A secondary cable operation and maintenance whole-cycle visual management system based on digital twinning

Abstract

The application discloses a secondary cable operation and maintenance whole-cycle visual management system based on digital twinning, and belongs to the technical field of cable monitoring. The system comprises a data acquisition module, a digital twinning module, a time-effectiveness evaluation module, a state evaluation module and a risk evaluation module. The data acquisition module is used for acquiring state data of each monitoring dimension of the secondary cable. The digital twinning module is used for establishing an entity object for each secondary cable in a digital twinning model. The time-effectiveness evaluation module is used for determining a decay rate parameter based on a differential evolution algorithm and calculating a time-effectiveness decay coefficient. The state evaluation module is used for calculating a normalized degradation value and performing weighted summation and averaging of the normalized degradation value of each monitoring dimension and the corresponding time-effectiveness decay coefficient to obtain a single-cable degradation score of the secondary cable. The risk evaluation module selects single-cable degradation scores of part of cables to form a degradation sequence and calculates a failure proximity of a loop based on the degradation sequence. The alarm module is used for generating a loop-level risk alarm. Through the application, the accuracy of secondary cable early warning display can be improved.

Description

Technical Field

[0001] This invention belongs to the field of cable monitoring technology, specifically relating to a visualization management system for the entire lifecycle of secondary cable maintenance based on digital twins. Background Technology

[0002] In existing digital twin operation and maintenance systems, the health status assessment of secondary cables relies on data from multiple monitoring dimensions, including data collected by online sensors and manual inspections. However, the status assessment of secondary cables primarily involves collecting data on insulation resistance, temperature, and partial discharge levels from online monitoring devices, as well as recording the cable appearance during manual inspections. These indicators are then weighted to obtain a health score for each cable, which is displayed in the digital twin model using color coding to indicate the degree of degradation. When a cable's health score falls below a preset threshold, the system issues an alarm and includes that cable in the maintenance plan. However, a circuit may contain multiple secondary cables. When multiple cables in a circuit are simultaneously in a slightly degraded state, the individual scores of each cable do not reach the alarm threshold, and traditional systems do not generate any alerts.

[0003] To address these issues, existing technologies assign importance weights to each cable in a loop and aggregate all scores within the loop into a comprehensive score using a weighted summation method. A loop-level alarm is generated when this comprehensive score exceeds a threshold. However, weighted aggregation misses the cable degradation distribution characteristics. For example, in one loop, a cable with a high weight is severely degraded while the others are in good condition; in another loop, multiple cables are simultaneously moderately degraded with highly similar degradation levels. Both scenarios might yield the same comprehensive score after weighted aggregation. However, the simultaneous approach to failure of multiple cables means they could fail sequentially within a very short time window, leaving maintenance personnel almost no time to address each cable individually. Conversely, a single cable with significant degradation can be eliminated simply by replacing that cable. The weighted aggregation method cannot distinguish between these two risk patterns, thus failing to provide maintenance personnel with an accurate basis for risk prioritization. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a digital twin-based full-cycle visual management system for secondary cable maintenance, thereby resolving the problems described in the background section.

[0005] To achieve the aforementioned objectives, this invention proposes a digital twin-based full-cycle visual management system for secondary cable maintenance, comprising: The data acquisition module is used to acquire status data of various monitoring dimensions of the secondary cable; The digital twin module is used to create an entity object for each secondary cable in the digital twin model. The entity object contains the cable's loop type and cable tray ownership information. The timeliness assessment module is used to determine the decay rate parameter based on the differential evolution algorithm when any monitoring dimension receives new collected data. Based on the duration since the data was generated, the standard update cycle of the monitoring dimension, and the decay rate parameter, it calculates the timeliness decay coefficient of the corresponding data of the monitoring dimension and treats it as a data update event. The status assessment module is used to normalize the status data of the monitoring dimensions to obtain normalized degradation values, and to weight and average the normalized degradation values ​​of each monitoring dimension with the corresponding time-related attenuation coefficient to obtain the degradation score of the secondary cable. The risk assessment module is used to sort all secondary cables in the loop according to the single cable degradation score, select some cables based on the single cable degradation score to form a degradation sequence, and calculate the failure proximity of the loop based on the degradation sequence. The alarm module is used to generate a loop-level risk alarm when the proximity of a failure exceeds a preset alarm threshold, and to mark the corresponding loop and secondary cable.

[0006] Furthermore, the timeliness decay coefficient of the data corresponding to the monitoring dimensions is calculated, including: Obtain the time difference between the current system time and the time of the most recent data record generation. Divide the time difference by the standard update cycle of the corresponding monitoring dimension to obtain the time offset ratio. Calculate the timeliness attenuation coefficient based on the attenuation rate parameter and the time offset ratio. The larger the attenuation rate parameter or the larger the time offset ratio, the smaller the timeliness attenuation coefficient.

[0007] Furthermore, the failure proximity of the circuit is calculated based on the degradation sequence, including: The individual cable degradation scores of each cable in the same secondary circuit are sorted from high to low, and the top m scores are selected to form a degradation sequence, where m is the number of sensitive circuits determined based on the total number of cables in the circuit and the circuit type. The average of all degradation scores in the degradation sequence is calculated as the head concentration, and the difference between the maximum and minimum scores in the degradation sequence is calculated as the head dispersion. The concentration weight and dispersion weight are determined according to the loop type. The failure proximity is calculated based on the concentration weight, dispersion weight, head concentration, and head dispersion. The higher the head concentration and the lower the head dispersion, the greater the loop failure proximity.

[0008] Furthermore, the decay rate parameter is determined based on the differential evolution algorithm, including: Each individual in the Differential Evolutionary Algorithm (DEA) contains a parameter vector. Each dimension of the parameter vector corresponds to the attenuation rate parameter value of the cable in the monitoring dimension. The DEA evaluates the fitness value of an individual based on a preset fitness function. The fitness function is calculated based on the comprehensive optimization weight and the average prediction bias. The comprehensive optimization weight represents the priority of attention assigned to different cables by the DEA during the optimization process, and the average prediction bias represents the error between the data aging rate predicted by the DEA during the optimization process and the actual data change rate monitored. The DEA improves search efficiency by using a subpopulation merging mechanism and improves iteration efficiency by using a subpopulation splitting mechanism during the optimization process.

[0009] Furthermore, the comprehensive optimization weight is calculated based on the failure proximity of the cable loop and the proximity gradient value of the cable. Specifically, the adjacent cables of the cable are determined based on the cable tray affiliation information of each cable in the digital twin model, and the proximity gradient value is calculated based on the failure proximity of the cable loop and the failure proximity of the adjacent cables.

[0010] Furthermore, a reference sample size threshold is set. In the mutation operation of the differential evolution algorithm, for the parameters in the parameter vector, if the cumulative number of data update events is less than the reference sample size threshold, the scaling factor starts with a preset maximum scaling value and decreases linearly as the number of data update events increases. When the number of data update events exceeds the reference sample size threshold, the scaling factor is set to a preset minimum scaling value.

[0011] Furthermore, the differential evolution algorithm improves search efficiency during the optimization process by employing a subpopulation merging mechanism, including: Individuals in the entire population are assigned to subpopulations with the same number of quadratic loops. Each subpopulation corresponds to one quadratic loop. The selection range of the difference basis vector in the mutation operation is limited to the subpopulation to which the individual belongs. At each preset interval of generations, the average value of the Euclidean distance between the individual parameter vectors in each subpopulation is calculated as a diversity index. When the diversity index of a subpopulation is lower than the preset diversity threshold, the subpopulations corresponding to the adjacent secondary circuits in the electrical topology diagram of the circuit corresponding to that subpopulation are retrieved, and the subpopulations with the highest diversity index that are not in a merged state are selected for merging.

[0012] Furthermore, improving iteration efficiency based on the subpopulation splitting mechanism includes: A minimum merging generation and a maximum merging iteration generation are preset. When the number of iterations of the merged subpopulation exceeds the minimum merging generation, and the improvement in the fitness value of the best individual within a consecutive preset generation is lower than a preset improvement threshold, and the change in the fitness value of the best individual in the subpopulation within a consecutive preset generation is less than a preset improvement threshold, or when the number of iterations of the merged subpopulation reaches the maximum merging iteration generation, the individuals in the merged subpopulation are sorted by fitness value and alternately assigned to two new subpopulations. Diversity detection and merging are not performed within the preset iteration period.

[0013] The beneficial effects of this invention are as follows: This invention first uses a timeliness assessment module to determine attenuation rate parameters for each monitoring dimension of each cable based on a differential evolution algorithm. It then calculates a timeliness attenuation coefficient by combining the duration of data generation with the standard update cycle, thus solving the problem of quantifiable timeliness caused by differences in collection frequency and change rate among different monitoring dimensions. Next, a state assessment module performs a weighted summation and average of the normalized degradation values ​​of each monitoring dimension with the timeliness attenuation coefficient to obtain a single-cable degradation score reflecting the true degree of cable degradation. Finally, a risk assessment module sorts the cables within the loop according to their single-cable degradation scores and selects the leading cables to form a degradation sequence. Based on the head concentration and head dispersion, it calculates the loop failure proximity, thus addressing the problem that weighted aggregation methods cannot distinguish between two risk modes. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the structure of the digital twin-based full-cycle visual management system for secondary cable maintenance according to the present invention. Figure 2 This is a schematic diagram of the dynamic merging and splitting mechanism of the subpopulation in this invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] like Figure 1 As shown, a digital twin-based full-cycle visualization management system for secondary cable maintenance includes: The data acquisition module is used to acquire status data for various monitoring dimensions of the secondary cable.

[0017] In this embodiment, the status data includes insulation resistance data, temperature data, partial discharge data, appearance status data, and circuit continuity status data. A decrease in insulation resistance data indicates potential aging or deterioration of the insulation material, or moisture corrosion. An increase in temperature data indicates potential increased contact resistance at the joints or excessive current carrying capacity of the cable. An increase in partial discharge data indicates potential impending insulation breakdown. Appearance status data includes the integrity of the cable sheath, color changes, mechanical damage, bending deformation, and joint corrosion, etc., which are quantified and recorded according to preset rating standards. For example, for sheath integrity, the absence of any visible damage on the sheath surface is defined as integrity level 1 and assigned a quantitative score of 0; minor scratches on the sheath surface with a length of 5 mm or less but not exposing the internal insulation layer are defined as integrity level 2 and assigned a quantitative score of 0.3. Circuit continuity status data reflects whether the electrical connectivity of the secondary circuit is normal.

[0018] The digital twin module is used to create an entity object for each secondary cable in the digital twin model. The entity object contains the cable's loop type and cable tray affiliation information.

[0019] In this embodiment, the entity object includes the cable's loop type, spatial path, loop affiliation, and cable tray affiliation information. The loop type includes protection loops, measurement loops, control loops, and signal loops, etc. The spatial path records the cable's laying path from the starting terminal block, through each cable tray segment, to the terminating terminal block in the form of a coordinate point sequence. The loop affiliation information includes the secondary loop number to which the cable belongs, and the cable tray affiliation information includes the numbers of each cable tray segment through which the cable passes.

[0020] The timeliness assessment module is used to determine the decay rate parameter based on the differential evolution algorithm when new data is received for any monitoring dimension. Based on the duration since the data was generated, the standard update cycle of the monitoring dimension, and the decay rate parameter, it calculates the timeliness decay coefficient of the corresponding data for the monitoring dimension and treats it as a data update event.

[0021] In this embodiment, the calculation of the timeliness decay coefficient of the data corresponding to the monitoring dimension includes: Obtain the time difference between the current system time and the time of the most recent data record generation. Divide the time difference by the standard update cycle of the corresponding monitoring dimension to obtain the time offset ratio. Calculate the timeliness attenuation coefficient based on the attenuation rate parameter and the time offset ratio. The larger the attenuation rate parameter or the larger the time offset ratio, the smaller the timeliness attenuation coefficient.

[0022] When any monitoring dimension receives newly acquired data, the system calculates the time-sensitivity attenuation coefficient. Taking the k-th monitoring dimension of a secondary loop cable as an example, when calculating the time-sensitivity attenuation coefficient, the system first retrieves the generation time of the most recent data record under the k-th monitoring dimension, then obtains the current system time, subtracts the generation time from the current system time, and obtains the time difference since the data was generated. A standard update cycle is set for each monitoring dimension to match its data acquisition method. The standard update cycle is the time interval between two adjacent acquisitions of this type of data under normal operation and maintenance conditions. For example, if a distributed fiber optic temperature measurement system completes one round of acquisition every 15 minutes, then the standard update cycle for the temperature dimension is 15 minutes. If the inspection procedure stipulates that the cable undergoes a visual inspection every 24 hours, then the standard update cycle for the appearance dimension is 24 hours. Dividing the above time difference by the standard update cycle of the monitoring dimension yields the time offset ratio.

[0023] In real-world scenarios, temperature data is affected by cable load, resulting in a rapid rate of change. In contrast, insulation resistance data reflects the slow aging process of insulation materials, exhibiting an extremely low rate of change within the acquisition period. To quantify this difference, an attenuation rate parameter is set for each monitoring dimension. The attenuation rate parameter ranges from 0 to 1; a higher parameter indicates faster attenuation of the data's timeliness within the acquisition period. Specific setting methods will be described later.

[0024] This embodiment calculates the time-related attenuation coefficient using the following method: Time-related attenuation coefficient = 1 - attenuation rate parameter × time offset ratio. According to this formula, for monitoring dimensions with a large attenuation rate parameter, the time-related attenuation coefficient is smaller at the end of the acquisition period; for monitoring dimensions with a small attenuation rate parameter, the time-related attenuation coefficient remains close to 1 at the end of the acquisition period. Furthermore, when the calculated time-related attenuation coefficient is less than 0 due to sensor malfunction, it is truncated to 0.

[0025] The status assessment module is used to normalize the status data of the monitoring dimensions to obtain normalized degradation values, and to weight and average the normalized degradation values ​​of each monitoring dimension with the corresponding time-related attenuation coefficient to obtain the degradation score of the secondary cable. By normalizing the status data of each monitoring dimension, a normalized degradation value is obtained. A normalized degradation value of 0 indicates that the monitoring indicator for that dimension is in good condition, while a normalized degradation value of 1 indicates that the monitoring indicator for that dimension is on the verge of degradation. In particular, the above normalization process not only maps the status data to the range of 0-1, but also aligns the status data during the mapping process. For example, a higher insulation resistance indicates better health. The orientation is reversed during normalization to ensure that the obtained normalized degradation value conforms to the meaning described above. The normalized degradation values ​​of each monitoring dimension of the cable are multiplied by their corresponding time-related attenuation coefficients, summed, and then divided by the sum of the time-related attenuation coefficients for all monitoring dimensions. This weighted average is used to obtain the degradation score for a single cable.

[0026] The risk assessment module is used to sort all secondary cables in a loop according to the single-cable degradation score, select some cables based on the single-cable degradation score to form a degradation sequence, and calculate the failure proximity of the loop based on the degradation sequence.

[0027] In this embodiment, the failure proximity of the circuit is calculated based on the deterioration sequence, including: The individual cable degradation scores of each cable within the same secondary circuit are sorted from high to low, and the top m scores are selected to form a degradation sequence, where m is the number of sensitive circuits determined based on the total number of cables in the circuit and the circuit type.

[0028] After completing the individual cable degradation scoring for all cables within the same secondary circuit, the individual cable degradation scores are sorted in descending order of value. After sorting, the top m cables with the highest individual cable degradation scores are selected from the sorted results. These m selected individual cable degradation scores form a degradation sequence, where m is a preset circuit sensitivity number. The value of the circuit sensitivity number m is determined based on the total number of cables in the circuit and the circuit type. For example, protection circuits, due to their high functional importance to the safe operation of the power grid, use a larger value for m. Measurement or signal circuits have relatively lower functional importance, and m can be set to a smaller value, such as 2 or 3. The purpose of selecting the top m cables instead of all cables is that the actual risk of the circuit is mainly determined by the cable with the most severe degradation.

[0029] The average of all degradation scores in the degradation sequence is calculated as the head concentration, and the difference between the maximum and minimum scores in the degradation sequence is calculated as the head dispersion. The concentration weight and dispersion weight are determined according to the loop type. The failure proximity is calculated based on the concentration weight, dispersion weight, head concentration, and head dispersion. The higher the head concentration and the lower the head dispersion, the greater the loop failure proximity.

[0030] After obtaining the degradation sequence, the average degradation score of all individual cables in the sequence is calculated, and this average is defined as the head concentration. The difference between the maximum and minimum scores in the degradation sequence is calculated, and this difference is defined as the head dispersion. The smaller the head dispersion, the closer the degradation scores of the cables are. Finally, the failure proximity is calculated based on the following formula: Failure Proximity = Concentration Weight × Head Concentration + Dispersion Weight × (1 - Head Dispersion). According to this formula, when the head concentration is high and the head dispersion is low, the failure proximity reaches a larger value, indicating that the circuit is in a state where multiple cables are simultaneously approaching failure. The concentration weight and dispersion weight are determined based on the circuit type. For example, in a protection circuit, synchronous degradation of multiple cables will cause the protection function to fail, so the dispersion weight is greater than the concentration weight.

[0031] The alarm module is used to generate a loop-level risk alarm when the proximity of a failure exceeds a preset alarm threshold, and to mark the corresponding loop and secondary cable.

[0032] An alarm threshold is set. When the failure proximity of a loop exceeds the alarm threshold, a loop-level risk alarm is generated. At this time, the digital twin system marks the cable contributing to the degradation sequence in the loop as the loop risk contributing cable. In this embodiment, the color hue of the cable 3D model is mapped to the single cable degradation score. The single cable degradation score transitions from green to yellow and then to red, from low to high. Taking a secondary loop as a unit, a loop-level risk labeling layer is superimposed along the spatial path of all cables in the digital twin 3D model. The display style of the labeling layer is determined based on the failure proximity. The failure proximity transitions from low to high, corresponding to the labeling layer changing from no display to dashed border, to solid border, and then to flashing highlight. The operation and maintenance visualization interface displays a loop-level operation and maintenance priority list sorted from high to low failure proximity. Each entry in the list includes the loop number, failure proximity value, loop risk contributing cable number, and corresponding single cable degradation score.

[0033] In the aforementioned condition assessment process, the calculation of the time-related attenuation coefficient depends on the attenuation rate parameters of each monitoring dimension for each cable. Therefore, it is necessary to determine the optimal attenuation rate parameter for each monitoring dimension of each cable. Based on this objective, this embodiment determines the optimal attenuation rate parameter using the following differential evolution algorithm.

[0034] Specifically, in this embodiment, the decay rate parameter is determined based on the differential evolution algorithm, including: Each individual in the Differential Evolutionary Algorithm (DEA) contains a parameter vector. Each dimension of the parameter vector corresponds to the attenuation rate parameter value of the cable in the monitoring dimension. The DEA evaluates the fitness value of an individual based on a preset fitness function. The fitness function is calculated based on the comprehensive optimization weight and the average prediction bias. The comprehensive optimization weight represents the priority of attention assigned to different cables by the DEA during the optimization process, and the average prediction bias represents the error between the data aging rate predicted by the DEA during the optimization process and the actual data change rate monitored. The DEA improves search efficiency by using a subpopulation merging mechanism and improves iteration efficiency by using a subpopulation splitting mechanism during the optimization process.

[0035] In the Differential Evolutionary Algorithm (DEA), the parameter vector of each individual contains the same dimension as the product of the number of cables and the number of monitoring dimensions. The d-th dimension of the parameter vector corresponds to the attenuation rate parameter value of a certain cable in a certain monitoring dimension. The DEA relies on historical data accumulated during system operation and maintenance. Whenever a new data acquisition is completed for a certain monitoring dimension of a cable, the most recently recorded normalized degradation value for that dimension before this acquisition is marked as the old value, and the normalized degradation value obtained in this acquisition is marked as the new value. The actual time interval between the two acquisitions is recorded, thus forming a historical data update event. In the initial stage of system operation, the attenuation rate parameters of each monitoring dimension use initial values ​​preset based on operational experience. When the cumulative historical data update events for each cable and each monitoring dimension reach the preset minimum sample size threshold, the system starts the DEA to optimize the attenuation rate parameters. Subsequently, as new data continues to be acquired, the algorithm is periodically re-executed to update the parameter values.

[0036] The fitness function of the differential evolution algorithm is as follows: Where F is the fitness value of the individual. The total number of cables, To monitor the total number of dimensions, Let i be the average prediction bias of the k-th monitoring dimension of cable i. The comprehensive optimization weights for cable i.

[0037] In this embodiment, the comprehensive optimization weight is calculated based on the failure proximity of the cable loop and the proximity gradient value of the cable. Specifically, the adjacent cables of the cable are determined based on the cable tray affiliation information of each cable in the digital twin model, and the proximity gradient value is calculated based on the failure proximity of the cable loop and the failure proximity of the adjacent cables.

[0038] Specifically, the comprehensive optimization weights are calculated based on the following formula: ,in, The failure proximity of the circuit to which cable i belongs. Let i be the proximity gradient value of cable i. This is the gradient weight adjustment coefficient. The proximity gradient value is determined based on the parameter calculation results of the previous round before each optimization round begins. The proximity gradient value is determined based on the physical adjacency relationship constructed from the cable tray affiliation information of each cable in the digital twin model. The specific calculation formula is as follows: ,in, Let i be the proximity gradient value of cable i. This is the set of physical neighbors of cable i, specifically the cables that are in the same cable tray segment as cable i. Let i be the adjacency weight between cable i and its physical neighbor cable j. The failure proximity of the circuit to which cable i belongs. Let be the failure proximity of the loop to which the physical neighbor cable j belongs. Based on the above formula, when the proximity gradient value is positive, it means that the cable is physically surrounded by high-risk cables.

[0039] The significance of the aforementioned comprehensive optimization weights lies in the fact that failure proximity allows the parameter dimensions corresponding to cables in high-risk loops to receive a larger contribution to the fitness value. This is because the accuracy of cable parameters in high-risk loops has a greater impact on operational decisions. Secondly, Cables located in high-risk physical neighborhoods are given additional optimization weights. In real-world operation and maintenance scenarios, multiple cables within the same cable tray segment share the same laying channel, ambient temperature, and heat dissipation conditions. When a degraded cable belonging to a high-risk loop exists within a cable tray segment, that segment may experience problems such as obstructed heat dissipation channels and moisture intrusion. Other cables within the same cable tray segment, even those belonging to lower-risk loops, may still be affected by the same environmental factors. If the attenuation rate parameters of these cables are not accurate enough, their current state assessment may be underestimated. Therefore, by applying additional optimization weights to these cables using proximity gradient values, their parameter accuracy is improved.

[0040] The average prediction bias is calculated using the following formula: ,in, Let be the total number of data update events that have occurred in the historical records for the kth monitoring dimension of cable i. Each time a new data collection is completed for this dimension, it is considered a data update event. The attenuation rate parameter of the k-th monitoring dimension for cable i encoded for the current individual. The time offset ratio is the ratio of the actual time interval between two adjacent data collections in the j-th data update event to the standard update cycle of that dimension. ,in, Let $\frac{k}{j}$ be the most recently recorded normalized degradation value of the $k$-th monitoring dimension of cable $i$ before the $j$-th data update event. Let be the normalized degradation value of cable i, which is actually collected in the j-th data update event for the k-th monitoring dimension. This represents the maximum reasonable change for the k-th monitoring dimension within a single standard update cycle. The maximum reasonable change is determined based on the physical characteristics of the monitored object and the equipment operating standards. For example, the maximum reasonable change for the temperature dimension is set based on the conductor's rated temperature rise standard.

[0041] The physical meaning of the above formula is, This represents the percentage of data that the system considers outdated. This represents the actual percentage deviation between the old and new data. Greater than A high fitness value indicates that the system overestimates the data's aging rate, meaning the attenuation rate parameter is too large, and the actual data changes are not as drastic as predicted. Conversely, a low fitness value indicates that the system underestimates the data's aging rate, and the actual data changes more drastically than predicted. The reason for using failure proximity as a weight is that the accuracy of cable parameters in high-risk loops has a greater impact on operational decisions; therefore, ensuring the accuracy of these cable parameters should be prioritized during optimization. Thus, a smaller fitness value means that the combination of attenuation rate parameters encoded by that individual cable more accurately matches the actual change patterns of data across all monitoring dimensions globally.

[0042] In this embodiment, a reference sample size threshold is set. In the mutation operation of the differential evolution algorithm, for the parameters in the parameter vector, if the cumulative number of data update events is less than the reference sample size threshold, the scaling factor starts with a preset maximum scaling value and decreases linearly as the number of data update events increases. When the number of data update events exceeds the reference sample size threshold, the scaling factor is set to a preset minimum scaling value.

[0043] In the mutation operation of the differential evolution algorithm, the scaling factor determines the magnitude of the perturbation of the difference vector on the individual. In the standard differential evolution algorithm, the scaling factor is a global constant, applying the same perturbation magnitude to all parameter dimensions. However, in this scenario, for parameter dimensions with a small number of historical data update events, the sample information used in the fitness function to evaluate the accuracy of the parameters in that dimension is insufficient. A larger scaling factor is needed to maintain a larger search range and avoid premature convergence to a local optimum due to insufficient information. For parameter dimensions with a sufficient number of historical data update events, a smaller scaling factor is used for fine-tuning to bring the parameters closer to their optimal values.

[0044] Based on the above analysis, the scaling factor for the d-th dimension parameter is set as follows: ,in, Let be the scaling factor for the d-th dimension parameter in the differential evolution algorithm. The preset maximum scaling value. This is the preset minimum scaling value. This represents the cumulative number of data update events in the historical record for the monitoring dimension corresponding to the cable of the d-th dimension parameter. This is a preset reference sample size threshold. Based on the above formula, when the number of historical data update events is much smaller than the reference sample size threshold, the scaling factor is close to the maximum scaling value, the mutation disturbance is large, and the search range is wide. When the number of historical data update events reaches or exceeds the reference sample size threshold, the scaling factor drops to the minimum scaling value, the mutation disturbance is small, and the search is concentrated in the neighborhood of the current solution. However, when the data volume exceeds the reference threshold, further reducing the scaling factor will lead to an overly limited search. The min function is used for truncation to ensure that a minimum search capability is maintained during the adjustment phase.

[0045] In this embodiment, during the cross-operation of the differential evolution algorithm, the standard differential evolution algorithm independently performs cross-determination on each dimension of the individual parameter vector. However, in the scenario of optimizing the attenuation rate parameter of secondary cables, an increase in joint contact resistance simultaneously leads to accelerated temperature rise and accelerated insulation aging, while insulation material degradation simultaneously causes accelerated decrease in insulation resistance and accelerated increase in partial discharge. This coupling relationship means that for cables under a specific degradation mode, there is an inherent correlation between the attenuation rate parameters of each monitoring dimension and the degradation mode, forming a parameter combination with physical interpretability.

[0046] In the mutation operation of differential evolution, the mutation vector is generated by adjusting the difference of the differential basis vectors by a scaling factor. The perturbations of each monitoring dimension of the same cable come from the same pair of differential basis vectors, preserving the correlation structure between dimensions. If a dimension-by-dimensional independent determination method is adopted in the subsequent crossover operation, some monitoring dimensions of the same cable may adopt the new values ​​of the mutated individual while the remaining monitoring dimensions retain the original values ​​of the target individual. This splices together parameter fragments from different individuals, producing parameter combinations without physical meaning, destroying the correlation structure between dimensions established in the mutation operation, and reducing the effectiveness of the experimental individuals.

[0047] To address the aforementioned issues, this embodiment clusters the parameter vectors of individuals on a cable-by-cable basis. Parameter dimensions corresponding to all monitoring dimensions of the same cable are grouped into the same cluster, thus dividing the individual parameter vectors into multiple clusters equal to the number of cables. During crossover operations, a random number is independently generated for each cluster and compared with a preset crossover probability. When the random number is less than the crossover probability, all parameter dimensions within that cluster adopt the values ​​of the corresponding dimensions from the mutated individual. When the random number is greater than or equal to the crossover probability, all parameter dimensions within that cluster retain the original values ​​of the corresponding dimensions from the target individual. Through this cluster-level crossover control, the attenuation rate parameters of all monitoring dimensions for the same cable are always synchronously updated or retained in each crossover operation, ensuring the integrity of the multi-dimensional parameter combinations for each cable in the generated test individual, corresponding to a physically interpretable degradation state mode.

[0048] In this embodiment, the differential evolution algorithm improves search efficiency during the optimization process by using a subpopulation merging mechanism, including: Individuals in the entire population are assigned to subpopulations with the same number of quadratic loops. Each subpopulation corresponds to one quadratic loop. The selection range of the difference basis vectors in the mutation operation is limited to the subpopulation to which the individual belongs.

[0049] In the differential evolution algorithm, each individual is a complete parameter vector encoding the attenuation rate parameters of all monitored dimensions of all cables globally. When the population size is fixed and the search space dimension is high, the distribution of individuals in the population is relatively sparse. Randomly selecting differential basis vectors across the entire population will lead to low search efficiency.

[0050] like Figure 2 As shown, to improve search efficiency, this embodiment divides the population into multiple subpopulations to narrow the selection range of the difference basis vectors. Specifically, individuals in the population are evenly distributed into several subpopulations with the same number of quadratic loops, and each subpopulation corresponds to one quadratic loop.

[0051] At each preset interval of generations, the average value of the Euclidean distance between the individual parameter vectors in each subpopulation is calculated as a diversity index. When the diversity index of a subpopulation is lower than the preset diversity threshold, the subpopulations corresponding to the adjacent secondary circuits in the electrical topology diagram of the circuit corresponding to that subpopulation are retrieved, and the subpopulations with the highest diversity index that are not in a merged state are selected for merging.

[0052] In the mutation operation, for a target individual belonging to a certain subpopulation, the selection range of the difference basis vector is limited to within that subpopulation. Since the comprehensive optimization weight of each cable in the fitness function is positively correlated with the failure proximity of its corresponding loop, individuals within the same subpopulation will preferentially converge on the parameter dimension corresponding to the cable in their own loop under selection pressure. As the iteration progresses, individuals within the subpopulation gradually converge to a local region in the search space. As a result, the amplitude of the constructed difference vector decreases, and the experimental individuals generated by the mutation operation are more concentrated in the reachable neighborhood of the target individual, making the search process more efficient. Furthermore, each subpopulation evolves independently within its own region, maintaining the diversity of the global population distribution in the search space and preventing the entire population from prematurely converging to a single solution.

[0053] In this embodiment, the iteration efficiency is improved based on the subpopulation splitting mechanism, including: A minimum merging generation and a maximum merging iteration generation are preset. When the number of iterations of the merged subpopulation exceeds the minimum merging generation, and the improvement in the fitness value of the best individual within a consecutive preset generation is lower than a preset improvement threshold, and the change in the fitness value of the best individual in the subpopulation within a consecutive preset generation is less than a preset improvement threshold, or when the number of iterations of the merged subpopulation reaches the maximum merging iteration generation, the individuals in the merged subpopulation are sorted by fitness value and alternately assigned to two new subpopulations. Diversity detection and merging are not performed within the preset iteration period.

[0054] However, limiting the selection of differential basis vectors to within subpopulations leads to convergence of individuals within subpopulations under selection pressure as iterations progress, resulting in decreased search efficiency. To address this issue, this embodiment calculates the average Euclidean distance between all pairs of individuals within each subpopulation at preset intervals, using this average distance as a diversity index. A higher diversity index indicates significant differences among individuals within a subpopulation, allowing for the construction of effective differential vectors. When the diversity index of a subpopulation falls below a preset diversity threshold, the search for that subpopulation is considered stalled. At this point, the subpopulations corresponding to the circuits of that subpopulation that share a terminal block or are directly connected via a terminal block in the secondary circuit electrical topology diagram are retrieved. The adjacent subpopulation with the highest diversity index is selected as the merging target, and the two subpopulations are merged into an expanded subpopulation. If the diversity indices of all adjacent subpopulations are below the diversity threshold, the adjacent subpopulation with the highest diversity index is selected for merging.

[0055] The reason for merging circuits with adjacent electrical topologies is that circuits sharing terminal blocks have cables that partially overlap or are close in their physical laying paths, passing through some of the same cable tray segments. These cables share similar optimal attenuation rate parameters in environmentally sensitive dimensions such as temperature. The parameter values ​​of individuals from adjacent circuit subpopulations in these dimensions can provide meaningful search direction references for currently stagnant subpopulations. In the merged subpopulation, individuals come from two subpopulations that have converged to different parameter regions, increasing the parameter differences between individuals and allowing the construction of significantly different difference vectors. Subpopulations already in a merged state do not participate in new merge operations, avoiding excessively large subpopulation sizes due to continuous merging of multiple subpopulations, which would reduce search efficiency.

[0056] To ensure sufficient information exchange after merging, the merged subpopulation must iterate for at least the preset minimum number of generations before a split determination can be made. Information exchange is considered complete when the merged subpopulation iterates for more than the minimum number of generations and the improvement in fitness value of the best individual within the subpopulation within the preset number of generations is less than a preset improvement threshold. If the merged subpopulation iterates for the maximum number of generations without meeting the above split conditions, a split operation is forcibly executed to prevent the merged state from remaining in place for too long. During the split operation, all individuals in the merged subpopulation are sorted by fitness value and alternately assigned to the two new subpopulations, resulting in an approximately balanced fitness distribution. The two new subpopulations then revert to independent subpopulations to continue the search. The split subpopulations enter a cooling-off period during which diversity detection and merging determination are not performed. The number of generations in the cooling-off period is equal to the minimum number of generations to prevent immediate re-triggering of merging due to insufficient diversity after splitting, thus avoiding frequent alternation between merging and splitting operations.

[0057] Through the aforementioned dynamic merging and splitting mechanism, each subpopulation can perform efficient searching with a small selection range during the normal iteration phase. When the search stalls, the candidate pool is expanded by merging to introduce new differential vector directions. After information exchange is completed, the subpopulation splits again to restore the efficient search mode.

[0058] It should be noted that the threshold parameters and related parameters of the differential evolution algorithm involved in this invention can be reasonably determined and adjusted by those skilled in the art based on the actual application scenario, system scale, number of cables, number of monitoring dimensions, and operation and maintenance experience, through conventional technical means such as experimental testing, simulation verification, or historical data statistical analysis, in order to achieve the optimal risk assessment effect and algorithm convergence performance.

[0059] The normalization method for state data of each monitoring dimension in this invention can employ conventional normalization techniques in the art, including but not limited to min-max normalization, Z-score standardization, or piecewise mapping methods based on expert knowledge. The specific method chosen depends on the distribution characteristics and physical meaning of the data for each monitoring dimension. The goal of normalization is to uniformly map the original state data with different dimensions and numerical ranges to the [0, 1] interval, and to perform directional alignment during the mapping process, so that a larger normalized value represents a more severe degree of degradation. Those skilled in the art can determine the normalization mapping relationship based on prior knowledge such as equipment standards, operating specifications, and failure thresholds for each monitoring dimension.

[0060] Furthermore, for the sake of concise presentation of the formulas and clear expression of the technical solutions, the division-to-zero protection measures for each formula are not described in detail in the specification. However, those skilled in the art will understand that in actual programming implementation, it is necessary to add judgment and fault tolerance processing for all possible situations where the division by zero may occur. This is a common practice in software engineering and does not affect the integrity and feasibility of the technical solution of this invention.

Claims

1. A fully visualized management system for the operation and maintenance of secondary cables based on digital twins, characterized in that, include: The data acquisition module is used to acquire status data of various monitoring dimensions of the secondary cable; The digital twin module is used to create an entity object for each secondary cable in the digital twin model. The entity object contains the cable's loop type and cable tray ownership information. The timeliness assessment module is used to determine the decay rate parameter based on the differential evolution algorithm when any monitoring dimension receives new collected data. Based on the duration since the data was generated, the standard update cycle of the monitoring dimension, and the decay rate parameter, it calculates the timeliness decay coefficient of the corresponding data of the monitoring dimension and treats it as a data update event. The status assessment module is used to normalize the status data of the monitoring dimensions to obtain normalized degradation values, and to weight and average the normalized degradation values ​​of each monitoring dimension with the corresponding time-related attenuation coefficient to obtain the degradation score of the secondary cable. The risk assessment module is used to sort all secondary cables in the loop according to the single cable degradation score, select some cables based on the single cable degradation score to form a degradation sequence, and calculate the failure proximity of the loop based on the degradation sequence. The alarm module is used to generate a loop-level risk alarm when the proximity of failure exceeds a preset alarm threshold, and to mark the corresponding loop and secondary cable. Calculate the timeliness decay coefficient of the data corresponding to the monitoring dimensions, including: Obtain the time difference between the current system time and the time of the most recent data record generation. Divide the time difference by the standard update cycle of the corresponding monitoring dimension to obtain the time offset ratio. Calculate the timeliness attenuation coefficient based on the attenuation rate parameter and the time offset ratio. The larger the attenuation rate parameter or the larger the time offset ratio, the smaller the timeliness attenuation coefficient. The failure proximity of a circuit is calculated based on the degradation sequence, including: The individual cable degradation scores of each cable in the same secondary circuit are sorted from high to low, and the top m scores are selected to form a degradation sequence, where m is the number of sensitive circuits determined based on the total number of cables in the circuit and the circuit type. The average of all degradation scores in the degradation sequence is calculated as the head concentration, and the difference between the maximum and minimum scores in the degradation sequence is calculated as the head dispersion. The concentration weight and dispersion weight are determined according to the loop type. The failure proximity is calculated based on the concentration weight, dispersion weight, head concentration, and head dispersion. The higher the head concentration and the lower the head dispersion, the greater the loop failure proximity.

2. The system according to claim 1, characterized in that, The decay rate parameters are determined based on the differential evolution algorithm, including: Each individual in the Differential Evolutionary Algorithm (DEA) contains a parameter vector. Each dimension of the parameter vector corresponds to the attenuation rate parameter value of the cable in the monitoring dimension. The DEA evaluates the fitness value of an individual based on a preset fitness function. The fitness function is calculated based on the comprehensive optimization weight and the average prediction bias. The comprehensive optimization weight represents the priority of attention assigned to different cables by the DEA during the optimization process, and the average prediction bias represents the error between the data aging rate predicted by the DEA during the optimization process and the actual data change rate monitored. The DEA improves search efficiency by using a subpopulation merging mechanism and improves iteration efficiency by using a subpopulation splitting mechanism during the optimization process.

3. The system according to claim 2, characterized in that, The comprehensive optimization weight is calculated based on the failure proximity of the cable loop and the proximity gradient value of the cable. Specifically, the adjacent cables are determined based on the cable tray affiliation information of each cable in the digital twin model, and the proximity gradient value is calculated based on the failure proximity of the cable loop and the failure proximity of the adjacent cables.

4. The system according to claim 2, characterized in that, A reference sample size threshold is set. In the mutation operation of the differential evolution algorithm, for the parameters in the parameter vector, if the cumulative number of data update events is less than the reference sample size threshold, the scaling factor starts with the preset maximum scaling value and decreases linearly as the number of data update events increases. When the number of data update events exceeds the reference sample size threshold, the scaling factor is set to the preset minimum scaling value.

5. The system according to claim 2, characterized in that, Differential evolutionary algorithms improve search efficiency during optimization by employing a subpopulation merging mechanism, including: Individuals in the entire population are assigned to subpopulations with the same number of quadratic loops. Each subpopulation corresponds to one quadratic loop. The selection range of the difference basis vector in the mutation operation is limited to the subpopulation to which the individual belongs. At each preset interval of generations, the average value of the Euclidean distance between the individual parameter vectors within each subpopulation is calculated as a diversity index. When the diversity index of a subpopulation is lower than the preset diversity threshold, the subpopulations corresponding to the adjacent secondary circuits in the electrical topology diagram of the circuit corresponding to that subpopulation are retrieved, and the subpopulations with the highest diversity index that are not in a merged state are selected for merging.

6. The system according to claim 2, characterized in that, Improving iteration efficiency based on subpopulation splitting mechanisms includes: A minimum merging generation and a maximum merging iteration generation are preset. When the number of iterations of the merged subpopulation exceeds the minimum merging generation, and the improvement in the fitness value of the best individual within a consecutive preset generation is lower than a preset improvement threshold, and the change in the fitness value of the best individual in the subpopulation within a consecutive preset generation is less than a preset improvement threshold, or when the number of iterations of the merged subpopulation reaches the maximum merging iteration generation, the individuals in the merged subpopulation are sorted by fitness value and alternately assigned to two new subpopulations. Diversity detection and merging are not performed within the preset iteration period.

Images