Smart algorithm-based power grid digital emergency resource intelligent deployment system

Abstract

The application relates to the technical field of power grid resource allocation, and particularly discloses a power grid digital emergency resource intelligent allocation system based on an intelligent algorithm, which acquires current sampling sequences of a tripped line and its associated side of the same bus synchronously, and constructs a transient data set containing space-time correlation information. By means of feature extraction and space-time correlation analysis, a saturated distortion interval of a current transformer is identified and locked. A physical information neural network is introduced, linear data before saturation is taken as input, and inversion calculation is carried out by fusing excitation characteristic physical constraints, so that the real primary current waveform hidden by the saturation is reconstructed. The traditional mode of extensive resource scheduling only according to a tripping signal is changed, and differentiated and accurate emergency command based on real causes of faults is realized, so that resource mismatch is avoided at the root, and the overall efficiency and flexibility of power grid fault recovery are significantly improved.

Description

Technical Field

[0001] This invention belongs to the field of power grid resource allocation technology, and relates to a digital emergency resource intelligent allocation system for power grids based on intelligent algorithms. Background Technology

[0002] In the safe and stable operation of power systems, line main protection and current transformers play irreplaceable roles. As the main protection for rapid fault clearing within the transmission line's fast-clearing zone, the reliability and speed of longitudinal differential protection are the cornerstones for preventing fault escalation and maintaining the integrity of the power grid architecture. Meanwhile, as the sensing organ connecting the primary system and secondary protection devices, the accuracy of current transformer transmission and its transient characteristics directly determine the authenticity of the information upon which protection criteria are based.

[0003] When the system encounters severe disturbances, especially when a severe short circuit occurs near the end of an adjacent line outside the protection zone, the resulting huge short-circuit current often contains a large-amplitude aperiodic component, causing the core flux to become sharply biased. At this time, even if the protected line itself is sound and without faults, its current transformer (CT) may be unable to transmit linearly due to deep saturation, resulting in a severely distorted secondary current waveform.

[0004] Existing technologies rely excessively on threshold judgments of single electrical quantities, such as identifying faults solely through the amplitude or waveform symmetry of differential current. However, this identification logic often proves inadequate in the face of full waveform distortion caused by high levels of aperiodic components. More critically, when a line at point B trips due to a misjudgment, the traditional fault handling process, based on the simplistic logic of "tripping equals fault," classifies both points A and B as equally serious repair events, triggering a two-way dispatch of large-scale repair resources. This passive, reactive resource scheduling model often results in limited high-level technical resources being idle at the mis-tripping point B, while the truly urgent fault point A cannot obtain sufficient technical support in a timely manner due to resource constraints, ultimately prolonging the overall fault repair time. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides a smart grid digital emergency resource allocation system based on intelligent algorithms to solve the above-mentioned technical problems.

[0006] To achieve the above and other objectives, the technical solution adopted by the present invention is as follows:

[0007] This invention provides an intelligent emergency resource allocation system for digital power grids based on intelligent algorithms. The system includes:

[0008] Tripped line judgment module: acquires the first current sampling sequence on each side of the tripped line, and simultaneously acquires the second current sampling sequence on the side associated with the tripped line on the same busbar; uses feature extraction algorithm to identify waveform distortion features in the first current sampling sequence, performs spatiotemporal correlation analysis between waveform distortion features and current change rate in the second current sampling sequence, determines whether the tripped line has current transformer saturation state and locks the saturation range;

[0009] Primary current reconstruction module: When it is determined that there is a saturation state of the current transformer, the linear segment sampling data before the start point of the saturation interval in the first current sampling sequence is extracted, and the linear segment sampling data is used as the input vector of the physical information neural network model. Using the physical information neural network model in combination with the preset physical constraint equation of the current transformer excitation characteristics, the composite loss function containing physical residuals is minimized through iteration, and the decay time constant data containing non-periodic components is calculated by inversion, and then the reconstructed true primary current estimation sequence is output.

[0010] Fault label qualitative module: Substitute the actual primary current estimation sequence into the preset differential protection judgment equation to solve for the differential current data and braking current data of the tripped line in the reconfiguration state; generate fault qualitative labels based on the comparison values ​​of the reconfigured differential current data and braking current data. The fault qualitative labels are divided into internal short circuit labels and saturation maloperation labels.

[0011] The power grid emergency repair and dispatch module retrieves and matches corresponding emergency resource profile data from a pre-set digital resource attribute database based on the fault characterization label. Among them, the saturation malfunction label matches the relay protection operation and maintenance profile data, and the internal short circuit label matches the line emergency repair project profile data. With the shortest response time and the lowest resource consumption as the multi-objective optimization function, the module uses an intelligent scheduling algorithm to perform spatiotemporal path planning on the matched emergency resource profile data, and outputs and executes digital emergency resource dispatch instructions for the tripped lines.

[0012] As described above, the intelligent power grid digital emergency resource allocation system based on intelligent algorithms provided by the present invention has at least the following beneficial effects:

[0013] 1. This invention collects the current information of the tripped line itself and simultaneously acquires real-time transient data of the associated neighboring area on the same bus. Then, through a designed feature extraction and spatiotemporal correlation algorithm, it cross-validates waveform distortion features and changes in neighboring current. This effectively overcomes the spatial deceptiveness of single waveform analysis, thereby accurately determining the saturation state of the current transformer and locking its distortion range. This lays a reliable perceptual foundation for subsequent processing and significantly improves the accuracy and anti-interference capability of initial diagnosis in complex fault scenarios.

[0014] 2. This invention substitutes the reconstructed current sequence into the protection criteria to generate reliable qualitative labels for "internal short circuit" or "saturation maloperation". Based on this label, different professional resource profiles are accurately matched from the digital resource library, and customized scheduling instructions are generated and executed through intelligent algorithms with the goal of optimizing response time and resource consumption. This achieves refined operation by accurately deploying high-level emergency repair resources to the actual fault point, while only dispatching lightweight professional review personnel to the maloperation point. This not only significantly improves the utilization efficiency of critical emergency resources and avoids resource congestion and idleness, but also shortens the overall fault recovery time and enhances the resilience and economy of power grid operation. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A schematic diagram of the system logic provided for this application. Detailed Implementation

[0017] The following description, in conjunction with the implementation of the present invention, is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the scope defined by the inventive concept, and all such modifications and additions should fall within the protection scope of the present invention.

[0018] In traditional relay protection and fault handling systems, fixed braking characteristic curves and static differential criteria cannot adapt to the nonlinear waveform distortion caused by current transformer saturation under complex operating conditions. When an external fault generates a through current with a high proportion of aperiodic components, the system cannot establish a dynamic mapping relationship between the distorted waveform on the secondary side and the actual current on the primary side, leading to a mismatch between the protection action logic and the actual fault state. This static discrimination mechanism reduces the accuracy of fault identification, causing the differential protection to mistakenly identify the differential current caused by saturation as an internal fault current, ultimately triggering a malfunction.

[0019] For example, in a scenario where a short circuit occurs outside the fault zone due to a severe thunderstorm, the aperiodic component of the through-current generated by a fault on an adjacent line accounts for more than 60%, leading to deep saturation of the local current transformer. In this situation, traditional protection devices still rely on a preset fixed-ratio braking characteristic for discrimination, failing to identify the saturation distortion characteristics in the waveform and misjudging the false differential current caused by saturation as an internal fault. The protection device trips erroneously, triggering the dispatch system to send a large repair fleet to the non-faulty line, while the actual fault location faces insufficient resource allocation.

[0020] If these problems are not addressed, erroneous tripping of non-faulty lines will lead to incorrect grid topology splitting, expanding the outage area. The resource dispatch system will respond blindly due to receiving incorrect fault labels, causing limited high-level repair resources to be idled and strained at non-faulty points, delaying the repair of the actual fault locations. The lack of physically constrained current reconfiguration mechanisms will also lead to misjudgments of fault nature, hindering the grid's intelligent transformation from reactive repair to proactive defense, ultimately creating a negative feedback loop of both safety hazards and resource waste.

[0021] When faced with the aforementioned problems, traditional systems use a single current sequence for fault diagnosis, making it impossible to distinguish between saturation and faults. To address this, this application utilizes a feature extraction algorithm to pinpoint the saturation range. Further analysis reveals that simply removing the saturation segment data is insufficient to reconstruct the true fault; a physical information neural network model must be introduced, combined with the physical constraint equations of excitation characteristics, to reconstruct the true primary current. By calculating the decay time constant of the aperiodic component and generating accurate fault characterization labels, subsequent resource allocation can be guided, allowing repair strategies to adaptively adjust according to the fault nature, thereby resolving the problems of false tripping and resource mismatch.

[0022] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Example 1:

[0024] Please see Figure 1 As shown, the intelligent emergency resource allocation system for power grid based on intelligent algorithms includes a tripped line judgment module, a primary current reconstruction module, a fault tag qualitative module, and a power grid emergency repair allocation module.

[0025] The various modules are connected via wired and / or wireless connections to enable data transmission between them;

[0026] Tripped line judgment module: acquires the first current sampling sequence of each side of the tripped line, and simultaneously acquires the second current sampling sequence of the side associated with the tripped line on the same bus.

[0027] Feature extraction algorithms are used to identify waveform distortion features in the first current sampling sequence. Spatiotemporal correlation analysis is then performed between these waveform distortion features and the current change rate in the second current sampling sequence to determine whether the tripped line experiences current transformer saturation and to pinpoint the saturation range. This includes:

[0028] For the first current sampling sequence, the slope change difference between adjacent sampling points is extracted to generate waveform distortion feature quantity; for the second current sampling sequence, the first derivative value of the current at the same moment is extracted to generate reference current change rate.

[0029] By mapping the waveform distortion characteristics to the rate of change of the reference current, a saturation identification sub-model is constructed.

[0030] Obtain the physical configuration parameters of the current transformer in the tripped line and determine its magnetization response weight under different loads; use the magnetization response weight to correct the coefficients of the saturation identification sub-model and generate a saturation state judgment sequence.

[0031] In the saturation state determination sequence, the first sampling point where the value deviates from the reference benchmark beyond the stable fluctuation range is identified as the saturation start point; the first sampling point where the value returns to the stable fluctuation range after the saturation start point is identified as the saturation end point; and the time series data between the saturation start point and the saturation end point is defined as the saturation interval.

[0032] The preferred construction logic for the saturation recognition sub-model is as follows:

[0033] Establish a historical fault sample set containing historical trip records of the power grid system. For each historical trip record, extract the historical waveform distortion features and historical reference current change rate of the corresponding time interval as the initial sample for model input. Mark the current transformer saturation value determined after the historical trip record is inspected and verified as the corresponding real saturation label value.

[0034] Calculate the algebraic difference between the historical waveform distortion characteristic and the historical reference current change rate to generate a historical characteristic difference sequence. Divide the historical characteristic difference sequence into multiple continuous difference intensity intervals according to the numerical value. Extract sample records in each difference intensity interval whose true saturation label values ​​meet the preset judgment conditions. Calculate the arithmetic mean of the true saturation label values ​​corresponding to the extracted sample records and set it as the standard saturation probability target value corresponding to the difference intensity interval.

[0035] A multi-layer nonlinear mapping function structure is constructed, with waveform distortion characteristics and reference current change rate as joint independent variables and saturation recognition index as dependent variable. An interpolation amplification factor and mapping bias are introduced into the multi-layer nonlinear mapping function structure as iterative optimization parameters. A gradient descent algorithm incorporating a time decay factor is used to iteratively fit the correspondence between the joint independent variables and the standard saturation probability target value. By iteratively minimizing the cross-entropy error between the saturation recognition index output by the multi-layer nonlinear mapping function structure and the standard saturation probability target value, the optimal interpolation amplification factor and optimal mapping bias in the convergent state are solved, thereby generating the saturation recognition sub-model.

[0036] Preferably, the logic for generating the saturation state determination sequence is as follows:

[0037] The data points in the first current sampling sequence and the second current sampling sequence are input into the saturation recognition sub-model in chronological order to obtain the initial saturation recognition index output by the saturation recognition sub-model at each sampling time point;

[0038] Extract the instantaneous amplitude data of the first current sampling sequence corresponding to each sampling time point, and search for the corresponding transient flux bias coefficient in the physical configuration parameters of the current transformer based on the instantaneous amplitude data;

[0039] The matched transient flux bias coefficient is multiplied with the magnetization response weight to generate a dynamic correction factor for the sampling time point corresponding to the instantaneous amplitude data.

[0040] The initial saturation recognition index corresponding to each sampling time point is multiplied by the dynamic correction factor at the same sampling time point to obtain the corrected saturation index corresponding to each sampling time point.

[0041] The modified saturation indices corresponding to all sampling time points are integrated and spliced ​​in chronological order to generate a saturation state determination sequence.

[0042] In one specific embodiment, the first current sampling sequence on each side of the tripped line is obtained through a distributed current transformer. Simultaneously, the second current sampling sequence of adjacent lines located on the same busbar associated side as the tripped line and which have not experienced a trip is acquired. ;

[0043] Identify waveform distortion features in the first current sampling sequence using feature extraction algorithms. Its calculation logic is to extract the difference in slope between adjacent sampling points, and the specific formula is as follows:

[0044]

[0045] This formula filters out the linear variation of the fundamental component through second-order difference operations, highlighting the nonlinear transition characteristics caused by the saturation of the transformer core; simultaneously, it extracts the rate of change of the reference current for the second current sampling sequence. The calculation formula is: Its dimension is ampere per second, which characterizes the actual system current evolution trend on the bus side.

[0046] When constructing the saturation identification sub-model, offline training is performed using a historical fault sample set. The historical waveform distortion features are mapped to the historical reference current change rate using a difference mapping. The core mapping function adopts a variation of logistic regression.

[0047]

[0048] The saturation recognition index has a value range of [0,1]. This is the difference amplification factor, used to adjust the model's sensitivity to small distortions, with units of... 'b' is the mapping bias, used to correct background bias caused by system noise; during model training, a gradient descent algorithm incorporating a time decay factor is used to... Iterative optimization is performed with b, using a preset time decay factor. The recommended value is between 0.95 and 0.98, with 0.97 being preferred. The underlying logic is to assign higher weights to samples closer to the time of the fault occurrence, thereby ensuring the model's accuracy in recognizing features in the early stages of transient saturation.

[0049] Furthermore, to eliminate the influence of differences in the physical characteristics of different models of current transformers, the physical configuration parameters of the current transformers, including rated transformation ratio, core cross-sectional area, and magnetization curve data, are obtained from the equipment parameter database to determine their magnetization response weights under the current load. The coefficients of the saturation identification sub-model are corrected using this weight, generating a saturation state determination sequence. The corrected formula is:

[0050]

[0051] in, The preset saturation current threshold for the current transformer is obtained from the physical configuration parameters. The dynamic correction coefficient is preset to perform linear compensation based on the secondary circuit load impedance, and its value is usually between 0.1 and 0.5.

[0052] Finally, determine the sequence in the saturation state. In this process, range locking is performed, and a reference threshold for the stable fluctuation range is set. This threshold is typically calibrated based on the residual unbalanced current during normal line operation, and is recommended to be between 0.15 and 0.25, preferably 0.2. First time exceeding If the duration exceeds 3 sampling points, that moment is recorded as the saturation start point. Subsequently, the numerical changes were monitored, and when the values ​​returned to normal... The point below which remains stable is denoted as the saturation termination point. ;Will[ [This is] limited to the saturation range.

[0053] It should be added that the magnetization response weights under the current load are... The calculation logic is as follows:

[0054]

[0055] in, The instantaneous value of the current at the current sampling time point is directly derived from the first current sampling sequence; This is the secondary load impedance of the current transformer, measured in ohms. It is obtained by reading the measured value of the secondary circuit load in the physical configuration parameters. This parameter reflects the impedance characteristics of the external circuit of the transformer and directly affects the magnitude of the induced electromotive force. The rated transformation ratio of the current transformer is derived from the equipment parameter database.

[0056] This refers to the number of turns in the secondary winding; typically, the primary winding has one turn. This is based on the rated turns ratio. The number of turns is calculated based on the standard number of turns.

[0057] This is the angular frequency of the power grid, measured in radians per second, calculated based on the standard frequency of the power grid. The cross-sectional area of ​​the iron core is in square meters and is derived from the equipment parameter library. This parameter determines the magnetic reluctance characteristics of the magnetic circuit. The saturation magnetic flux density threshold, in Tesla, is extracted by inflection point analysis of the magnetization curve data and represents the critical magnetic flux value for the iron core to enter the saturation region. This is a preset magnetization correction coefficient, dimensionless, whose preset logic is based on the nonlinear hysteresis loop characteristics of the core material.

[0058] Primary current reconstruction module: When it is determined that there is a current transformer saturation state, the linear segment sampling data in the first current sampling sequence before the starting point of the saturation interval is extracted, and the linear segment sampling data is used as the input vector of the physical information neural network model.

[0059] By utilizing a physical information neural network model combined with pre-defined physical constraint equations for the excitation characteristics of current transformers, and iteratively minimizing a composite loss function containing physical residuals, the decay time constant data containing aperiodic components is calculated. This results in the output of a reconstructed true primary current estimation sequence, including:

[0060] The linear segment sampling data is used as the boundary training condition input into the physical information neural network model to establish a composite loss function that includes the data fitting error term and the physical constraint residual term.

[0061] Based on the equivalent impedance data of the secondary circuit and the magnetization characteristic data of the iron core of the tripped line, Kirchhoff's current law is used to construct the correlation logic between the rate of change of magnetic flux and the primary current, which is used as the basis for judging the physical constraint residual term.

[0062] In the iterative calculation process of the physical information neural network model, the decay time constant data of the non-periodic component is used as the variable to be optimized. The connection weights of the neural network are adjusted by the gradient descent algorithm until the value of the composite loss function converges to the preset error tolerance range.

[0063] The decay time constant data corresponding to the convergence state is obtained, and the optimized physical information neural network model is used to map and calculate the instantaneous value of the primary current at each sampling time point in the saturation interval to obtain the transient current reconstruction value containing the fundamental component and the decaying DC component.

[0064] The linear segment sampling data and the reconstructed transient current values ​​within the saturation interval are sequentially concatenated according to the sampling time order to generate the reconstructed true primary current estimation sequence.

[0065] Preferably, in the iterative calculation process of the physical information neural network model, the decay time constant data of the non-periodic component is used as the variable to be optimized, and the connection weights of the neural network are adjusted through the gradient descent algorithm until the value of the composite loss function converges to a preset error tolerance range, including:

[0066] Obtain linear segment sampling data With penalty factor Construct a composite loss function: ,in This represents the primary current mapping value output by the physical information neural network model at the k-th sampling time. The decay time constant data to be optimized is given by M, where M is the total number of sampling points. The derivative of the primary current mapping value output by the physical information neural network model at the k-th sampling time with respect to time is used to characterize the instantaneous rate of change of the primary current at that time.

[0067] The gradient increment of the neural network connection weights is obtained by differentiating the composite loss function L using the gradient descent algorithm. and the update increment of decay time constant data ,in and These are the preset learning rates. and These are the partial derivatives of the composite loss function with respect to the neural network connection weights and decay time constants, respectively; W is the current connection weight of the physical information neural network.

[0068] Perform variable update operations based on the gradient increment and update increment to obtain the updated weights. and updated decay time constant data ;

[0069] Updated weights Compared with the updated decay time constant data Feedback is fed into the physical information neural network model for the next round of iterative calculation, and the value of the composite loss function L is monitored in real time;

[0070] When the value of the composite loss function L meets the stopping criterion of being less than the preset error tolerance range, the iterative calculation is locked, and the current... The decay time constant data was determined by the inversion solution.

[0071] Preferably, the calculation process for the transient current reconstruction value, which includes the fundamental component and the attenuated DC component, is as follows:

[0072] Obtain the decay time constant data corresponding to the convergence state. Optimization weight matrix of physical information neural network model This allows us to obtain the reconstructed transient current values ​​at each sampling time point within the saturation interval. ;

[0073] Where k is the time index of the sampling point within the saturation interval, and f is the preset system rated frequency. For a preset fixed sampling period, This is the amplitude data of the primary current fundamental wave. This is the initial phase angle data for the primary current. This is the initial amplitude data for the non-periodic components;

[0074] Using a physical information neural network model combined with optimized weight matrix Feature mapping is performed on the sampled data of the linear segment to extract the envelope variation trend and zero-crossing distribution characteristics of the primary current. The fundamental amplitude data of the primary current is determined by nonlinear regression calculation. Primary current initial phase angle data and initial amplitude data of non-periodic components The optimal estimated solution;

[0075] Decay time constant data and the calculated fundamental current amplitude data Primary current initial phase angle data Initial amplitude data of non-periodic components Substituting into the above formula for calculating the transient current reconstruction value, and through point-by-point mapping calculation, the transient current reconstruction value corresponding to each sampling time point within the saturation interval is output. .

[0076] In one specific embodiment, after determining that a current transformer is in a saturation state, the primary current reconstruction module first extracts the linear segment sampling data located before the saturation start point from the first current sampling sequence. As the input vector for the physical information neural network model, this linear segment typically contains at least 1 / 4 cycle sampling points to ensure that sufficient initial phase angle and amplitude features can be extracted.

[0077] Subsequently, by using a physical information neural network model combined with the pre-set physical constraint equations for the excitation characteristics of the current transformer, the composite loss function L, which includes physical residuals, is minimized iteratively.

[0078] Specifically, the network structure and training logic configuration of the physical information neural network model are as follows:

[0079] The model consists of one input layer, three to five fully connected hidden layers (preferably four layers with 64 neurons each), and one output layer. The hidden layers use the Tanh hyperbolic tangent function as the activation function to ensure the smoothness and differentiability of the network's solution for higher-order derivatives of the current.

[0080] The composite loss function of the model is defined as: .

[0081] in, For the data fitting error term, the residual between the actual value of the linear segment sampled data and the network prediction value is calculated using the mean square error. The physical constraint residuals represent the network predictions substituted into Kirchhoff's current law and the core magnetization equation (i.e., ...). The resulting equation residuals; 2 and The weights for data loss and physical loss are set to 0.5 initially and adjusted using a dynamic adaptive weight allocation strategy during training.

[0082] During training, the Adam optimization algorithm is used for backpropagation, with an initial learning rate set to 0.001. The decay time constant data of the non-periodic components are used as the variables to be optimized (i.e., the learnable parameters), and simultaneously participate in gradient descent iterations until the aforementioned composite loss function is reached. If the value is less than the preset error tolerance range for 10 consecutive iterations, the network is considered to have converged. At this point, the current model parameters and decay time constant are extracted for subsequent current reconstruction mapping.

[0083] In the iterative calculation process of the physical information neural network model, the decay time constant data of the non-periodic component is used. As the variable to be optimized, the connection weights W of the neural network are adjusted using the gradient descent algorithm. The value of is determined by calculating the loss function in each iteration. partial derivatives Get update increment ,in The preset learning rate is recommended to be between 0.001 and 0.01, preferably 0.005. When the value of the composite loss function L converges to the preset error tolerance range, the current value is locked. The numerical value serves as the final decay time constant data. .

[0084] The fault label qualitative module substitutes the actual primary current estimation sequence into the preset differential protection judgment equation to solve for the differential current data and braking current data of the tripped line in the reconfiguration state. Based on the comparison value of the reconfigured differential current data and braking current data, a fault qualitative label is generated. The fault qualitative label is divided into internal short circuit label and saturation maloperation label.

[0085] In one specific embodiment, a forward mapping calculation is performed on the linear segment sampling data using a physical information neural network model to obtain the denoised feature sequence. Subsequently, based on the acquired decay time constant data... And a physical model of the transient current of the power system, constructing a model containing the parameters to be determined. , and The system of statically indeterminate equations. To improve solution efficiency, the nonlinear model is linearized using trigonometric identities, i.e., let... .

[0086] Next, we define three linear coefficients to be solved: (Amplitude of fundamental active component). (Amplitude of fundamental reactive component), and (Initial amplitude of the non-periodic component). Based on this, a matrix operation equation is constructed. Each row of the observation matrix A corresponds to a sampling time k, and its construction formula is:

[0087]

[0088] In the above formula, f is the system's rated frequency, which is fixed at 50Hz; The sampling period is k; k is the index of the sampling point within the linear segment. Matrix B represents the feature sequence output by the physical information neural network model. The column vector formed.

[0089] The coefficient vector above is solved using the least squares iterative logic. The calculation formula is as follows:

[0090]

[0091] This formula ensures the overall optimality of parameter estimation by minimizing the sum of squared residuals between the feature sequence and the physical model. Wherein, the matrix... for The optimal estimated coefficients can be obtained directly by inverting the symmetric positive definite matrix.

[0092] Finally, by analyzing the coefficient vector Perform an inverse transform to calculate the target physical parameters. Primary current fundamental amplitude data. The calculation formula is:

[0093]

[0094] Primary current initial phase angle data The calculation formula is:

[0095]

[0096] Initial amplitude data of non-periodic components Direct value .in The calculation needs to be combined with and The sign of the initial phase angle determines its quadrant, ensuring the uniqueness of the initial phase angle.

[0097] Preferably, the operation logic of the fault tag qualitative module includes:

[0098] Phasor feature information of the real primary current estimation sequence within the same sampling cycle is extracted. The reconstructed differential current data is obtained by calculating the vector sum of the phasors on each side, and the reconstructed braking current data is obtained by simultaneously calculating the arithmetic mean of the amplitudes of the phasors on each side.

[0099] The reconstructed differential current data and the reconstructed braking current data are mapped to coordinates to construct a motion characteristic trajectory model, where the reconstructed braking current data is the horizontal axis and the reconstructed differential current data is the vertical axis.

[0100] Obtain the proportional braking coefficient K corresponding to the tripped line, and use the proportional braking coefficient to reconstruct the braking current data. Perform linear scaling calculations to generate dynamic action threshold values. ; The preset minimum operating current setting;

[0101] Based on the magnitude distribution of coordinate points in the motion characteristic trajectory model, the true physical state of the tripped line is determined in reverse:

[0102] If the reconstructed differential current data falls into the braking constraint zone below the dynamic action threshold value, it is determined that the differential protection action of the tripped line is caused by the saturation of the current transformer, and a saturation false trip tag is generated.

[0103] If the reconstructed differential current data falls into the action triggering zone above the dynamic action threshold value, it is determined that the tripped line has real electrical short circuit characteristics, and an internal short circuit label is generated.

[0104] Preferably, the calculation process for the proportional braking coefficient corresponding to the tripped line is as follows:

[0105] Extract the peak current of the second current sampling sequence within the locked saturation range. And combined with the preset value of the ratio difference error obtained from the physical configuration parameters. Synchronization error preset value and the preset relay protection reliability coefficient The proportional braking coefficient corresponding to the tripped line is obtained. ;in, The preset minimum operating current setting is used to correct the braking sensitivity in the low current range; the proportional braking coefficient K is used to quantify the proportion of the maximum unbalanced current generated by non-fault factors in the reconfiguration state.

[0106] In one specific embodiment, the true primary current estimation sequences for each side output by the primary current reconstruction module are obtained. The discrete Fourier transform algorithm is then used to extract the phasor feature information of the true primary current estimation sequences within each sampling sliding window to obtain the current phasors for each side. The reconstructed differential flow data is then obtained by solving the vector summation operation. Simultaneously, the arithmetic mean of the current phasor amplitudes on each side is calculated to obtain the reconstructed braking current data. , where N is the number of sides of the tripped line.

[0107] Subsequently, the system quantifies the transient unbalanced current level caused by non-fault factors by calculating the proportional restraint coefficient corresponding to the tripped line. The formula for calculating the proportional restraint coefficient K... The peak current extracted from the second current sampling sequence within the locked saturation range, in amperes, is obtained by using a maximum value retrieval algorithm to search for the point with the maximum absolute value of the second current sequence within the saturation range, representing the maximum current level under external fault or heavy load conditions. This is the preset value for the ratio error of the current transformer, which is a dimensionless constant. According to national standards, it is usually taken as 0.1, representing a 10% rated error limit. The preset value for synchronization error is a dimensionless constant. The preset logic is based on the impact of sampling clock jitter and communication delay on phasor angle difference. The recommended value range is 0.02 to 0.08, with 0.05 being preferred. The reliability coefficient of relay protection is a dimensionless constant. The preset logic is to leave a safety margin to cope with extreme unpredictable factors. It is recommended to set it between 1.2 and 1.5, with 1.3 being preferred. The preset minimum operating current setting, in amperes, is used to avoid the minimum unbalanced current during normal line operation. Its recommended value is typically set to 0.2 to 0.5 times the rated current. The proportional braking coefficient K is necessary because it is introduced through the denominator. It achieves nonlinear smoothing of braking characteristics in the low current range, avoiding the loss of sensitivity caused by an excessively large K value when there are small current fluctuations.

[0108] Next, the calculated proportional braking coefficient K is combined with the reconstructed braking current data to obtain the dynamic action threshold value.

[0109] Finally, a motion characteristic trajectory model will be constructed to reconstruct the differential flow data. With dynamic action threshold value Point-by-point numerical comparison is performed: if the reconstructed differential current data remains below the dynamic action threshold, it indicates that the differential current is entirely covered by the calculated transient error envelope. The protection action of this tripped line is determined to be caused by a false differential current due to current transformer saturation, generating a saturation false trip tag. If the reconstructed differential current data exceeds the dynamic action threshold and enters the action trigger zone, it indicates that even after deducting the effects of saturation and synchronization errors, a significant differential current vector sum still exists. The tripped line is determined to have genuine electrical short-circuit characteristics, generating an internal short-circuit tag. This qualitative method based on physical error model inversion solves the problem of traditional criteria failing to distinguish between faults within and outside the zone under severe saturation.

[0110] Power grid emergency repair and dispatch module: Based on the fault characterization label, it retrieves and matches the corresponding emergency resource profile data from the preset digital resource attribute database. Among them, the saturation malfunction label matches the relay protection operation and maintenance profile data, and the internal short circuit label matches the line emergency repair project profile data.

[0111] Using the shortest response time and lowest resource consumption as multi-objective optimization functions, an intelligent scheduling algorithm is used to perform spatiotemporal path planning on the matched emergency resource profile data, outputting and executing a sequence of digital emergency resource allocation instructions for the tripped lines, including:

[0112] Based on the emergency resource profile data obtained by matching fault qualitative labels, extract the current geographic coordinate data, availability status markers, and resource level parameters of each emergency resource entity;

[0113] The current geographic coordinate data is spatially mapped to the fault coordinate location of the tripped line. The estimated response time for each emergency resource entity to reach the fault coordinate location is calculated, and the corresponding transfer cost value is calculated in combination with the resource level parameters.

[0114] A comprehensive scheduling cost function containing the expected response time and transfer cost is constructed. Multi-objective iterative optimization logic is used to sort the cost function values ​​of different emergency resource entities in descending order, and the optimal scheduling object with the lowest cost function value is selected.

[0115] Extract the business communication protocol of the optimal scheduling object, and convert the fault coordinate location, fault characterization label and corresponding graded handling plan of the tripped line into instruction control frames that conform to the business communication protocol, thereby generating a digital emergency resource allocation instruction sequence.

[0116] The sequence of digital emergency resource allocation instructions is sent to the mobile execution terminal of the optimal scheduling object through the digital allocation system interface, and the changes in the geographical coordinate data of the optimal scheduling object are monitored in real time.

[0117] Preferably, the optimal scheduling object with the lowest cost function value is selected, including:

[0118] For each emergency resource entity participating in the allocation route planning, its corresponding expected response time and transfer cost are extracted. The deviation standardization method is used to calculate the time normalization coefficient of the expected response time in all entity samples and the cost normalization coefficient of the transfer cost in all entity samples.

[0119] Obtain the weight allocation strategy corresponding to the fault qualitative label. When the fault qualitative label is the internal short circuit label, increase the weight of the time normalization coefficient value and decrease the weight of the cost normalization coefficient value. When the fault qualitative label is the saturation malfunction label, balance the weight ratio of the time normalization coefficient value and the cost normalization coefficient value.

[0120] The weighted time normalization coefficient value and the weighted cost normalization coefficient value are summed to construct a comprehensive scheduling cost function, and the scheduling cost value corresponding to each emergency resource entity is calculated.

[0121] Establish a scheduling cost sequence containing all emergency resource entities, and use bubble sort or quick sort algorithm to compare the values ​​and change the positions of the scheduling cost sequence to achieve ascending order of scheduling cost values ​​for different emergency resource entities and generate an ordered cost linked list.

[0122] Extract the scheduling cost value at the top of the ordered cost chain and define the emergency resource entity pointed to by the scheduling cost value as the optimal scheduling object.

[0123] In one specific embodiment, the power grid emergency repair and dispatch module first receives the fault characterization tags output by the fault tag characterization module, and performs a classification search in a preset digital resource attribute database based on the tag type. When a saturation malfunction tag is received, the system automatically associates it with the relay protection operation and maintenance profile data, aiming to dispatch secondary equipment professional technicians to conduct transformer drive experiments and setting value verification; when an internal short circuit tag is received, it is associated with the line emergency repair project profile data, aiming to dispatch live-line working vehicles and distribution network maintenance and testing personnel to isolate the fault point and replace spare parts.

[0124] After matching the emergency resource profile data of the corresponding category, the system extracts the current geographic coordinate data of each emergency resource entity. Available status flags and resource level parameters Subsequently, the module uses spatial mapping logic to calculate the coordinates of each resource entity reaching the fault location of the tripped line. Expected response time Compared with the figure of transportation cost Among them, the expected response time The calculation is based on:

[0125]

[0126] In the above formula, The preset average driving speed is expressed in kilometers per hour. Its preset logic is based on the historical average vehicle speed of the regional traffic network, and a value range of 30 km / h to 60 km / h is recommended. To ensure response redundancy under complex road conditions, this embodiment preferably uses 40 km / h. The numerator is the Euclidean distance between two points, expressed in kilometers. (Transportation cost value) The calculation formula is:

[0127]

[0128] in, The unit price is a preset unit price for transportation per kilometer, in yuan per kilometer, which is determined based on the current material transfer quota of the power logistics system. This is a resource level parameter, a dimensionless positive integer. The higher the level, the higher the administrative priority and guarantee intensity of resource allocation, and the higher the corresponding unit cost.

[0129] Next, in order to eliminate the impact of dimensional differences on multi-objective optimization, the time normalization coefficient value was calculated using the deviation standardization method. With cost normalization coefficient value The calculation process is as follows: ,in and These are the longest and shortest response times in the set of candidate resources, respectively; similarly, the dimensionless values ​​are obtained. Based on this, a comprehensive scheduling cost function is constructed. :

[0130]

[0131] in, 1 and 1 represents the time weighting coefficient and the cost weighting coefficient, respectively, both being dimensionless constants, and their sum equals 1. The preset logic of the weighting allocation strategy is: under the internal short-circuit tag, to restore power as quickly as possible, timeliness should be prioritized, and the weighting should be adjusted accordingly. 1 is set to the range of 0.7 to 0.9, preferably 0.8, correspondingly 1 is set to 0.2; under the saturation false trip label, since there is no actual risk of power outage, the focus is on balancing economy and efficiency, and will... 1 and 1 is set to 0.5.

[0132] Subsequently, the system establishes a scheduling cost sequence and uses the bubble sort algorithm to process all available resource entities. The values ​​are sorted in ascending order to generate an ordered cost linked list. The record with the lowest value is extracted from the first end of the list, and the entity it points to is defined as the optimal scheduling object. Finally, the module extracts the business communication protocol of the optimal scheduling object and encapsulates the fault coordinate location, fault characterization label, and corresponding emergency response plan into a command control frame. The command control frame is sent to the mobile execution terminal of the optimal scheduling object through the digital dispatch system interface, and the real-time geographic coordinates of the optimal scheduling object are obtained through GPS real-time incremental scanning function to monitor whether its trajectory converges to the fault coordinate location, thereby completing the closed-loop dispatch of emergency resources.

[0133] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0134] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0135] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart grid digital emergency resource allocation system based on intelligent algorithms, characterized in that, include: Tripped line judgment module: acquires the first current sampling sequence on each side of the tripped line, and simultaneously acquires the second current sampling sequence on the side associated with the tripped line on the same busbar. The waveform distortion features in the first current sampling sequence are identified by feature extraction algorithm. The waveform distortion features are then analyzed in a spatiotemporal correlation with the current change rate in the second current sampling sequence to determine whether the tripped line is in a current transformer saturation state and to lock the saturation range. Primary current reconstruction module: When it is determined that there is a saturation state of the current transformer, the linear segment sampling data before the start point of the saturation interval in the first current sampling sequence is extracted, and the linear segment sampling data is used as the input vector of the physical information neural network model. Using the physical information neural network model in combination with the preset physical constraint equation of the current transformer excitation characteristics, the composite loss function containing physical residuals is minimized through iteration, and the decay time constant data containing non-periodic components is calculated by inversion, and then the reconstructed true primary current estimation sequence is output. Fault label qualitative module: Substitute the actual primary current estimation sequence into the preset differential protection judgment equation to solve for the differential current data and braking current data of the tripped line in the reconfiguration state; Based on the comparison values ​​between differential current data and braking current data, a fault characterization label is generated; Power grid emergency repair and dispatch module: Based on the fault characterization label, retrieve and match the corresponding emergency resource profile data from the preset digital resource attribute database; Using the shortest response time and lowest resource consumption as the multi-objective optimization function, the intelligent scheduling algorithm performs spatiotemporal path planning on the matched emergency resource profile data, and outputs and executes emergency resource allocation instructions for the tripped lines.

2. The intelligent power grid digital emergency resource intelligent allocation system based on intelligent algorithms according to claim 1, characterized in that, Determining whether the tripped line has a current transformer saturation state and locking in the saturation range includes: For the first current sampling sequence, the slope change difference between adjacent sampling points is extracted to generate waveform distortion feature quantity; for the second current sampling sequence, the first derivative value of the current at the same moment is extracted to generate reference current change rate. By mapping the waveform distortion characteristics to the rate of change of the reference current, a saturation identification sub-model is constructed. Obtain the physical configuration parameters of the current transformer in the tripped line and determine its magnetization response weight under different loads; use the magnetization response weight to correct the coefficients of the saturation identification sub-model and generate a saturation state judgment sequence. In the saturation state determination sequence, the first sampling point where the value deviates from the reference benchmark beyond the stable fluctuation range is identified as the saturation start point; the first sampling point where the value returns to the stable fluctuation range after the saturation start point is identified as the saturation end point; and the time series data between the saturation start point and the saturation end point is defined as the saturation interval.

3. The intelligent power grid digital emergency resource intelligent allocation system based on intelligent algorithms according to claim 2, characterized in that, The construction logic of the saturation recognition sub-model is as follows: Establish a historical fault sample set containing historical trip records of the power grid system. For each historical trip record, extract the historical waveform distortion features and historical reference current change rate of the corresponding time interval as the initial sample for model input. Mark the current transformer saturation value determined after the historical trip record is inspected and verified as the corresponding real saturation label value. Calculate the algebraic difference between the historical waveform distortion characteristic and the historical reference current change rate to generate a historical characteristic difference sequence. Divide the historical characteristic difference sequence into multiple continuous difference intensity intervals according to the numerical value. Extract sample records in each difference intensity interval whose true saturation label values ​​meet the preset judgment conditions. Calculate the arithmetic mean of the true saturation label values ​​corresponding to the extracted sample records and set it as the standard saturation probability target value corresponding to the difference intensity interval. A multi-layer nonlinear mapping function structure is constructed, with waveform distortion characteristics and reference current change rate as joint independent variables and saturation recognition index as dependent variable. An interpolation amplification factor and mapping bias are introduced into the multi-layer nonlinear mapping function structure as iterative optimization parameters. A gradient descent algorithm incorporating a time decay factor is used to iteratively fit the correspondence between the joint independent variables and the standard saturation probability target value. By iteratively minimizing the cross-entropy error between the saturation recognition index output by the multi-layer nonlinear mapping function structure and the standard saturation probability target value, the optimal interpolation amplification factor and optimal mapping bias in the convergent state are solved, thereby generating the saturation recognition sub-model.

4. The intelligent power grid digital emergency resource intelligent allocation system based on intelligent algorithms according to claim 2, characterized in that, The logic for generating the saturation state determination sequence is as follows: The data points in the first current sampling sequence and the second current sampling sequence are input into the saturation recognition sub-model in chronological order to obtain the initial saturation recognition index output by the saturation recognition sub-model at each sampling time point; Extract the instantaneous amplitude data of the first current sampling sequence corresponding to each sampling time point, and search for the corresponding transient flux bias coefficient in the physical configuration parameters of the current transformer based on the instantaneous amplitude data; The matched transient flux bias coefficient is multiplied with the magnetization response weight to generate a dynamic correction factor for the sampling time point corresponding to the instantaneous amplitude data. The initial saturation recognition index corresponding to each sampling time point is multiplied by the dynamic correction factor at the same sampling time point to obtain the corrected saturation index corresponding to each sampling time point. The modified saturation indices corresponding to all sampling time points are integrated and spliced ​​in chronological order to generate a saturation state determination sequence.

5. The intelligent power grid digital emergency resource intelligent allocation system based on intelligent algorithms according to claim 1, characterized in that, The steps for obtaining the reconstructed true primary current estimation sequence include: The linear segment sampling data is used as the boundary training condition input into the physical information neural network model to establish a composite loss function that includes the data fitting error term and the physical constraint residual term. Based on the equivalent impedance data of the secondary circuit and the magnetization characteristic data of the iron core of the tripped line, Kirchhoff's current law is used to construct the correlation logic between the rate of change of magnetic flux and the primary current, which is used as the basis for judging the physical constraint residual term. In the iterative calculation process of the physical information neural network model, the decay time constant data of the non-periodic component is used as the variable to be optimized. The connection weights of the neural network are adjusted by the gradient descent algorithm until the value of the composite loss function converges to the preset error tolerance range. The decay time constant data corresponding to the convergence state is obtained, and the optimized physical information neural network model is used to map and calculate the instantaneous value of the primary current at each sampling time point in the saturation interval to obtain the transient current reconstruction value containing the fundamental component and the decaying DC component. The linear segment sampling data and the reconstructed transient current values ​​within the saturation interval are sequentially concatenated according to the sampling time order to generate the reconstructed true primary current estimation sequence.

6. The intelligent power grid digital emergency resource intelligent allocation system based on intelligent algorithms according to claim 5, characterized in that, In the iterative calculation process of the physical information neural network model, the decay time constant data of the non-periodic component is used as the variable to be optimized. The connection weights of the neural network are adjusted through the gradient descent algorithm until the value of the composite loss function converges to the preset error tolerance range, including: Obtain linear segment sampling data With penalty factor Construct a composite loss function: ,in This represents the primary current mapping value output by the physical information neural network model at the k-th sampling time. The decay time constant data to be optimized is given by M, where M is the total number of sampling points. The derivative of the primary current mapping value output by the physical information neural network model at the k-th sampling time with respect to time is used to characterize the instantaneous rate of change of the primary current at that time. The gradient increment of the neural network connection weights is obtained by differentiating the composite loss function L using the gradient descent algorithm. and the update increment of decay time constant data ,in and These are the preset learning rates. and These are the partial derivatives of the composite loss function with respect to the neural network connection weights and decay time constants, respectively; W is the current connection weight of the physical information neural network. Perform variable update operations based on the gradient increment and update increment to obtain the updated weights. and updated decay time constant data ; Updated weights Compared with the updated decay time constant data Feedback is fed into the physical information neural network model for the next round of iterative calculation, and the value of the composite loss function L is monitored in real time; When the value of the composite loss function L meets the stopping criterion of being less than the preset error tolerance range, the iterative calculation is locked, and the current... The decay time constant data was determined by the inversion solution.

7. The intelligent power grid digital emergency resource intelligent allocation system based on intelligent algorithms according to claim 5, characterized in that, The calculation process for the transient current reconstruction value, which includes the fundamental component and the attenuated DC component, is as follows: Obtain the decay time constant data corresponding to the convergence state. Optimization weight matrix of physical information neural network model This allows us to obtain the reconstructed transient current values ​​at each sampling time point within the saturation interval. ; Where k is the time index of the sampling point within the saturation interval, and f is the preset system rated frequency. For a preset fixed sampling period, This is the amplitude data of the primary current fundamental wave. This is the initial phase angle data for the primary current. This is the initial amplitude data for the non-periodic components; Using a physical information neural network model combined with optimized weight matrix Feature mapping is performed on the sampled data of the linear segment to extract the envelope variation trend and zero-crossing distribution characteristics of the primary current. The fundamental amplitude data of the primary current is determined by nonlinear regression calculation. Primary current initial phase angle data and initial amplitude data of non-periodic components The optimal estimated solution; Decay time constant data and the calculated fundamental current amplitude data Primary current initial phase angle data Initial amplitude data of non-periodic components Substituting into the above formula for calculating the transient current reconstruction value, and through point-by-point mapping calculation, the transient current reconstruction value corresponding to each sampling time point within the saturation interval is output. .

8. The intelligent power grid digital emergency resource intelligent allocation system based on intelligent algorithms according to claim 1, characterized in that, The operational logic of the fault tag qualitative module includes: Phasor feature information of the real primary current estimation sequence within the same sampling cycle is extracted. The reconstructed differential current data is obtained by calculating the vector sum of the phasors on each side, and the reconstructed braking current data is obtained by simultaneously calculating the arithmetic mean of the amplitudes of the phasors on each side. The reconstructed differential current data and the reconstructed braking current data are mapped to coordinates to construct a motion characteristic trajectory model, where the reconstructed braking current data is the horizontal axis and the reconstructed differential current data is the vertical axis. Obtain the proportional braking coefficient corresponding to the tripped line, and use the proportional braking coefficient to perform linear proportional calculation on the reconstructed braking current data to generate a dynamic action threshold value. Based on the magnitude distribution of coordinate points in the motion characteristic trajectory model, the true physical state of the tripped line is determined in reverse: If the reconstructed differential current data falls into the braking constraint zone below the dynamic action threshold value, it is determined that the differential protection action of the tripped line is caused by the saturation of the current transformer, and a saturation false trip tag is generated. If the reconstructed differential current data falls into the action triggering zone above the dynamic action threshold value, it is determined that the tripped line has real electrical short circuit characteristics, and an internal short circuit label is generated.

9. The intelligent power grid digital emergency resource intelligent allocation system based on intelligent algorithms according to claim 8, characterized in that, The calculation process for the proportional braking coefficient corresponding to this tripped line is as follows: Extract the peak current of the second current sampling sequence within the locked saturation range. And combined with the preset value of the ratio difference error obtained from the physical configuration parameters. Synchronization error preset value and the preset relay protection reliability coefficient The proportional braking coefficient corresponding to the tripped line is obtained. ;in, The preset minimum operating current setting is used to correct the braking sensitivity in the low current range.

10. The intelligent power grid digital emergency resource intelligent allocation system based on intelligent algorithms according to claim 8, characterized in that, Using the shortest response time and lowest resource consumption as multi-objective optimization functions, an intelligent scheduling algorithm is employed to perform spatiotemporal path planning on the matched emergency resource profile data, including: Based on the emergency resource profile data obtained by matching fault qualitative labels, extract the current geographic coordinate data, availability status markers, and resource level parameters of each emergency resource entity; The current geographic coordinate data is spatially mapped to the fault coordinate location of the tripped line. The estimated response time for each emergency resource entity to reach the fault coordinate location is calculated, and the corresponding transfer cost value is calculated in combination with the resource level parameters. A comprehensive scheduling cost function is constructed, which includes the expected response time and the transfer cost. Multi-objective iterative optimization logic is used to sort the cost function values ​​of different emergency resource entities in descending order, and the optimal scheduling object with the lowest cost function value is selected.

11. The intelligent power grid digital emergency resource intelligent allocation system based on intelligent algorithms according to claim 10, characterized in that, The optimal scheduling object with the lowest cost function value is selected, including: For each emergency resource entity participating in the allocation route planning, its corresponding expected response time and transfer cost are extracted. The deviation standardization method is used to calculate the time normalization coefficient of the expected response time in all entity samples and the cost normalization coefficient of the transfer cost in all entity samples. Obtain the weight allocation strategy corresponding to the fault qualitative label. When the fault qualitative label is the internal short circuit label, increase the weight of the time normalization coefficient value and decrease the weight of the cost normalization coefficient value. When the fault qualitative label is the saturation malfunction label, balance the weight ratio of the time normalization coefficient value and the cost normalization coefficient value. The weighted time normalization coefficient value and the weighted cost normalization coefficient value are summed to construct a comprehensive scheduling cost function, and the scheduling cost value corresponding to each emergency resource entity is calculated. Establish a scheduling cost sequence containing all emergency resource entities, and use bubble sort or quick sort algorithm to compare the values ​​and change the positions of the scheduling cost sequence to achieve ascending order of scheduling cost values ​​for different emergency resource entities and generate an ordered cost linked list. Extract the scheduling cost value at the top of the ordered cost chain and define the emergency resource entity pointed to by the scheduling cost value as the optimal scheduling object.

Images