the coordinated action timing instruction table using the correction bias.

[0040] Preferably, before step S5, a feature dimensionality reduction and filtering step is included: step S41, obtaining the peak interval parameter and attenuation slope parameter of each node in the transient physical state parameter sequence; step S42, eliminating data points where the attenuation slope parameter is greater than a first preset threshold and the peak interval parameter is less than a second preset threshold, and generating a purification condition vector; step S43, extracting the feature components of the purification condition vector and each predicted topological distortion submatrix, and calculating the correlation feature distance between the purification condition vector and the feature components.

[0041] Preferably, it also includes a topology reduction and degradation step: step S81, determining whether the distance of each associated feature is greater than the preset activation threshold in the pre-compiled action sequence set; step S82, stopping the comparison of discrete causal time-series chains when the distance of each associated feature is greater than the activation threshold within a continuous 3ms time window; step S83, monitoring the absolute value of the transient physical parameters of a single controlled energy isolation terminal, and sending an action trigger signal to the controlled energy isolation terminal when it exceeds the preset safety extreme value threshold.

[0042] Preferably, the method employs a spatiotemporal graph convolutional network, which uses temporal convolutional layers to capture the temporal variation characteristics of transient physical state parameter sequences and uses spatial convolutional layers to capture the topological spatial coupling characteristics between each controlled energy isolation terminal.

[0043] Preferably, in step S2, a graph neural network is used to model the physical properties of each controlled energy isolation terminal, mapping the physical entity to a controlled energy isolation terminal in the logical mapping space.

[0044] Preferably, the action trigger signal in step S6 includes a precise break time for each controlled energy isolation terminal participating in the collaboration, and the break time is allocated based on the logical topological distance between each controlled energy isolation terminal in the controlled topology network.

[0045] Preferably, in step S5, the order-preserving determination is performed using a node state transition function: if any directly adjacent node pair in the controlled topology network... State transition time stamps satisfy logical relationships If so, it is determined that the discrete causal time-series chain matches the standard time-series chain.

[0046] Example 1: In a flexible DC distribution network with high-frequency topology reconfiguration characteristics, a fuse action coordinated control method based on an artificial intelligence system addresses timing race conflicts during fault transients by using a pre-generated action sequence set. Its operating environment includes multiple controlled energy isolation terminals distributed across a controlled network. When the system is in a steady-state cycle of grid-connected energy storage battery clusters, the control unit acquires the initial state topology matrix reflecting the network's logical connections and inputs the historical physical parameter sequences of the controlled energy isolation terminals into a spatiotemporal trajectory prediction model to generate pre-compiled actions containing a predicted topology distortion submatrix and a coordinated action timing instruction table. In application scenarios where the system faces transient short-circuit faults, the physical current state of multi-level distributed nodes approaches the thermal collapse critical point within milliseconds. The centralized artificial intelligence model is affected by the delay in multi-dimensional feature calculation and random electromagnetic interference in the communication channel, resulting in communication delay and out-of-order arrival of data packets when the collected operating condition data is reported to the control node. This physical sampling delay causes the real-time data matrix to be misaligned on the time axis. If the feature distance is matched according to the traditional absolute timestamp, the misaligned data stream will cause the artificial intelligence model to have a recognition blind spot and trigger the failure of coordinated action commands or the over-level tripping.

[0047] When a transient challenge occurs, the control unit continuously acquires the transient condition data stream uploaded by the topology boundary nodes and executes the matching trigger logic to extract the absolute value of the derivative of each node that exceeds the preset noise threshold. The system identifies the first arrival points of mutations, strips away the underlying hardware absolute timestamp parameters carried by each mutation first arrival point, and generates a discrete causal timing chain reflecting the energy evolution logic solely based on the physical order in which these mutation signals arrive at the logic processing unit. When faced with an inverted situation where near-end signals arrive later than far-end signals due to random link delays, the logic processing unit does not blindly follow the apparent reception timing. Instead, it automatically loads the out-of-order node identification array into the initial state topology matrix. Based on the pre-characterized absolute spatial topological physical distance between nodes within the matrix, it divides this distance by the known constant electromagnetic field of the energy fluctuations in the physical transmission medium. The wave propagation rate is used to calculate the inherent reference spatial propagation time difference between each controlled node. This reference time difference is then used to perform reverse delay compensation on the communication reception arrival time axis. After completely eliminating random disturbances caused by communication congestion, an ordered causal sequence following the laws of real physical wave propagation is reconstructed. This mechanism transforms the asynchronous delay of the physical world into ordered causal characteristics in the domain of logic control. This allows the system to no longer rely on the absolute synchronization of the entire network's sampling channels. Instead, it uses the node arrangement order in the discrete causal time sequence chain to perform matching with the standard time sequence chain in the pre-compiled submatrix, determining the discrete causal time sequence chain observed in real time. Any pair of directly adjacent nodes State transition time stamps satisfy logical relationships Furthermore, when this inequality holds true for all constraint node pairs along a specific diffusion path, the control unit retrieves the uniquely bound cooperative action timing instruction table and sends an action trigger signal to the corresponding controlled energy isolation terminal according to the preset time offset parameters therein, where the noise threshold... It is determined based on the variance envelope of historical sampled waveforms and is used to filter high-frequency white noise interference in signal transmission.

[0048] By placing the highly complex spatiotemporal trajectory prediction process in the steady-state cycle and reducing the triggering criteria of the transient response stage to a discrete time-series chain comparison based on causal logic, the technical solution achieves time decoupling between computation time and physical evolution. This architecture extracts discrete causal features from non-aligned operating condition data and constructs a logical matching logic with asynchronous tolerance in the artificial intelligence system. This eliminates the constraint of network-wide data sampling synchronization on the reliability of collaborative control. Without increasing the cost of high-precision clock synchronization hardware, it utilizes the topological invariant characteristics of physical energy evolution to offset the random delay caused by communication channel congestion, maintaining the collaborative determinism and safety isolation performance of multi-level action nodes under harsh operating conditions.

[0049] Example 2: This experiment verifies the collaborative performance of the technical solution under a non-ideal communication environment in a 10kV flexible DC distribution network physical simulation platform containing four voltage source converter nodes and six sets of controlled energy isolation terminals. The experimental data originates from real-time operating condition sequences acquired by a high-precision electrical parameter acquisition terminal in the physical simulation platform. The sampling frequency of this terminal is set to 100kHz, and the current measurement accuracy is not less than 0.1% to meet the requirement of capturing millisecond-level fault transient characteristics. To construct an experimental environment with realistic engineering interference, a random delay operator satisfying the Rayleigh distribution is introduced into the communication link between each measurement node and the control unit. The jitter range of the data packet arrival time is set between 2ms and 15ms. Gaussian white noise with a signal-to-noise ratio of 20dB is superimposed on the current signals acquired by each node. The setting decision logic involves a technical balance between sensitivity to signal mutations and the risk of false triggering. When the historical sampling waveform of the monitored node exhibits high levels of stable random fluctuations, in order to ensure the accuracy of identifying the first point of mutation, the control unit calculates the variance envelope of historical operating data within a sliding window. A background noise model is established. Specifically, when dealing with steady-state operating conditions containing high-order harmonics and non-stationary fluctuations, the control unit internally presets a discrete operating condition evaluation window with a width of 10 milliseconds. This window is continuously slid along the acquisition time axis in microsecond-level steps. The mathematical variance value of the digital current sampling within each sliding window frame is calculated sequentially. The system calls a cubic spline interpolation algorithm to perform smooth curve fitting on the local variance extrema points output by all continuously sliding evaluation windows. Finally, a continuous variance envelope representing the evolution trend of electromagnetic background fluctuations along the path is dynamically output. This will affect The core factors for determining the values ​​are identified as the noise floor power spectral density of the sampling system and the intensity of communication interference; a noise threshold is then set. and There is a positive correlation mapping relationship, that is ,in, As the safety gain coefficient, in this experiment, for a typical application example of energy storage grid connection impact conditions, it was determined by calibrating the mean variance of 500 steady-state cycles. The value of is 3.5, thereby obtaining a discrete judgment criterion for distinguishing physical faults from electromagnetic interference, ensuring that the physical first point of arrival extracted in a complex electromagnetic environment has causal order-preserving characteristics.

[0050] The system performs feature-based dimensionality reduction filtering on high-frequency electromagnetic pulse trains that penetrate the aforementioned amplitude thresholds. Based on the physical characteristics of high-frequency attenuation of electromagnetic waves in distributed impedance media, the high-frequency interference waveform caused by non-topological energy evolution exhibits a sharp amplitude drop and dense polarity reversal in the time domain. The control unit extracts the time difference between adjacent extreme points in the transient physical state parameter sequence as the peak interval parameter. The absolute ratio of the amplitude difference between adjacent extreme points to the peak interval parameter is calculated as the attenuation slope parameter. The control unit determines the measured Greater than the first preset threshold and When the value is less than the second preset threshold, the corresponding high-frequency oscillation data points are removed, and a purification condition vector is reconstructed. The control unit extracts the purification condition vector. eigencomponents of each predicted topological distortion submatrix Calculate the correlation feature distance between the two. Specific associated feature distance The calculation formula is as follows: The parameter definitions and physical constraints of the above formulas are as follows: The correlation feature distance characterizes the dispersion of real-time operating conditions and predicted states in the multi-dimensional topological space. Its value range is constrained to [0,2]. A smaller value indicates a higher degree of fit to the topological distortion path. The purification condition vector is in the first... The scalar components of dimension are represented by the number . The real-time sampling peak values ​​of transient physical parameters of the controlled energy isolation terminal are consistent with the physical dimensions of the underlying sensors. For the characteristic components in the th The scalar components of dimension represent the baseline parameter values ​​of the predicted topological distortion submatrix output by the model at the corresponding nodes, with units of 1 and 2. Maintain consistency. This is a feature dimension index, representing the terminal number with an independent physical sampling channel, and its value range is positive integers. The total feature dimension represents the total number of controlled energy isolation terminals participating in the current coordinated regulation. The value range is positive integers. The control unit performs calculations based on normalized inner product logic, removes the interference parameters caused by the absolute amplitude drift of a single node, and outputs pure spatial direction feature deviation data as the absolute quantitative basis for subsequent logical judgment.

[0051] In a short-circuit challenge scenario where the simulated fault point is located in the middle of a branch, the original input data faced by the sample group of this invention is characterized by time-domain overlap of the physical parameter derivative sequences reported by each terminal due to random time delays. The peak points of the original current derivatives obtained by the sampling terminals exhibit a disordered distribution on the receiving time axis of the control unit, which is opposite to the physical evolution order. Among them, the terminals located near the fault end... The sudden change signal arrived at the remote terminal after a delay of 12.4ms due to link congestion. The signal arrives at 3.2ms. During the execution of the matching trigger logic, the control unit extracts... Physical first arrival time marker and Physical first arrival time marker and obtain through calculation Node relative to The consistency between the physical evolution slope of the node and the preset standard value is achieved by comparing the discrete causal time-series chain after removing the clock stamp. The sample of this invention completes the matching of the predicted topological distortion sub-matrix within 1.5ms and generates a cooperative action command containing a precise time offset, so that the action deviation of the multi-level fuse is controlled within 0.8ms. In contrast, the control group that removes the discrete causal time-series chain logic relies on absolute timestamp matching and generates logic blocking when faced with random time delay jitter of more than 10ms, resulting in a cooperative action command issuance delay of more than 45.2ms.

[0052] The experimental data show a gradient change in technical effectiveness with increasing interference intensity. When the communication delay jitter is in the low-intensity range of 2ms to 5ms, the command coordination error of the present invention's sample group remains stable at 0.2ms. When the interference intensity increases to the high-intensity range of 10ms to 15ms, the logic matching failure rate of the control group rises to 34.5%, while the present invention's sample group, through constraint node pairs on a specific diffusion path, executes... The order-preserving determination of the relationship shows that the coordination accuracy only fluctuates slightly to 0.85ms. This performance reveals the adaptability of the technical solution to random time delays when processing non-aligned operating data. Further increasing the interference simulation, when the noise threshold... When the range is set to be more than 5 times the variance envelope, the system is observed to be unresponsive, with a 3.1ms delay in first-point acquisition. This confirms that the noise threshold range defined in this invention is a working window that balances response speed and disturbance rejection stability. Causal logic is restored in asynchronous data streams through a physical invariant matching mechanism.

[0053] Example 3: This example combines Figures 1 to 2 This document describes a method for coordinated control of fuse operation based on an artificial intelligence system, such as... Figure 1 As shown, step S1 obtains the steady-state characteristics of the controlled topology network and generates an initial state topology matrix containing the adjacency relationships between controlled energy isolation terminals; step S2 inputs the historical physical parameter sequence of the controlled energy isolation terminals and the initial state topology matrix into a pre-trained spatiotemporal trajectory prediction model, and outputs a pre-compiled action sequence set containing a predicted topology distortion sub-matrix and a cooperative action timing instruction table; step S3 collects the transient physical state parameter sequence of each controlled energy isolation terminal arriving asynchronously in the controlled topology network, and extracts the first arrival of abrupt changes where the absolute value of the physical parameter derivative of each controlled energy isolation terminal exceeds a preset noise threshold. Step S4 involves stripping the underlying hardware absolute timestamp carried by the first mutation arrival point and generating a discrete causal timing chain reflecting the physical energy evolution logic based on the physical order of arrival of each mutation arrival point to the logic processing unit. Step S5 involves comparing the arrangement order of each controlled energy isolation terminal in the discrete causal timing chain with the arrangement order of the standard timing chain in the pre-compiled action sequence set. Step S6 involves determining that when the arrangement order in the discrete causal timing chain matches a specific standard timing chain, retrieving the bound cooperative action timing instruction table and sending an action trigger signal to the corresponding controlled energy isolation terminal according to the preset time offset parameters therein.

[0054] like Figure 2 As shown, the artificial intelligence control unit internally deploys a spatiotemporal trajectory prediction model, a pre-compiled action sequence set, and an asynchronous communication network for the logic processing unit. This network serves as a signal transmission medium, connecting the artificial intelligence control unit and the controlled energy isolation terminal. Controlled energy isolation terminal Bidirectional connection, controlled energy isolation terminal and controlled energy isolation terminal Each device includes a transient physical state parameter sequence module for data acquisition and a hardware action execution mechanism for physical blocking. Each controlled energy isolation terminal reports the first arrival point data stream of the mutation to the artificial intelligence control unit via an asynchronous communication network. After executing discrete causal time-series chain comparison logic internally, the artificial intelligence control unit feeds back a coordinated action trigger signal to the corresponding controlled energy isolation terminal via the asynchronous communication network to drive the hardware action execution mechanism to complete the coordinated protection action under non-ideal asynchronous communication conditions.

[0055] Example 4: In a distributed energy storage power station containing 32 heterogeneous controlled energy isolation terminals, an AI-based method for coordinated control of fuse action achieves global timing alignment by offline calibration of the hardware action delay of each controlled energy isolation terminal. The technical challenge in this scenario lies in the millisecond-level discreteness of the fuse-breaking time among the controlled energy isolation terminal hardware from different physical batches. This hardware heterogeneity is injected as a constraint feature parameter into the spatiotemporal trajectory prediction model during the steady-state cycle derivation phase. The spatiotemporal trajectory prediction model employs a three-layer graph convolutional neural network architecture, and its input tensor... From the initial state topology matrix With physical parameter characteristic matrix Composition, in which the physical parameter characteristic matrix Includes historical current change rate of each controlled energy isolation terminal, bus voltage fluctuation value, and pre-calibrated hardware operation delay. The control unit obtains the average fuse-breaking time of each controlled energy isolation terminal under standard load through offline testing and defines it as the hardware action delay. .

[0056] During model inference, the first-layer graph convolution operator uses adjacency relationships to spatially aggregate node features, the second-layer temporal convolution operator extracts the evolution trend of the time series, and finally the output layer generates a predicted topological distortion sub-matrix and determines the time bias in the cooperative action timing instruction table based on the load importance weights of each node. Time bias The calculation is based on the following: ,in, This is the time offset amount sent to the controlled energy isolation terminal. The global baseline coordination time set for the system. To address the hardware operation delay of the controlled energy isolation terminal, For the estimated average communication link loss delay, all parameters are in milliseconds. This method of parameterizing the physical characteristics of heterogeneous hardware and using them as model input features eliminates the need for the control unit to calculate hardware compensation logic in real time during the fault transient matching phase. By calling the cooperative action timing instruction table containing the compensated time parameters, the impact of hardware discreteness on cooperative accuracy can be offset, ensuring the physical isolation reliability of the energy storage battery cluster under extreme topology distortion scenarios.

[0057] Example 5: In the initial startup phase of a controlled network containing multiple distributed deployment nodes, the fuse action coordinated control method based on an artificial intelligence system establishes a time reference through a pre-calibration step. The control unit sends synchronization pulse signals to each controlled energy isolation terminal and retrieves response messages. The average communication link loss delay is determined using half of the round-trip time. Under zero-load conditions, each controlled energy isolation terminal was induced to generate three no-load actions. The physical response time series of each controlled energy isolation terminal from receiving the command to contact separation was extracted, and the arithmetic mean was taken as the hardware action delay of each node. The obtained parameters are stored in the index table corresponding to the pre-compiled action sequence set, so that the global baseline coordination time set by the system is achieved. According to the formula Transformed into unique time offsets for each node .

[0058] When the system encounters physical parameter deviations due to changes in cable physical characteristics or fluctuations in environmental electromagnetic noise, the control unit maintains the noise threshold through a dynamic baseline reconstruction mechanism. The system continuously acquires the steady-state current waveform over five slip cycles and calculates its variance envelope to assess sensitivity. After monitoring When the deviation of the mean value from the original calibration value exceeds 15%, the control unit automatically generates a safety gain coefficient. The fine-tuning signal will be acquired in real time. Reused as the input tensor of the spatiotemporal trajectory prediction model Furthermore, the timing instruction table for coordinated actions is updated during the background deduction phase. This adaptive calibration process for non-ideal physical environments eliminates the risk of logic lockout caused by the solidification of initial parameters, ensuring that the discrete causal timing chain reflecting the energy evolution logic remains in a matching state with the predicted topological distortion sub-matrix even after multiple maintenance cycles.

[0059] Example 6: When the controlled topology network is in a dynamic debugging scenario after the addition of a new distributed power source, the control unit corrects the parameter distribution of the spatiotemporal trajectory prediction model by constructing a standardized offline training and incremental update procedure. The control unit extracts a heterogeneous dataset containing 10,000 normal samples and 2,000 fault samples from the historical operating condition database. Each sample contains the current derivative sequence and bus voltage change rate of each controlled energy isolation terminal. The spatiotemporal trajectory prediction model adopts an integrated architecture containing a three-layer graph convolutional neural network and a two-layer temporal convolutional neural network. During the training phase, the initial state topology matrix is... The non-zero elements in the matrix serve as the spatial weight benchmark between nodes, and are used as follows: The learning rate is used to adjust the model weights, and the cosine similarity index of the predicted topological distortion submatrix to be output is used. The optimization process stops when the value stabilizes above 0.95. Based on this, the system generates a time bias that includes topological feature weights during the background deduction phase using the updated model. The parameter sequence is mapped to the registers of each controlled energy isolation terminal, so that the coordinated action timing instruction table reflects the electrical path characteristics after the new power supply is connected. Based on the preparation of the offline training sample input, for the label extraction set required for supervised learning, the system automatically traces the final network break topology map of each group of historical fault samples in the heterogeneous dataset when physical isolation is successfully performed, statically converts it into a binary adjacency state matrix and labels it as the true value of the predicted topology distortion submatrix. At the same time, the optimal timing cut-off point that does not cause cascading tripping in the historical record is extracted as the true value label of the coordinated action instruction. The control unit guides and establishes the parameter convergence mapping path of the spatiotemporal trajectory prediction model in the process of graph convolution and temporal backpropagation by calculating the joint loss function of topological space cross-entropy and action timing mean square error between the output predicted sequence and the above two types of real calibration data.

[0060] In test conditions facing high-frequency electromagnetic pulse group interference ranging from 1MHz to 10MHz, the technical solution calibrates the noise threshold. Safety gain coefficient To maintain the extraction stability of discrete causal time series chains, the control unit acquires the original current waveform under this environment and calculates the variance envelope. The ratio of the extreme value of the derivative of the abrupt signal to the variance of the background noise was detected. When the gain is below 15dB, the system will adjust the safety gain factor. The step size was adjusted from 3.5 to 4.2 to filter out non-faulty physical first-arrival points caused by electromagnetic coupling, in order to determine the noise threshold. During the process, the control unit performs logical judgment; if the signal fluctuation amplitude within three consecutive sampling windows does not exceed... If the boundary is reached, the current mutation point is determined to be an interference signal and the matching logic is intercepted. To prevent the aforementioned extremely high amplitude transient electromagnetic pulses from directly breaking down the local autonomous protection extreme value bottom line set by the absolute value of a single transient physical parameter, the system additionally incorporates a time-domain integral anti-race interlock control program between the absolute value over-limit path and the trigger node of the actuator. This program forces that any over-limit absolute value of a single physical parameter must not only cross the safety extreme value threshold, but also maintain its over-limit state for a full system fundamental frequency period before the interlock can be released and an autonomous action trigger command can be issued. This eliminates the system over-level defense deadlock between the global dimensionality reduction anti-interference mechanism and the local direct protection function caused by high-frequency pulses. The parameter compensation procedure for boundary conditions enables the system to extract the order-preserving causal characteristics reflecting the energy evolution logic in an environment with degraded signal-to-noise ratio, so that the cooperative action command maintains a preset precision time alignment state in the asynchronous out-of-order data stream.

[0061] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A method for coordinated control of fuse operation based on an artificial intelligence system, characterized in that, Includes the following steps: Step S1: Obtain the steady-state characteristics of the controlled topology network and generate an initial state topology matrix containing the adjacency relationships between controlled energy-isolated terminals; Step S2: Input the historical physical parameter sequence and initial state topology matrix of the controlled energy isolation terminal into the pre-trained spatiotemporal trajectory prediction model, and output a pre-compiled action sequence set. The pre-compiled action sequence set includes a predicted topology distortion sub-matrix representing the state evolution path and a cooperative action timing instruction table bound to the predicted topology distortion sub-matrix. Step S3: Collect the transient physical state parameter sequence of each controlled energy isolation terminal in the controlled topology network that arrives asynchronously, and extract the first arrival point of the mutation where the absolute value of the physical parameter derivative of each controlled energy isolation terminal exceeds the preset noise threshold. Step S4: Remove the absolute timestamp of the underlying hardware carried by the first mutation point, and generate a discrete causal time sequence chain that reflects the physical energy evolution logic based on the physical order in which each mutation first point arrives at the logic processing unit. Step S5: Compare the order of each controlled energy isolation terminal in the discrete causal time-series chain with the order of the standard time-series chain in the pre-compiled action sequence set; Step S6: When it is determined that the order of the discrete causal timing chain is consistent with the specific standard timing chain, the bound cooperative action timing instruction table is retrieved, and an action trigger signal is sent to the corresponding controlled energy isolation terminal according to the preset time offset parameter.

2. The method for coordinated control of fuse action based on an artificial intelligence system according to claim 1, characterized in that, The initial state topology matrix contains the logical topological distance between each controlled energy isolation terminal, load constraint parameters, and topological invariant features characterizing the adjacency relationship of nodes.

3. The method for coordinated control of fuse action based on an artificial intelligence system according to claim 1, characterized in that, Following step S6, a dynamic bias compensation step is also included: Step S71, obtaining the actual response time sequence from the controlled energy isolation terminal receiving the action trigger signal to the completion of the feedback physical action; Step S72, calculating the residual sequence between the actual response time sequence and the predetermined reference response time; Step S73, determining that when five consecutive sample values ​​of the residual sequence maintain the same polarity, extracting the arithmetic mean of the residual sequence as the correction bias; Step S74, updating the time bias parameter of the corresponding controlled energy isolation terminal in the coordinated action timing instruction table using the correction bias.

4. The method for coordinated control of fuse action based on an artificial intelligence system according to claim 1, characterized in that, Before step S5, a feature dimensionality reduction and filtering step is also included: Step S41, obtain the peak interval parameter and attenuation slope parameter of each node in the transient physical state parameter sequence; Step S42, eliminate data points whose attenuation slope parameter is greater than the first preset threshold and whose peak interval parameter is less than the second preset threshold, and generate a purification condition vector; Step S43, extract the feature components of the purification condition vector and each predicted topological distortion submatrix, and calculate the correlation feature distance between the purification condition vector and the feature components.

5. The method for coordinated control of fuse action based on an artificial intelligence system according to claim 4, characterized in that, It also includes a topology reduction and degradation step: Step S81, determine whether the distance of each associated feature is greater than the preset activation threshold in the pre-compiled action sequence set; Step S82, when the distance of each associated feature is greater than the activation threshold within a continuous 3ms time window, stop comparing discrete causal time-series chains; Step S83, monitor the absolute value of the transient physical parameters of a single controlled energy isolation terminal, and when it exceeds the preset safety extreme value threshold, send an action trigger signal to the controlled energy isolation terminal.

6. The method for coordinated control of fuse action based on an artificial intelligence system according to claim 1, characterized in that, The method employs a spatiotemporal graph convolutional network, which uses temporal convolutional layers to capture the temporal variation characteristics of transient physical state parameter sequences and spatial convolutional layers to capture the topological spatial coupling characteristics between each controlled energy isolation terminal.

7. The method for coordinated control of fuse action based on an artificial intelligence system according to claim 1, characterized in that, In step S2, a graph neural network is used to model the physical properties of each controlled energy isolation terminal, mapping the physical entities to controlled energy isolation terminals in the logical mapping space.

8. The method for coordinated control of fuse action based on an artificial intelligence system according to claim 1, characterized in that, The action trigger signal in step S6 includes the precise break time for each controlled energy isolation terminal participating in the collaboration. The break time is allocated based on the logical topological distance between each controlled energy isolation terminal in the controlled topology network.

9. The method for coordinated control of fuse action based on an artificial intelligence system according to claim 1, characterized in that, In step S5, the order-preserving determination is performed using the node state transition function: if any directly adjacent node pair in the controlled topology network... State transition time stamps satisfy logical relationships If the discrete causal time-series chain matches the standard time-series chain, then it is determined that the two chains are successfully matched.

Images