A method for monitoring the state of an air-filled high-voltage switch cabinet

Abstract

The application relates to the technical field of power grid dispatching, and discloses a state monitoring method for a gaseous high-voltage switch cabinet. The method comprises the following steps: coupling and analyzing switch cabinet state monitoring data to obtain a coupling unit; obtaining a joint response function through model prediction; obtaining a reduced-order operator and a baseline response by solving the Jacobian of the function at a current working point, substituting the baseline value of an inactivated device and performing boundary consistency substitution to obtain a joint response prediction value; normalizing and fusing the prediction value to obtain a dispatchable health factor, and screening to obtain a dispatch decision scheme. The application improves the degree of coincidence between the health risk assessment result of the candidate dispatch scheme and the actual operation state of the device, and simultaneously improves the real-time evaluation efficiency of large-scale candidate schemes.

Description

Technical Field

[0001] This application relates to the field of power grid dispatching technology, and in particular to a method for monitoring the status of gas-filled high-voltage switchgear. Background Technology

[0002] Existing gas-insulated high-voltage switchgear condition monitoring technologies typically deploy various sensors on the switchgear body, including those for gas pressure, gas moisture content, partial discharge, mechanical characteristics, and contact and busbar temperatures. The raw signals collected are preprocessed through filtering and normalization by the distribution automation station terminal. Then, machine learning models deployed locally or in the cloud are used to model the historical and current monitoring data of a single switchgear, outputting the current health score or the probability of failure in the future for the switchgear, which is provided for maintenance personnel to refer to when formulating maintenance plans.

[0003] The existing health prediction models only use the equipment's historical and current monitoring data as input, and the output is a static health score that is unrelated to specific scheduling operations. This cannot reflect the response of the switchgear after specific scheduling actions such as load transfer and network reconfiguration. At the same time, the existing technology does not consider the physical conduction relationships such as heat conduction, current distribution conduction, and airflow coupling that objectively exist between multiple switchgear in the same compartment, the same bus section, or the same heat dissipation duct. The health assessment of each switchgear is carried out independently. When the candidate scheme on the scheduling side involves multiple switchgear with physical coupling relationships, the method of evaluating each switchgear independently cannot reflect the impact of coupling superposition on the overall response, and it is easy to underestimate the actual risk of the candidate scheme.

[0004] Even with the introduction of scheduling action parameters into the health prediction model and the use of graph convolution and other methods to model the coupling and transmission relationships between multiple switchgear, the sheer number of candidate schemes for grid reconfiguration or load transfer, coupled with the varying equipment combinations involved in each candidate scheme, means that re-executing a complete graph neural network inference for each candidate scheme would result in a rapid increase in computational overhead with the number of candidate schemes, making it difficult to meet the response time requirements typically required for scheduling decisions. Conversely, if a reduced-order coupling response operator is pre-solved offline for each coupling unit under the premise of a fixed physical coupling structure to avoid repetitive inference for each candidate scheme, the problem arises because candidate schemes typically only activate the coupling unit. The reduced-order coupling response operator is obtained by solving the entire coupling unit uniformly. When directly reusing this operator, for switchgear in the coupling unit that is not actually scheduled by the current candidate scheme, if the candidate action parameter value is substituted, the scheduling action that has never occurred will be fabricated. If the zero value is substituted, the current actual operating condition of the equipment will be lost. Both processing methods will cause the response prediction results of the other equipment in the coupling unit to deviate from the actual physical state of the equipment. How to make the response prediction results of the actual scheduling scenario where some equipment is activated and some equipment is not activated consistent with the actual operating condition of the equipment while reusing the pre-solved reduced-order coupling response operator has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a condition monitoring method for gas-insulated high-voltage switchgear. The technical problem to be solved is that in existing gas-insulated high-voltage switchgear condition monitoring methods, equipment health assessment and power grid dispatching decisions are disconnected, and the response prediction of candidate dispatching schemes in multi-equipment coupling scenarios is difficult to balance computational efficiency and consistency with the actual operating conditions of the equipment. By separating the physical coupling structure of the equipment from the dynamic dispatching decision, pre-solving reusable reduced-order coupling response operators and incorporating partial activation boundary consistency, the method improves the consistency between the health risk assessment results of candidate dispatching schemes and the actual operating status of the equipment, while also improving the real-time evaluation efficiency for large-scale candidate schemes.

[0006] This application provides a method for monitoring the status of a gas-filled high-voltage switchgear, the method comprising: Step S1: Perform coupling relationship analysis on the multi-dimensional status monitoring data of the switchgear to obtain the coupling unit; Step S2: Perform response prediction processing on the coupled unit using the coupled response prediction model to obtain the joint response prediction function relationship; Step S3: Perform Jacobian solution processing on the joint response prediction function relationship at the current operating point of the coupling unit to obtain the reduced-order coupling operator and the baseline response. Based on the candidate scheduling action parameters in the candidate scheduling scheme, substitute the measured baseline values ​​into the coupling devices in the coupling unit that are not activated by the candidate scheduling scheme. Perform partial activation boundary consistency substitution processing on the reduced-order coupling operator to obtain the joint response prediction value of the coupling unit corresponding to the candidate scheduling scheme. Step S4: Normalize and fuse the candidate scheduling schemes according to the joint response prediction values ​​of the coupled units to obtain the schedulable health factors, and select the scheduling decision schemes from the schedulable health factors.

[0007] This invention, when analyzing the coupling relationship of multi-dimensional status monitoring data of switchgear, divides the equipment into coupling units based on the actual installation and networking relationship. This makes the joint response prediction function relationship obtained by the subsequent coupling response prediction model no longer the response obtained in isolation for a single switchgear, but the joint response after the interaction of multiple switchgears with physical transmission relationships within the same coupling unit. This allows the health risk assessment results to truly reflect the comprehensive impact of candidate scheduling schemes on the entire coupling area, without missing the stress transmission between equipment due to independent evaluation of each switchgear. In the specific application scenario of gas-insulated high-voltage switchgear, it is common to see multiple devices arranged in the same compartment, connected to the same busbar section, and cooled by the same air duct. If a general single-point health prediction model is applied without considering this scenario, even if the model has high prediction accuracy, it will not be able to answer the question of what additional impact a scheduling action of a certain device will have on adjacent devices. Therefore, the value of the coupled response prediction model lies not in the model's fitting ability, but in its ability to transform the inherent physical coupling relationship of the switchgear compartment, busbar section, and cooling air duct into the model's input structure. This makes the algorithm truly fit the actual working condition of the centralized arrangement and close coupling of gas-insulated switchgear. This is also the key point to grasp when considering the introduction of such algorithms in similar application scenarios. That is, the model structure must correspond to the actual physical layout of the equipment. Simply stacking feature dimensions without considering the physical layout will not bring the same contribution.

[0008] Based on this, the present invention obtains a reduced-order coupling operator and baseline response by performing Jacobian solution on the joint response prediction function relationship at the current operating point of the coupling unit. This allows the operator to be repeatedly invoked when facing different candidate scheduling schemes, without having to re-execute the complete model inference for each candidate scheme. This feature enables health risk assessment to have the basic capability to support real-time comparison of large-scale candidate schemes, which meets the real-world requirements for response speed in scheduling decision-making scenarios such as network reconfiguration and load transfer. Furthermore, considering the reality that some switchgear cabinets within the coupling unit are activated by candidate schemes while others are not, this invention substitutes the measured baseline values ​​of the unactivated coupling devices and performs partial activation boundary consistency substitution processing on the reduced-order coupling operator. This ensures that the joint response prediction value can reflect the response of the activated devices after performing scheduling actions, while preventing unselected adjacent devices from generating non-existent response deviations. This processing reflects the deep integration of algorithm features and application scenarios. Because the reduced-order coupling operator itself is a general linear numerical processing method, it is only necessary to design the boundary consistency substitution step specifically when considering the scenario characteristic that the candidate scheduling scheme for switchgear cabinets usually only involves some devices. Otherwise, directly applying this operator in actual scheduling scenarios will result in distortion. Finally, by normalizing and fusing the predicted values ​​of the joint response of the coupled units, a schedulable health factor is obtained, which is then used to screen and obtain scheduling decision schemes. For the first time, the equipment status monitoring data is presented in a quantitative form that can be directly used in scheduling decision comparison. This allows the algorithm output to form a direct correspondence with the decision variables of the specific scheduling scenario of network reconfiguration and load transfer, rather than remaining at the level of alarm information at the equipment operation and maintenance level. This demonstrates that only when the evaluation algorithm is translated into a specific numerical form that can be used for scheduling decisions can it truly make a practical contribution in the field of power grid scheduling. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of one embodiment of the gas-filled high-voltage switchgear status monitoring method in this application. Figure 2 This is a schematic diagram comparing the prediction errors of different prediction methods across multiple monitoring dimensions in the embodiments of this application; Figure 3 This is a schematic diagram illustrating how the coupling unit can schedule the evolution of health factors over a monitoring cycle and trigger preventative maintenance markers in an embodiment of this application. Detailed Implementation

[0011] This application provides a method for monitoring the status of a gas-filled high-voltage switchgear. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0012] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the gas-filled high-voltage switchgear status monitoring method in this application includes: Step S1: Perform coupling relationship analysis on the multi-dimensional status monitoring data of the switchgear to obtain the coupling unit; Specifically, the step of analyzing the coupling relationships of multi-dimensional status monitoring data of switchgear to obtain coupling units involves addressing the fact that the coupling relationship is not a statistical correlation, but rather refers to the actual physical conduction paths between switchgear units. These paths specifically manifest as heat conduction within the same switchgear compartment, current distribution and conduction on the same busbar segment, and airflow coupling within the same cooling duct. These three types of relationships are chosen as the basis for judgment because they correspond to the three main physical mechanisms of stress conduction within the switchgear, avoiding the establishment of coupling relationships based on abstract feature similarities lacking physical basis. This ensures that the division of coupling units has an interpretable physical origin. The acquisition cycle of the multi-dimensional status sensors is set to sixty seconds. This value is derived from the typical timescale of recurring partial discharge pulses and the response speed of temperature rise changes in the switchgear. A value shorter than this would result in unnecessary consumption of edge terminal storage and computing power, while a value longer might miss the recurring characteristics of partial discharge pulses. This step addresses the problem mentioned in the background technology of existing monitoring data being disconnected from the actual physical layout of the equipment, and the failure to recognize stress superposition effects due to multiple devices being treated in isolation.

[0013] Step S2: Perform response prediction processing on the coupled unit using the coupled response prediction model to obtain the joint response prediction function relationship; Specifically, the step of obtaining the joint response prediction function relationship by performing response prediction processing on the coupled units through the coupled response prediction model refers to a mapping relationship with the current state of the equipment and a candidate scheduling action that has not yet been executed as input variables. Its output is not the current health level of the equipment, but the predicted increment of the equipment state quantity after executing the candidate action. The candidate scheduling action parameters are limited to three components: load increment, operation number increment, and runtime. This setting corresponds to the three main aging paths of the switchgear: load increment corresponds to the thermal aging of contacts and buses, operation number increment corresponds to mechanical wear, and runtime corresponds to insulation cumulative aging. These three already cover the main degradation mechanisms of the switchgear, and there is no need to introduce more redundant dimensions to increase the model complexity.

[0014] Step S3: Perform Jacobian solution processing on the joint response prediction function relationship at the current operating point of the coupled unit to obtain the reduced-order coupling operator and the baseline response. Based on the candidate scheduling action parameters in the candidate scheduling scheme, substitute the measured baseline values ​​into the coupled devices in the coupled unit that are not activated by the candidate scheduling scheme, and perform partial activation boundary consistency substitution processing on the reduced-order coupling operator to obtain the joint response prediction value of the coupled unit corresponding to the candidate scheduling scheme. Specifically, at the current operating point of the coupled unit, the joint response prediction function relationship is processed by Jacobian calculation to obtain the reduced-order coupling operator and baseline response. Then, the measured baseline values ​​are substituted into the coupled devices not activated by candidate scheduling schemes. The reduced-order coupling operator is essentially the first-order rate of change of the joint response prediction function relationship with respect to the action parameters near a specific operating point. Caching this rate of change avoids re-executing a complete model inference for each candidate scheme evaluation. The update cycle of this operator is set to fifteen minutes. This value is taken from the typical time constant of field load fluctuations and equipment temperature drift. A cycle shorter than this is insufficient to reflect equipment state changes, while a cycle longer will lag behind the actual equipment degradation progress. Fifteen minutes is a compromise between computational overhead and prediction timeliness. The reason for substituting measured baseline values ​​into the coupled devices not activated by candidate schemes, instead of setting them to zero or directly substituting candidate values, is that setting them to zero would erase the device's existing true operating baseline, while substituting candidate values ​​would fabricate actions that the device has not actually performed. Both approaches would cause the prediction results to deviate from the actual physical state of the equipment.

[0015] Step S4: Normalize and fuse the candidate scheduling schemes based on the joint response prediction values ​​of the coupled units to obtain the schedulable health factor, and then select the scheduling decision scheme from the schedulable health factor.

[0016] Specifically, the step of normalizing and fusing candidate scheduling schemes based on the joint response prediction values ​​of coupled units to obtain schedulable health factors, and then using these schedulable health factors to select scheduling decision schemes, differs fundamentally from the static health score in existing technologies. It is not an isolated value given for the current state of a single device, but rather a path-level risk quantity obtained after comprehensively considering the responses of all coupled devices involved in a specific candidate scheduling path. The physical degradation limits used for normalization are derived from the warning thresholds determined by equipment technical specifications and historical fault statistics, while the weight vector used for weighted summation comes from the statistical proportion of each monitoring dimension in historical faults.

[0017] In one specific embodiment, step S1 includes: By using a multi-dimensional state sensor array consisting of gas sensors, partial discharge sensors, mechanical characteristic sensors, and temperature sensors deployed in the switchgear, gas state parameters, partial discharge parameters, mechanical characteristic parameters, and temperature parameters are collected to obtain multi-dimensional state monitoring data of the switchgear. The multi-dimensional status monitoring data of the switchgear is processed into feature vectors to obtain multi-dimensional status feature vectors. Based on the switchgear compartment number, busbar segment number, and heat dissipation duct number recorded in the equipment ledger, the coupling type of the switchgear corresponding to the multidimensional state feature vector is determined by the relationship between shared compartment, shared busbar segment, and shared duct, and a physical coupling diagram is obtained. Based on the physical coupling graph, coupling weight assignment and connected subgraph partitioning are performed on the switch cabinet to obtain coupling units.

[0018] Specifically, the gas state parameters include two items: SF6 gas pressure deviation rate and SF6 gas moisture content; the partial discharge parameters include two items: UHF partial discharge amplitude and ultrasonic partial discharge signal energy; the mechanism characteristic parameters include three items: opening time deviation, closing time deviation, and energy storage motor current peak value; and the temperature parameters include two items: contact temperature and bus temperature. Each of these nine original parameters is normalized according to its historical value range under normal operating conditions, mapping each original parameter to a dimensionless value between zero and one. These parameters are then arranged and combined in a fixed order: gas state parameters, partial discharge parameters, mechanism characteristic parameters, and temperature parameters, forming a multidimensional state feature vector. Each component of this vector maintains the same arrangement position and dimensional meaning across different switchgear units, ensuring comparability of feature vectors between different switchgear units. The equipment ledger is a structured record table registered by maintenance personnel after the equipment installation or modification at the station. It records the switchgear compartment number, the busbar segment number it is connected to, and the heat dissipation duct number of each switchgear. These three numbers remain fixed until the equipment undergoes physical modification and do not change with the collection of monitoring data. This data is existing data that can be directly read and accessed by this method. The criteria for determining shared compartment relationships are whether the switchgear compartment numbers of the two switchgears are the same; the criteria for determining shared busbar segment relationships are whether the busbar segment numbers connected to the two switchgears are the same; and the criteria for determining shared duct relationships are whether the heat dissipation duct numbers of the two switchgears are the same. As long as the two switchgears have at least one of the above three numbers in common, an edge is established between the nodes corresponding to the two switchgears. The relationship type of the same number is used as the coupling type label of the edge, thus forming a physical coupling graph with switchgears as nodes and number comparison results as edges.

[0019] The coupling weight assignment process calculates a weight value for each edge in the physical coupling graph. This weight value is calculated by weighting three physical quantities: the distance between the shared equipment in the common compartment of the two switchgear connected by the edge, the current carrying ratio of the shared busbar section, and the cross-sectional ratio of the shared air duct. The values ​​of each of the three physical quantities are first normalized to between 0 and 1. The closer the distance, the higher the current carrying ratio, and the larger the cross-sectional ratio of the shared air duct, the closer the weight value is to 1, indicating a stronger physical conduction relationship between the two switchgears. The connected subgraph partitioning process disconnects edges in the physical coupling graph whose weight values ​​are lower than a preset coupling strength threshold, retaining only edges whose weight values ​​are not lower than the threshold. The set of nodes that are still interconnected through the retained edges after disconnection each forms a coupling unit. Switchgears that do not retain connections with any other switchgear form a coupling unit containing only themselves. The preset coupling strength threshold is set to 0.3. This value is taken from the lower quartile of the weight distribution of various coupling relationships in the historical statistics of the station. If the threshold is set too low, switchgear with negligible actual transmission effects will be mistakenly included in the same coupling unit, leading to unnecessary cross-interference in subsequent joint response prediction. If the threshold is set too high, switchgear with obvious transmission relationships will be incorrectly split into independent units, omitting the stress superposition effect. Taking the lower quartile as the boundary can retain most of the connection relationships that have been verified to have obvious transmission effects in history, while removing connection relationships with obviously low weights.

[0020] In one specific embodiment, step S2 includes: The state coding unit encodes the multi-dimensional state monitoring data of the switchgear in the coupling unit to obtain the state hidden vector of the switchgear in the coupling unit. Obtain the candidate scheduling action parameters from the candidate scheduling schemes, and concatenate the state latent vector and the candidate scheduling action parameters to obtain the concatenated feature vector; The concatenated feature vector is input into the local response subnetwork for response prediction processing to obtain the local response prediction value of the switch cabinet within the coupled unit; Based on the physical coupling graph corresponding to the coupling unit, graph convolution propagation processing is performed on the local response prediction values ​​to obtain the joint response prediction function relationship.

[0021] Specifically, the state coding unit adopts a single-layer gated recurrent unit network with an input dimension of 9, corresponding to the 9 components in the multidimensional state feature vector obtained by normalization and combination. The hidden layer dimension is set to 16, which is taken from the empirical estimate of the number of aging modes that may be reflected by the 9 original monitoring quantities and their combination. Too low a dimension will cause the model to be unable to fully express the differences between different degradation modes such as partial discharge, gas degradation, mechanical wear, and temperature anomalies. Too high a dimension will cause overfitting under the condition of limited sample size of historical data in the field. 16 dimensions achieve a balance between the two. The state encoding unit encodes the feature vector sequence of the switchgear for the most recent 20 acquisition cycles. The 20 cycles correspond to a time window of about 20 minutes under the previously set 60-second acquisition interval. The length of this window is taken from the typical observation time required for the partial discharge signal to form a stable repetitive pattern from its appearance. If the window is too short, it will not be able to capture the repetitive features of the partial discharge. If the window is too long, it will introduce outdated data in which the equipment state has changed. The 16-dimensional real vector output after encoding is the state latent vector. This vector is not a direct mapping of a single monitoring quantity, but a comprehensive state expression obtained by compressing the interaction of 9 monitoring quantities in the time dimension. Each dimension component in the vector does not correspond to a single physical quantity, but is a combination feature of multiple monitoring quantities automatically learned during the training process of the gated loop unit. The candidate scheduling action parameters use the three previously defined components: the proposed load transfer increment, the proposed number of operations, and the expected duration of operation. This 3-dimensional vector is combined with the 16-dimensional state latent vector by concatenating the first and last vectors along the vector dimensions to obtain a 19-dimensional concatenated feature vector. This concatenation method allows the subsequent network to simultaneously receive two types of information in the same computation: the current real state of the device and the hypothetical actions that have not yet been executed. This provides the input-level basis for distinguishing between the natural evolution of the device and the response caused by the actions.

[0022] The concatenated feature vectors are input into a local response subnetwork, which is configured as a two-layer feedforward neural network. The first layer has an input dimension of 19 and an output dimension of 32. The 32 dimension is taken from an approximate doubling of the input dimension to expand the feature interaction space before compression, so that the interaction terms between the original state information and action parameter information in the 19-dimensional input can be fully expressed. The output of the first layer is processed by a linear rectified function and then fed into the second layer. The second layer has an input dimension of 32 and an output dimension of 8. The 8-dimensional output corresponds to the contact temperature increment, bus temperature increment, SF6 gas pressure deviation rate increment, and SF6 moisture content, respectively. The predicted increments of eight monitored quantities—increment, UHF partial discharge amplitude increment, ultrasonic partial discharge signal energy increment, opening time deviation increment, and closing time deviation increment—are for the next 24 hours. This 24-hour prediction period is taken from the daily rolling scheduling cycle typically used for network reconfiguration and load transfer decisions, which is consistent with the time granularity of actual scheduling scenarios. The cumulative operation count percentage is not included in the output because this value can be directly calculated from the proposed increase in operation count component in the candidate scheduling action parameters and the rated operation count recorded in the equipment ledger, without the need for network prediction. After obtaining the local response prediction values, the local response prediction values ​​of all switchgear within the same coupling unit are stacked vertically according to the arrangement order of the switchgear within the coupling unit, forming a matrix with 8 rows equal to the number of switchgears in the coupling unit. This matrix, along with the adjacency submatrix corresponding to the coupling unit previously extracted based on the physical coupling diagram, is substituted into the graph convolution propagation calculation. During the calculation, the adjacency submatrix is ​​first normalized according to the number of connected edges of each node to obtain the degree matrix. The degree matrix is ​​then used to normalize both sides of the adjacency submatrix to avoid numerical imbalance caused by excessive differences in the number of connected edges of nodes. Finally, the normalized adjacency submatrix is ​​multiplied by the local response stacking matrix. The result of the multiplication reflects the local response prediction values ​​of each switchgear. The local response of the switchgear is propagated to its coupled neighbors through the physical coupling edges, and the result is multiplied with a trainable weight matrix and processed by a linear rectification function to obtain the joint response of all switchgear in the coupled unit after considering the coupling propagation effect. This joint response is a function relationship composed of the current weights of the graph convolutional layer, the coupling submatrix, and the local response stacking matrix. This is the joint response prediction function relationship of the coupled unit. The specific value of this function relationship will change with the actual monitoring status of the switchgear in the coupled unit and the value of the candidate scheduling action parameters. It is not a fixed value, but retains the dependence on these two types of input variables.

[0023] In one specific embodiment, step S3 involves performing a Jacobian solution on the joint response prediction function relationship at the current operating point of the coupling unit to obtain the reduced-order coupling operator and the baseline response, including: The candidate scheduling action parameters of the switchgear in the coupling unit are concatenated to obtain the candidate scheduling action parameter concatenation vector; Based on the implicit state vector of the switch cabinet within the coupling unit, the current operating point of the coupling unit is calibrated to obtain the current operating point. Based on the current operating point, the partial derivative of the joint response prediction function relationship with respect to the concatenated vector of candidate scheduling action parameters is solved to obtain the reduced-order coupling operator. The candidate scheduling action parameter concatenation vector after being zeroed is substituted into the joint response prediction function relationship for value processing to obtain the baseline response.

[0024] Specifically, the candidate scheduling action parameter concatenation vector is obtained by combining the 3D candidate scheduling action parameters of each switchgear in the coupling unit according to the order of the switchgear in the coupling unit. The candidate scheduling action parameter concatenation vector corresponding to a coupling unit containing 3 switchgears has a dimension of 9, and the candidate scheduling action parameter concatenation vector corresponding to a coupling unit containing 2 switchgears has a dimension of 6. The arrangement order of this vector is consistent with the arrangement order of the switchgears in the aforementioned graph convolution propagation calculation, ensuring that the switchgear identity corresponding to each component in the subsequent matrix operation is not misaligned. The current working point calibration process refers to extracting the 16-dimensional state latent vector of each switchgear in the coupling unit at the current moment in the same arrangement order and fixing its value, which serves as the reference point for the state latent vector part to remain unchanged when calculating the partial derivative in the subsequent process. This reference point only uses the latest state latent vector obtained in the current acquisition cycle when solving the reduced-order coupling operator, and does not use the state latent vector of the historical cycle, ensuring that the solved reduced-order coupling operator reflects the response characteristics of the equipment in the current actual state, rather than the response characteristics in the outdated state.

[0025] Partial derivative calculation refers to taking the first-order partial derivative of each output component in the joint response prediction function with respect to each component in the candidate scheduling action parameter concatenation vector, given that the implicit state vector is fixed at the current operating point. The partial derivatives of all output components with respect to all input components are then arranged by row and column to form a reduced-order coupling operator matrix. The number of rows in the matrix equals the number of switchgear in the coupling unit multiplied by 8, and the number of columns equals the dimension of the candidate scheduling action parameter concatenation vector. Taking a coupling unit containing 3 switchgears as an example, the reduced-order coupling operator matrix has 24 rows and 9 columns. The value in the first row and first column of the matrix represents the rate of change of the first output component of the first switchgear relative to the load increment of the first switchgear itself. The value in the first row and fourth column of the matrix represents the rate of change of the first output component of the first switchgear relative to the load increment of the second switchgear. The latter value being non-zero reflects the physical coupling and transmission relationship between the devices within the coupling unit. The baseline response is the output value calculated by setting all components of the concatenated vector of candidate scheduling action parameters to 0 and then inputting it into the joint response prediction function relationship. The number of rows is the same as the number of rows of the reduced-order coupling operator matrix, and the number of columns is 1. This value represents the predicted response of each switchgear in the coupling unit after 24 hours of operation without any additional scheduling action and only based on the current state latent vector according to the natural evolution law. The reduced-order coupling operator matrix and the baseline response together form the basis for subsequent response prediction when different candidate scheduling action parameter concatenated vector values ​​are substituted. There is no need to re-execute the complete calculation process of state encoding and graph convolution propagation for each specific candidate scheduling scheme.

[0026] In one specific embodiment, step S3 involves substituting measured baseline values ​​into the couplers in the coupler unit that were not activated by the candidate scheduling scheme, based on the candidate scheduling action parameters in the candidate scheduling scheme. This includes: The candidate scheduling schemes are parsed to obtain the set of active switchgear. Based on the set of activated switchgear, the activation status of the switchgear in the coupling unit is calibrated to obtain the inactive coupling equipment. The measured baseline values ​​were extracted from the multi-dimensional status monitoring data of the switchgear without activated coupling devices to obtain the measured baseline values. Based on the candidate scheduling action parameter splicing vector and the measured baseline value, the action parameters of the switchgear in the coupling unit are substituted and assembled to obtain the substitution vector.

[0027] Specifically, the candidate scheduling scheme clearly records the switchgear numbers involved in the proposed scheduling operation and their corresponding candidate scheduling action parameter values. The candidate scheduling scheme undergoes activation switchgear parsing, which involves reading the switchgear numbers recorded in the candidate scheduling scheme and determining the set of these numbers as the activated switchgear set. The activation status marking process compares the numbers of all switchgear within the coupled unit with the activated switchgear set one by one. Switchgear whose numbers are present in the activated switchgear set are marked as activated, while switchgear whose numbers are not present are marked as inactive coupled devices. For example, in a coupled unit containing 3 switchgears, if the candidate scheduling scheme only applies a scheduling action to the first switchgear, then the activated switchgear set only contains the number of the first switchgear; the second and third switchgears are both marked as inactive coupled devices.

[0028] The measured baseline value extraction process involves extracting three current status values ​​from the multi-dimensional status monitoring data of each inactive coupled device: the current actual load level, the current operation count converted from the current cumulative operation count, and the actual running time since the last maintenance. These three values ​​correspond one-to-one with the three components of load increment, operation count increment, and running time in the candidate scheduling action parameters. The difference is that the former is the current actual operating status value of the device, while the latter is the additional scheduling increment to be applied by the candidate scheme. Inactive coupled devices were not selected by the candidate scheme and there is no additional scheduling increment applied. Therefore, their current status value is filled in the corresponding position instead of 0. The purpose is to ensure that the position corresponding to the device in the matrix multiplication operation can reflect its actual contribution to the overall response of the coupled unit when the current actual operating condition remains unchanged, rather than being mistakenly regarded as a blank position that does not participate in any calculation. The action parameter substitution and assembly process involves keeping the original candidate values ​​of the 3D components corresponding to the activated switchgear in the candidate scheduling action parameter splicing vector unchanged, replacing the 3D components corresponding to the inactive coupling devices with the measured baseline values ​​of those devices, and recombining them according to the same switchgear arrangement order as the candidate scheduling action parameter splicing vector to obtain a substitution vector with the same dimension as the candidate scheduling action parameter splicing vector. Taking the aforementioned coupling unit containing 3 switchgears as an example, the substitution vector has a dimension of 9, where the 1st to 3rd dimensions are taken from the original values ​​of the candidate scheduling action parameters of the 1st switchgear, the 4th to 6th dimensions are taken from the measured baseline values ​​of the 2nd switchgear, and the 7th to 9th dimensions are taken from the measured baseline values ​​of the 3rd switchgear. This substitution vector matches the dimension of the previously solved 24-row 9-column reduced-order coupling operator matrix.

[0029] In one specific embodiment, step S3 involves performing partial activation boundary consistency substitution processing on the reduced-order coupling operator to obtain the joint response prediction value of the coupling units corresponding to the candidate scheduling scheme, including: The reduced-order coupling operator and the substituted vector are subjected to matrix-vector multiplication to obtain the coupled conduction response increment. The coupled conducted response increment and the baseline response are superimposed to obtain the preliminary response prediction value of the coupled unit; Based on the measured baseline values ​​of the unactivated coupling devices, the component corresponding to the unactivated coupling devices in the preliminary response prediction values ​​of the coupling unit is subjected to boundary consistency correction processing to obtain the boundary correction response prediction values. The predicted values ​​of the boundary correction response are calibrated using candidate schemes to obtain the joint response prediction values ​​of the coupled units corresponding to the candidate scheduling schemes.

[0030] Specifically, the matrix-vector multiplication process involves multiplying the previously obtained 24-row, 9-column reduced-order coupling operator matrix with the 9-dimensional substitution vector to obtain a 24-dimensional column vector. This column vector is denoted as the coupling-transferred response increment. The value of each row is equal to the sum of the products of the corresponding row of the reduced-order coupling operator matrix and the corresponding column of the substitution vector. For example, the value corresponding to the output component in the first row of the second switchgear is equal to the sum of the products of that row and each of the nine components of the substitution vector. This summation includes both the contribution of the second switchgear's own candidate action parameters or current status values, and the portion transmitted from the first and third switchgear through the coupling relationship. The coupling-transferred response increment is then added to the previously obtained 24-dimensional baseline response, item by item, according to the corresponding row positions, to obtain the preliminary response prediction value of the coupled unit. This superposition process ensures that the preliminary response prediction value simultaneously includes the baseline portion of the equipment's natural evolution and the linearly approximated action and coupling-transferred portions.

[0031] Since the reduced-order coupling operator matrix is ​​a first-order approximation obtained for small changes in the action parameters near 0, with the state implicit vector fixed at the current operating point, this approximation will produce errors in regions where the action parameters are far from 0. The measured baseline values ​​substituted into the unactivated coupling device are often not small quantities close to 0. For example, if the actual load of a switchgear is already operating at a relatively high proportion of its rated capacity, directly using the reduced-order coupling operator to linearly extrapolate this larger value will cause the estimated value of the device's own response to deviate from the true value. Therefore, it is necessary to perform boundary consistency correction on the 8-dimensional components corresponding to the unactivated coupling device in the preliminary response prediction value of the coupling unit. Specifically, the measured baseline value of the unactivated coupling device is substituted separately into the joint response prediction function relationship to recalculate once, obtaining the accurate response value of the device under its actual current state, and then from... In the preliminary response prediction value of the coupled unit, the portion obtained by multiplying the sub-block belonging to the diagonal position of the device itself in the reduced-order coupling operator matrix by its measured baseline value is subtracted from the 8-dimensional component corresponding to the device. Only the portion transmitted from the candidate action parameters of other activated switchgear through the coupling relationship is retained. This retained portion is added to the previously recalculated accurate response value to obtain the corrected response value of the inactive coupled device. The components corresponding to the activated switchgear retain their values ​​from the preliminary response prediction value of the coupled unit without modification. The components corresponding to all switchgear are recombined in their original order to obtain the boundary correction response prediction value. This process allows the response of the activated device to retain the computational efficiency of linear approximation, while the response of the inactive device is changed to accurate calculation. The coupling transmission part still uses linear approximation, balancing computational efficiency and prediction accuracy. The candidate scheme calibration process for the boundary correction response prediction value involves establishing a correspondence between the 24-dimensional vector and its corresponding candidate scheduling scheme number and recording and storing it. The result after calibration is the joint response prediction value of the coupled unit corresponding to the candidate scheduling scheme.

[0032] Figure 2 This diagram illustrates the comparison of prediction errors across multiple monitoring dimensions using different prediction methods in this application's embodiments. The horizontal axis represents five monitoring dimensions: contact temperature, busbar temperature, SF6 pressure, partial discharge amplitude, and tripping time. The vertical axis represents the proportion of the prediction root mean square error to the corresponding monitoring dimension's range. For each monitoring dimension, the diagram provides the prediction error values ​​for the three methods: the unit-by-unit independent prediction method, the full-map repeated inference method, and the reduced-order coupling operator assembly method described in this invention. The unit-by-unit independent prediction method does not consider the physical coupling relationship between switchgear. While the full-map repeated inference method considers the coupling relationship, it does not perform boundary consistency processing for situations where some devices in the candidate scheduling scheme are not activated. The prediction error of the method described in this invention across all five monitoring dimensions is lower than the prediction errors of the other two methods.

[0033] In one specific embodiment, step S4 includes: Based on the preset physical degradation limit, the predicted value of the joint response of the coupled units is normalized to obtain the normalized response value. Based on the preset weight vector, the normalized response values ​​are weighted and summed to obtain the path risk index; Based on path risk indicators, health factors are transformed to obtain schedulable health factors. Based on preset health factor thresholds, candidate scheduling schemes are screened to obtain scheduling decision schemes.

[0034] Specifically, preset physical degradation limits are set for the predicted increments of the aforementioned eight monitoring quantities, namely: contact temperature increment limit of 15 degrees Celsius, busbar temperature increment limit of 12 degrees Celsius, SF6 gas pressure deviation rate increment limit of 5%, SF6 moisture content increment limit of 3 microliters per liter, ultra-high frequency partial discharge amplitude increment limit of 10 dB, ultrasonic partial discharge signal energy increment limit of 8 energy units, opening time deviation increment limit of 2 milliseconds, and closing time deviation increment limit of 2 milliseconds. The above eight values ​​are taken from the smaller of the upper limit of normal operation given in the equipment manufacturer's technical specifications and the lower limit of the increment value range observed before the fault occurred in the historical fault statistics. The reason for taking the smaller value is to enable the limit to be triggered at an earlier stage before the equipment actually fails, and to avoid missing the time window for preventive treatment due to the limit setting being too lenient. Normalization is performed on the joint response prediction value or boundary correction response prediction value of each coupled unit involved in the candidate scheduling scheme. The predicted increment of each monitoring quantity of each switch cabinet is divided by the preset physical degradation limit corresponding to the monitoring quantity. If the division result is less than 0, it is taken as 0. If the division result is greater than 1, the original value is retained without upper limit truncation. The value obtained is the normalized response value. The larger the value, the closer the predicted increment is to its degradation limit.

[0035] The preset weight vector contains eight weight values ​​corresponding to eight monitoring quantities. Each weight value is calculated based on the proportion of times each monitoring dimension appears as the primary cause of a fault in the historical fault records of this station. The sum of the eight weight values ​​is 1. A higher weight value indicates a stronger historical correlation between that monitoring quantity and equipment faults. The weighted summation process involves multiplying the normalized response values ​​of all switchgear and all eight monitoring quantities affected by the candidate scheduling scheme by their respective weight values, summing them, and then dividing by the total number of switchgear affected by the candidate scheduling scheme. The quotient is the path risk index, which is a non-negative number above 0. A larger value indicates a higher overall equipment degradation risk after the candidate scheduling scheme is implemented. The health factor conversion process involves subtracting the path risk index from 1. If the result is less than 0, it is set to 0. The resulting value is the schedulable health factor, which ranges from 0 to 1. A value closer to 1 indicates a lower equipment health risk corresponding to the candidate scheduling scheme. The preset health factor threshold is 0.6. This value is taken from the upper limit of the schedulable health factor value range corresponding to scheduling schemes that have experienced unplanned outages in historical execution records. Setting the threshold at this upper limit ensures that candidate schemes with schedulable health factors below this threshold correspond to risk levels where equipment failures have historically occurred. The screening process compares the schedulable health factors of all candidate scheduling schemes with the preset health factor threshold. Candidate scheduling schemes with schedulable health factors not lower than the preset health factor threshold are retained in the executable scheme set. They are then sorted from low to high according to their respective network loss indicators. The candidate scheduling scheme ranked first is selected as the scheduling decision scheme and issued for execution. Candidate scheduling schemes with schedulable health factors lower than the preset health factor threshold are removed from the executable scheme set and are not issued as scheduling decision schemes.

[0036] In one specific embodiment, the preset physical degradation limit is determined based on the equipment technical specification limits and historical fault statistical early warning thresholds corresponding to the monitoring dimensions of the switchgear multi-dimensional status monitoring data. The preset weight vector is determined based on the historical fault proportion statistics of the monitoring dimensions corresponding to the switchgear multi-dimensional status monitoring data, and the sum of all components in the preset weight vector is 1. If the schedulable health factor is lower than the preset health factor threshold, the candidate scheduling schemes corresponding to the schedulable health factor are marked to obtain high-risk candidate scheduling schemes, and preventive maintenance instructions are generated based on the high-risk candidate scheduling schemes.

[0037] Specifically, the equipment technical specification limits are taken from the normal operating limits of the corresponding monitoring dimensions recorded in the technical parameter manual provided by the switchgear manufacturer with the equipment. For example, for contact temperature, the equipment technical specification limit is 75 degrees Celsius. The historical fault statistical warning threshold is taken from the historical fault records of this station and switchgear of the same model. Within the 24-hour observation window before the fault occurred, the increment of the monitored quantity relative to the average value of the 7 days before the fault occurred is taken. The increment values ​​of this item in all historical fault samples are arranged from smallest to largest, and the value at the lower quartile of the arrangement is taken as the historical fault statistical warning threshold. Taking contact temperature as an example again, if the lower quartile of the increment value of this item in the historical fault samples is 18 degrees Celsius, then the historical fault statistical warning threshold is 18 degrees Celsius. The preset physical degradation limit is determined by taking the smaller of the incremental value corresponding to the limit in the equipment technical specifications and the historical fault statistics warning threshold. For example, if the difference between the incremental value corresponding to the limit in the equipment technical specifications and the lower operating limit for the contact temperature is 20 degrees Celsius and the historical fault statistics warning threshold is 18 degrees Celsius, the smaller of the two, 18 degrees Celsius, is the final preset physical degradation limit for that item. The reason for taking the smaller value has been explained in the supplementary explanation of step four above and will not be repeated here.

[0038] Historical fault percentage statistics refer to the statistical analysis of all historical fault records of the station and switchgear of the same model. For each fault, the direct cause is determined according to the monitoring dimension. The number of times each of the eight monitoring dimensions appears as a direct cause is divided by the total number of fault records, resulting in eight percentage values. These eight percentage values ​​are directly used as the component values ​​in the preset weight vector. Since the sum of the eight percentage values ​​is equal to 1 by definition, the condition that the sum of all components in the preset weight vector is 1 is guaranteed by the statistical method itself, requiring no additional normalization. When the schedulable health factor is lower than the preset health factor threshold, a high-risk marker field is added to the candidate scheduling scheme corresponding to that schedulable health factor. The marker field records the candidate scheduling scheme number, the corresponding schedulable health factor value, and the monitoring dimension that triggered the marker. The candidate scheduling scheme after adding the marker field is the high-risk candidate scheduling scheme. The process of generating preventive maintenance instructions based on high-risk candidate scheduling schemes involves reading the monitoring dimensions and corresponding switchgear numbers of the trigger markers in the high-risk candidate scheduling schemes, and then filling the switchgear number, the monitoring dimension name of the trigger marker, and the corresponding normalized response value into the preventive maintenance instruction. The preventive maintenance instruction includes three items: the switchgear number to be inspected, the monitoring dimensions that need to be checked, and the corresponding normalized response value. This is used to prompt maintenance personnel to arrange maintenance and verification of the switchgear and monitoring dimensions that trigger high-risk markers after the scheduling action corresponding to the candidate scheduling scheme has been excluded from execution.

[0039] Figure 3This is a schematic diagram illustrating the evolution of the schedulable health factor of the coupled unit during the monitoring cycle and the triggering of preventive maintenance marking in an embodiment of this application. The horizontal axis represents the monitoring cycle number, and the vertical axis represents the schedulable health factor value. The diagram shows the curves of the schedulable health factor of each of the two coupled units during the monitoring cycle, as well as the horizontal line containing the preset health factor threshold. One curve shows that the schedulable health factor of the coupled unit continuously decreases and falls below the preset health factor threshold over multiple monitoring cycles. The monitoring cycle intervals where this curve falls below the preset health factor threshold are marked with shaded areas in the diagram. The corresponding candidate scheduling schemes within this interval are marked as high-risk candidate scheduling schemes, and preventive maintenance instructions are generated accordingly. The other curve shows that the schedulable health factor of the coupled unit remains above the preset health factor threshold throughout the entire monitoring cycle, without triggering the marking process.

[0040] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring the status of a gas-filled high-voltage switchgear, characterized in that, The method includes: Step S1: Perform coupling relationship analysis on the multi-dimensional status monitoring data of the switchgear to obtain the coupling unit; Step S2: Perform response prediction processing on the coupled unit using the coupled response prediction model to obtain the joint response prediction function relationship; Step S3: Perform Jacobian solution processing on the joint response prediction function relationship at the current operating point of the coupling unit to obtain the reduced-order coupling operator and the baseline response. Based on the candidate scheduling action parameters in the candidate scheduling scheme, substitute the measured baseline values ​​into the coupling devices in the coupling unit that are not activated by the candidate scheduling scheme. Perform partial activation boundary consistency substitution processing on the reduced-order coupling operator to obtain the joint response prediction value of the coupling unit corresponding to the candidate scheduling scheme. Step S4: Normalize and fuse the candidate scheduling schemes according to the joint response prediction values ​​of the coupled units to obtain the schedulable health factors, and select the scheduling decision schemes from the schedulable health factors.

2. The method for monitoring the status of a gas-filled high-voltage switchgear according to claim 1, characterized in that, Step S1 includes: By using a multi-dimensional state sensor array consisting of gas sensors, partial discharge sensors, mechanical characteristic sensors, and temperature sensors deployed in the switchgear, gas state parameters, partial discharge parameters, mechanical characteristic parameters, and temperature parameters are collected to obtain multi-dimensional state monitoring data of the switchgear. The multi-dimensional status monitoring data of the switchgear is processed into feature vectors to obtain multi-dimensional status feature vectors; Based on the switch cabinet compartment number, bus section number, and heat dissipation duct number recorded in the equipment ledger, the coupling type of the switch cabinets corresponding to the multidimensional state feature vector is determined by the relationship between shared compartment, shared bus section, and shared duct, and a physical coupling diagram is obtained. Based on the physical coupling diagram, the switch cabinet is subjected to coupling weight assignment and connected subgraph partitioning to obtain coupling units.

3. The method for monitoring the status of a gas-filled high-voltage switchgear according to claim 1, characterized in that, Step S2 includes: The state coding unit encodes the multi-dimensional state monitoring data of the switch cabinet within the coupling unit to obtain the state hidden vector of the switch cabinet within the coupling unit. Obtain candidate scheduling action parameters from the candidate scheduling schemes, and concatenate the state latent vector and the candidate scheduling action parameters to obtain a concatenated feature vector. The spliced ​​feature vector is input into the local response sub-network for response prediction processing to obtain the local response prediction value of the switch cabinet in the coupling unit; Based on the physical coupling graph corresponding to the coupling unit, the local response prediction value is processed by graph convolution propagation to obtain the joint response prediction function relationship.

4. The method for monitoring the status of a gas-filled high-voltage switchgear according to claim 1, characterized in that, In step S3, the joint response prediction function relationship is subjected to Jacobian solution processing at the current operating point of the coupling unit to obtain the reduced-order coupling operator and the baseline response, including: The candidate scheduling action parameters of the switch cabinet within the coupling unit are concatenated to obtain a concatenated vector of candidate scheduling action parameters. Based on the implicit state vector of the switch cabinet within the coupling unit, the current operating point of the coupling unit is calibrated to obtain the current operating point. Based on the current operating point, the partial derivative of the joint response prediction function relationship with respect to the concatenated vector of candidate scheduling action parameters is calculated to obtain a reduced-order coupling operator. The candidate scheduling action parameter concatenation vector, after being zeroed out, is substituted into the joint response prediction function relationship for value processing to obtain the baseline response.

5. The method for monitoring the status of a gas-filled high-voltage switchgear according to claim 4, characterized in that, In step S3, based on the candidate scheduling action parameters in the candidate scheduling scheme, the measured baseline values ​​are substituted into the coupling devices in the coupling unit that are not activated by the candidate scheduling scheme, including: The candidate scheduling scheme is parsed to obtain a set of activated switchgear; Based on the set of activated switch cabinets, the activation status of the switch cabinets in the coupling unit is calibrated to obtain the inactive coupling devices. The measured baseline values ​​are extracted from the multi-dimensional status monitoring data of the switchgear without activated coupling devices to obtain the measured baseline values. Based on the candidate scheduling action parameter splicing vector and the measured baseline value, the action parameters of the switch cabinet in the coupling unit are substituted and assembled to obtain the substitution vector.

6. The method for monitoring the status of a gas-filled high-voltage switchgear according to claim 5, characterized in that, In step S3, the reduced-order coupling operator undergoes partial activation boundary consistency substitution processing to obtain the joint response prediction value of the coupling units corresponding to the candidate scheduling scheme, including: Perform matrix-vector multiplication on the reduced-order coupling operator and the substituted vector to obtain the coupled conduction response increment; The coupled conduction response increment and the baseline response are superimposed to obtain the preliminary response prediction value of the coupled unit; Based on the measured baseline value of the unactivated coupling device, the component corresponding to the unactivated coupling device in the preliminary response prediction value of the coupling unit is subjected to boundary consistency correction processing to obtain the boundary correction response prediction value. The predicted boundary correction response is processed by candidate scheme calibration to obtain the joint response prediction of the coupled units corresponding to the candidate scheduling scheme.

7. The method for monitoring the status of a gas-filled high-voltage switchgear according to claim 1, characterized in that, Step S4 includes: Based on the preset physical degradation limit, the predicted value of the joint response of the coupling unit is normalized to obtain the normalized response value. Based on a preset weight vector, the normalized response values ​​are weighted and summed to obtain a path risk index. Based on the path risk indicators, health factors are converted to obtain schedulable health factors. Based on a preset health factor threshold, the candidate scheduling schemes are screened to obtain a scheduling decision scheme.

8. The method for monitoring the status of a gas-filled high-voltage switchgear according to claim 7, characterized in that, The preset physical degradation limit is determined based on the equipment technical specification limit and historical fault statistical early warning threshold of the corresponding monitoring dimension of the switchgear multi-dimensional status monitoring data.

9. The method for monitoring the status of a gas-filled high-voltage switchgear according to claim 7, characterized in that, The preset weight vector is determined based on the historical fault percentage statistics of the corresponding monitoring dimensions of the multi-dimensional status monitoring data of the switch cabinet, and the sum of each component in the preset weight vector is 1.

10. The method for monitoring the status of a gas-filled high-voltage switchgear according to claim 7, characterized in that, If the schedulable health factor is lower than the preset health factor threshold, the candidate scheduling schemes corresponding to the schedulable health factor are marked to obtain high-risk candidate scheduling schemes, and preventive maintenance instructions are generated based on the high-risk candidate scheduling schemes.

Images