An automated method and system for filling a cable conduit with a thermally conductive material

Abstract

This invention discloses an automated filling method and system for thermally conductive materials inside cable ducts, relating to the field of plastic processing technology. The method includes: collecting operating status parameters of the cable laid in the high-voltage duct; using unsupervised multi-point iterative search to determine multiple typical operating conditions and cable heat generation and transfer models; matching a high thermal conductivity carbon mesh material combination for simulation; constructing a thermally conductive material filling strategy library based on the simulation results; retrieving the optimal filling strategy in real time; initiating automated injection operations; identifying synchronous observations; and calibrating the filling strategy in real time. This invention solves the technical problems of existing technologies that cannot flexibly adjust the thermally conductive material filling strategy according to the real-time operating status of the cable duct, resulting in low filling efficiency and poor accuracy. It achieves the technical effect of flexibly matching the material filling strategy based on the real-time operating status of the cable duct and performing automated filling, thereby improving filling accuracy and efficiency.

Description

Technical Field

[0001] This invention relates to the field of plastic processing technology, and specifically to an automated filling method and system for filling cable conduits with thermally conductive materials. Background Technology

[0002] With the widespread application of high-voltage cables, power systems have placed higher demands on the heat dissipation performance of these cables. Filling with thermally conductive materials can effectively enhance the heat dissipation capacity of cables, reduce their temperature, and extend their service life. However, current technologies mostly rely on manual injection molding, which cannot accurately select suitable thermally conductive materials based on the actual operating conditions of the cable. This not only results in low efficiency and poor precision but may also lead to material waste or uneven filling, affecting the cable's heat dissipation performance and current carrying capacity. Summary of the Invention

[0003] This application provides an automated filling method and system for filling cable ducts with thermally conductive materials, which solves the technical problems of existing technologies that cannot flexibly adjust the filling strategy of thermally conductive materials according to the real-time operating status of cable ducts, resulting in low filling efficiency and poor accuracy.

[0004] The first aspect of this application provides an automated filling method for filling cable ducts with thermally conductive materials. The method includes: collecting a set of operating state parameters for high-voltage cable duct installations using load conditions, laying environment, and duct structure dimensions as indexes; performing unsupervised multi-point iterative search based on the set of operating state parameters to determine multiple typical operating conditions and multiple typical cable heat generation and transfer models; matching and screening the porosity, thermal conductivity, mechanical properties, and filling slurry ratio parameters of a high thermal conductivity carbon mesh material according to the multiple typical operating conditions to determine multiple matching thermally conductive material combinations, wherein each typical operating condition corresponds to one matching thermally conductive material combination; and performing matching and screening based on the multiple typical cable heat generation and transfer models and multiple matching thermally conductive material combinations. A full-scale pipe heat generation and heat transfer simulation was conducted to construct multiple thermally conductive material filling strategies. These strategies were then mapped and associated with various typical operating conditions to build an automated filling strategy library. Each thermally conductive material filling strategy included filling injection flow rate, injection pressure, filling sequence, and allowable solidification time window. Real-time operating status characteristic parameters were collected, and the automated filling strategy library was searched. A real-time automated filling strategy was loaded, automated injection operation was initiated, and synchronous observations were performed to obtain a synchronous observation sequence. An observation offset detector and a control action uncertainty detector were used to identify the synchronous observation sequence, and the real-time automated filling strategy was calibrated based on the identification results.

[0005] A second aspect of this application provides an automated filling system for filling cable ducts with thermally conductive materials. The system includes: an operating status parameter acquisition module, used to acquire a set of operating status parameters for high-voltage cable duct laying, indexed by load conditions, laying environment, and duct structure dimensions; a typical operating condition determination module, used to perform unsupervised multi-point iterative search based on the operating status parameter set to determine multiple typical operating conditions and multiple typical cable heat generation and heat transfer models; a thermally conductive material combination matching module, used to match and screen the porosity, thermal conductivity, mechanical properties, and filling slurry ratio parameters of high thermal conductivity carbon mesh materials according to the multiple typical operating conditions to determine multiple matching thermally conductive material combinations, wherein each typical operating condition corresponds to one matching thermally conductive material combination; and a filling strategy library construction module, used to construct a filling strategy library based on the multiple typical cable heat generation models. A full-scale pipe heat generation and heat transfer simulation was performed using a heat transfer model and multiple matching thermally conductive materials. Multiple thermally conductive material filling strategies were constructed, and these strategies were mapped and associated with multiple typical operating conditions to build an automated filling strategy library. Each thermally conductive material filling strategy includes filling injection flow rate, injection pressure, filling sequence, and allowable solidification time window. An automated filling observation module was used to collect real-time operating status characteristic parameters, search the automated filling strategy library, load real-time automated filling strategies, initiate automated injection operations, and perform synchronous observations to obtain a synchronous observation sequence. A strategy calibration module was used to identify the synchronous observation sequence using an observation offset detector and a control action uncertainty detector, and calibrate the real-time automated filling strategy based on the identification results.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] This application provides an automated filling method and system for filling thermally conductive materials into cable ducts, relating to the field of plastic processing technology. By collecting cable operating status parameters, using unsupervised multi-point iterative search to determine typical operating conditions and cable heat generation and transfer models, and selecting matching high thermal conductivity carbon mesh material combinations for simulation, an automated filling strategy library is constructed. The optimal filling strategy is retrieved in real time, and the filling process is calibrated through automated injection and synchronous observation. This solves the technical problems of existing technologies that cannot flexibly adjust the thermally conductive material filling strategy according to the real-time operating status of the cable duct, resulting in low filling efficiency and poor accuracy. It achieves the technical effect of flexibly matching material filling strategies based on the real-time operating status of the cable duct and performing automated filling, thereby improving filling accuracy and efficiency. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic diagram of an automated filling method for filling thermally conductive material into a cable duct, provided in an embodiment of this application;

[0010] Figure 2 This is a schematic diagram of an automated filling system for filling thermally conductive material into a cable duct, as provided in an embodiment of this application.

[0011] Figure labeling: 11 Operation status parameter acquisition module, 12 Typical working condition determination module, 13 Thermal conductive material combination matching module, 14 Filling strategy library construction module, 15 Automated filling observation module, 16 Strategy calibration module. Detailed Implementation

[0012] This application provides an automated filling method and system for filling cable ducts with thermally conductive materials, which solves the technical problems of existing technologies that cannot flexibly adjust the filling strategy of thermally conductive materials according to the real-time operating status of cable ducts, resulting in low filling efficiency and poor accuracy.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] It should be noted that the terms "first," "second," etc., in the specification 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 of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server 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 modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0015] Example 1, as Figure 1As shown, this application provides an automated filling method for filling cable conduits with thermally conductive material, the method comprising:

[0016] P10: Collect a set of operating status parameters for high-voltage duct-laid cables, indexed by load conditions, laying environment, and duct structure dimensions.

[0017] Specifically, to achieve comprehensive monitoring and data collection of the operating status of high-voltage cable ducts, it is first necessary to define the key index parameters in the data collection process. These parameters include load conditions, laying environment, and duct structure dimensions. These index parameters can comprehensively reflect the changes in the cable's status during actual operation, providing a solid data foundation for subsequent model building and filling strategy formulation.

[0018] Load conditions primarily involve the magnitude of the current and the voltage level during cable operation. The current directly affects the cable's heat generation, while the voltage is closely related to the cable's power transmission efficiency. During data acquisition, high-precision current transformers and voltage sensors are installed at key cable nodes to monitor current and voltage changes in real time. These sensors record data at millisecond-level sampling frequencies and use built-in filtering algorithms to remove noise, ensuring data accuracy and reliability. The acquired current and voltage data will be labeled as time-series data for subsequent analysis of cable operating status changes under different load conditions.

[0019] Secondly, the laying environment has a significant impact on the heat dissipation performance of cables. The laying environment includes external factors such as temperature, humidity, and wind speed, all of which directly affect the cable's heat dissipation capacity. In the case of duct laying, changes in airflow and temperature field within the duct will affect the cable's heat dissipation effect; therefore, collecting relevant environmental data is crucial. Environmental monitoring equipment can be deployed to record ambient air temperature and humidity data in real time, while anemometers can be installed to monitor wind speed and airflow velocity inside and outside the duct. Furthermore, air convection rate is also an important factor, and relevant data can be collected using specialized airflow monitoring equipment. These environmental parameters help to understand the heat exchange between the cable and the external environment, further guiding the selection and optimization of thermally conductive materials.

[0020] Next, the structural dimensions of the cable conduit are another key factor affecting the cable's heat dissipation performance. The inner diameter, outer diameter, length, and material of the conduit all influence the cable's heat dissipation efficiency. The conduit's dimensions determine the size of the gap between the cable and the conduit, which directly affects the effectiveness of the filler material and the heat conduction path. To accurately obtain the conduit's dimensions, laser measuring equipment can be used to precisely measure its various geometric parameters. This measurement data will be imported into computer-aided design (CAD) software to generate a three-dimensional model of the cable conduit. This model allows for a visual analysis of the spatial relationship between the cable and the conduit, providing precise geometric parameters for subsequent filler material design.

[0021] By comprehensively collecting data on load conditions, laying environment, and duct structure dimensions, a complete set of operating status parameters can be formed. This set includes multi-dimensional data such as current load, temperature, humidity, wind speed, airflow rate, duct geometry, and material, which comprehensively reflect the thermal characteristics of the cable under different operating conditions. All collected data should be transmitted in real time to a data center for aggregation and storage via Internet of Things (IoT) technology or Wireless Sensor Network (WSN). To ensure data accuracy and timeliness, all sensors should be high-precision and capable of regular calibration to avoid data deviation. This data not only provides a theoretical basis for optimizing the cable's heat dissipation performance but also provides a reliable basis for the selection of thermally conductive materials and the optimization of filling strategies.

[0022] P20: Based on the set of operating state parameters, perform unsupervised multi-point iterative search to determine multiple typical operating conditions and multiple typical cable heat generation and heat transfer models.

[0023] Furthermore, step P20 in this embodiment of the application also includes:

[0024] P21: Map each operating state parameter in the set of operating state parameters to a D-dimensional space to construct an operating state parameter space, where D is the dimension of the operating state parameter and is a positive integer; P22: Select multiple initial search points from the operating state parameter space through random sampling; P23: Perform iterative joint verification on the multiple initial search points. When the verification is successful, traverse the multiple initial search points in the operating state parameter space to perform unsupervised multi-point iterative search to determine multiple typical operating conditions; P24: Construct multiple cable heat generation models and multiple cable heat transfer models based on the multiple typical operating conditions; P25: Map and couple the multiple cable heat generation models and multiple cable heat transfer models to obtain the multiple typical cable heat generation and heat transfer models.

[0025] It should be understood that in order to extract representative typical operating conditions from the set of operating state parameters of high-voltage cable ducts and to construct corresponding cable heat generation and heat transfer models, unsupervised multi-point iterative search can be used to ensure the accuracy and reliability of extracting key information from complex data.

[0026] First, based on the set of operating state parameters, each operating state parameter is mapped to a D-dimensional space, constructing an operating state parameter space. Here, D represents the dimension of the operating state parameter and is a positive integer. For example, operating state parameters may include multiple dimensions such as current, voltage, ambient temperature, humidity, and pipe spacing. In this way, complex multi-parameter data is transformed into a set of points in a high-dimensional space, forming a high-dimensional operating state parameter space. In this space, each operating state parameter has a corresponding coordinate value, and all these parameters are jointly expressed through the constructed D-dimensional space, thus forming a complete, high-dimensional state description.

[0027] Next, to find possible typical operating conditions from this operating state parameter space, multiple initial search points can be selected through random sampling. Each initial search point represents a possible operating condition, and random sampling ensures that the search process is unbiased and can broadly cover the possibilities of various operating conditions. These initial search points do not require pre-defined specific rules or assumptions; instead, they are selected through uniform sampling of the operating state parameter space, ensuring sufficient representativeness of operating conditions from the entire space. In specific implementations, uniform random sampling or Latin hypercube-based sampling methods can be used to ensure a uniform distribution of samples in the high-dimensional space.

[0028] Subsequently, these initial search points undergo iterative joint verification. Joint verification is a method for comprehensively evaluating whether multiple parameter points meet specific conditions. It can effectively aggregate and optimize various search points through clustering, similarity metrics, or neighborhood-based verification methods to find the optimal typical operating conditions. In this step, the purpose of verification is to determine whether these initial search points can represent typical operating states. First, the verification process includes multi-dimensional evaluation of each initial search point, such as checking whether it conforms to known physical laws, such as the relationship between current and voltage, and whether it is within a reasonable range of environmental parameters, such as the normal range of temperature and humidity. If an initial search point fails verification, it is marked as an invalid point, and a new point is randomly selected from the operating state parameter space to supplement it.

[0029] Once all initial search points have been validated, an unsupervised multi-point iterative search is further performed within the runtime parameter space. Unsupervised multi-point iterative search is a data mining technique that refines and optimizes search results through multiple iterations. Specifically, clustering-based algorithms, such as K-Means or DBSCAN, can be used to cluster the initial search points, thereby identifying different typical operating conditions. In each iteration, the algorithm adjusts the positions of the search points based on the current clustering results and re-clusters until a preset convergence condition is met, such as the change in cluster centers being less than a certain threshold.

[0030] Next, based on several determined typical operating conditions, multiple cable heat generation models and multiple cable heat transfer models were further constructed. The cable heat generation model mainly describes the heat generation mechanism of the cable under different load conditions, while the cable heat transfer model focuses on the heat transfer process in the cable and its surrounding environment. The construction of these two models is based on operating state parameters under typical conditions, achieved through a combination of theoretical analysis and experimental verification. Specifically, the heat generation model can be described by establishing the cable's heat balance equation, considering the Joule heating effect of the current and the thermal resistance characteristics of the cable material. The heat transfer model can be constructed through finite element analysis (FEA) to simulate the conduction, convection, and radiation processes of heat in the cable, filling material, and surrounding environment.

[0031] In the construction of these models, multiple typical operating conditions were applied to different cable heating and heat conduction scenarios to ensure the adaptability and accuracy of the models in actual operation. The heating behavior and heat transfer methods of cables may vary significantly under different operating conditions. Therefore, modeling typical operating conditions allows for better prediction of key parameters such as cable heat loss and temperature changes. The cable heat generation model considers the influence of factors such as load, cable material, size, and ambient temperature on heat generation, while the cable heat transfer model combines the cable's thermal conductivity with the external environment to simulate the heat diffusion path, ensuring comprehensive control over cable thermal management.

[0032] Finally, multiple cable heat generation models and multiple cable heat transfer models are mapped and coupled to obtain several typical cable heat generation and heat transfer models. Mapping and coupling is a technique for associating and integrating different models. In this step, through mapping and coupling, the heat generation and heat transfer models are organically combined to form a complete cable thermal characteristic model. In specific implementation, a multiphysics coupling analysis method can be used, where the output of the cable heat generation model, such as the heat generation rate, is used as the input of the cable heat transfer model, while the temperature distribution of the cable heat transfer model is fed back into the cable heat generation model, thereby achieving dynamic interaction and coupling between the two models. In this way, typical cable heat generation and heat transfer models can comprehensively reflect the thermal behavior of cables under different operating conditions, providing solid theoretical support for the subsequent formulation of thermal conductivity material matching and filling strategies.

[0033] Furthermore, in this embodiment, step P23 further includes iteratively verifying the multiple initial search points:

[0034] P23-1: Enumerate each of the multiple initial search points pairwise to obtain an initial search point enumeration combination set; P23-2: Perform similarity identification on each of the initial search point enumeration combination sets to obtain an enumeration combination similarity set; P23-3: Count the number of enumeration combination similarities in the enumeration combination similarity set that are greater than or equal to a preset similarity threshold, and determine whether the statistical result is greater than or equal to the preset threshold. If yes, the verification fails; P23-4: If no, the verification passes.

[0035] Optionally, the process of iteratively joint verification of multiple initial search points is further refined. Through systematic similarity analysis, the diversity and independence of the initial search points are ensured, and model bias or local optima problems caused by excessive similarity of search points are avoided.

[0036] First, multiple initial search points are enumerated pairwise to obtain a set of possible combinations. This process generates a set containing all possible combinations by systematically comparing each pair of initial search points. For example, if there are n initial search points, the generated set of combinations will contain n(n-1) / 2 combinations. Each combination represents a comparison between a pair of initial search points.

[0037] Next, similarity identification is performed on each enumerated set of initial search point combinations to obtain an enumerated combination similarity set. Similarity identification is achieved by calculating the similarity between each pair of initial search points using various methods, such as Euclidean distance, cosine similarity, or other similarity metrics suitable for high-dimensional data. This quantifies the similarity value of two initial search point combinations, thereby determining their spatial relationship. All these similarity values ​​are then aggregated to form the enumerated combination similarity set.

[0038] Subsequently, the number of combinations with similarities greater than or equal to a preset similarity threshold is counted. This preset similarity threshold is a key parameter, set based on the system's design requirements and the tolerance level in practical applications. Its purpose is to filter out highly similar combinations, as these combinations may not provide new information for subsequent optimization searches. If the similarity between two search points is greater than or equal to the preset similarity threshold, they are considered too close in the runtime parameter space and may not provide sufficient diversity. The statistical results will indicate how many pairs of search points exceed this similarity threshold.

[0039] Finally, the validation is determined based on the statistical results. If the number of similarity points greater than or equal to a preset similarity threshold is greater than or equal to a preset number threshold, the initial search point set is considered to have failed validation. This means there are too many overly similar search points, which may affect the effectiveness of subsequent iterative searches. In this case, the initial search points need to be reselected to increase their diversity. Conversely, if the number of similarity points greater than or equal to a preset similarity threshold is less than a preset number threshold, the initial search point set is considered to have passed validation, and subsequent unsupervised multi-point iterative searches can continue.

[0040] This process ensures that the selection of initial search points is representative and avoids redundant and repeated search points from affecting the efficiency of the optimization process. It also ensures that unsupervised multi-point iterative search can cover a wide range of working conditions, thereby improving the accuracy and practicality of the final optimization results.

[0041] Furthermore, upon successful verification, an unsupervised multi-point iterative search is performed within the operating state parameter space, traversing multiple initial search points to determine multiple typical operating conditions. Steps P23-4 in this embodiment further include:

[0042] P23-41: Traverse each of the multiple initial search points and perform an iterative search process within the running state parameter space; wherein, the iterative search process includes:

[0043] P23-42: Calculate the local state density and determine the search update direction based on the distribution of operating state parameters around the initial search point; P23-43: Update the position of the initial search point along the direction of increasing state density, so that the initial search point gradually moves towards the high-density area of ​​operating state parameters to obtain iterative search points; P23-44: Repeat the search update movement iteration process for iterative search points until the position change of the search point is less than a preset threshold or the maximum number of iterations is reached to obtain multiple target search points; P23-42: Use the operating state parameters corresponding to the multiple target search points as multiple typical operating conditions.

[0044] Specifically, once the initial search point is validated, an unsupervised multi-point iterative search begins within the runtime state parameter space. The core of this process is to progressively update the position of the search point, moving it towards a high-density region of runtime state parameters, ultimately determining several typical operating conditions.

[0045] First, an iterative search process is performed on each of the multiple initial search points within the runtime state parameter space. Each initial search point represents a possible operating condition, therefore, a search needs to be performed on each of these initial search points to discover the most representative typical operating condition. Each initial search point will undergo an independent iterative search process within the runtime state parameter space. At this point, the core of the search process is to explore the runtime state parameter space surrounding the initial search point, searching for high-density regions similar to that point, thereby finding more relevant typical operating conditions.

[0046] During the iterative search, the local state density is first calculated based on the distribution of operating state parameters around the initial search point. Local state density reflects the distribution and clustering of operating state parameters within a specific region. Higher density indicates that the operating state parameters are more concentrated within that region, suggesting that the region may be a more stable and typical operating condition area. After calculating the local density, the system determines the update direction of the search based on the density distribution, moving towards areas with higher density. This step is crucial because it ensures that the search point is concentrated in potentially typical operating condition areas, thereby improving search efficiency.

[0047] Next, the initial search point is updated along the direction of increasing density. This update process is not random, but guided by the density direction, causing the search point to gradually move towards high-density regions in the operating state parameter space. As the search point approaches the high-density region, it can more accurately capture the characteristics of typical operating conditions. The step size of each position update can be dynamically adjusted according to the rate of change of local state density to ensure the stability and efficiency of the search process. The updated search points are called iterative search points, and these points represent new possible operating conditions found in the current iteration.

[0048] Then, the search update and movement process is repeated for each iterative search point until the position change of the search point is less than a preset threshold or the maximum number of iterations is reached. The preset threshold is used to determine whether the search point has converged to a high-density region, while the maximum number of iterations is used to limit the time complexity of the search process. When the position change of the search point is less than the preset threshold, it indicates that the search point has stabilized in a high-density region; when the maximum number of iterations is reached, even if the search point has not fully converged, it is considered to be close to a high-density region. Finally, multiple target search points are obtained.

[0049] Finally, the operating state parameters corresponding to these target search points are used as multiple typical operating conditions. Each target search point represents a high-density region in the operating state parameter space, which typically contains a large number of data points with similar characteristics. Therefore, the operating state parameters corresponding to the target search points can effectively reflect the typical characteristics of the cable under different operating conditions, and are thus identified as typical operating conditions. Through these typical operating conditions, subsequent cable thermal management optimization can be more precise, enabling targeted adjustments based on different operating conditions to improve the cable's current carrying capacity and heat dissipation efficiency.

[0050] Through the above process, this application determines several typical operating conditions in the operating state parameter space through unsupervised multi-point iterative search. This process not only ensures that the search points can gradually converge to high-density regions, but also improves search efficiency by dynamically adjusting the search direction and step size.

[0051] P30: Based on the multiple typical working conditions, the porosity, thermal conductivity, mechanical properties and filling slurry ratio parameters of the high thermal conductivity carbon mesh material are matched and screened to determine multiple matching thermal conductivity material combinations, wherein each typical working condition corresponds to one matching thermal conductivity material combination.

[0052] Optionally, after identifying several typical operating conditions, the next key task is to match and screen the properties of high thermal conductivity carbon mesh materials according to these operating conditions. That is, to determine the most suitable combination of thermally conductive materials based on the specific requirements of each typical operating condition in order to optimize the heat dissipation performance of the cable.

[0053] First, for each typical operating condition, analyze its corresponding operating parameters, such as current load, ambient temperature, and pipe spacing. These parameters determine the heat generation and dissipation requirements of the cable under that condition. For example, under high-load operating conditions, the cable generates more heat, so materials with high thermal conductivity are needed to quickly conduct heat; while under conditions with high ambient temperature, the porosity of the material may need to be adjusted appropriately to optimize airflow and assist in heat dissipation.

[0054] Next, the porosity of the high thermal conductivity carbon mesh material was screened based on the heat dissipation requirements of typical operating conditions. Porosity affects the thermal conductivity and weight of the material. Generally, higher porosity can reduce the material density, but may sacrifice some thermal conductivity. Therefore, a balance needs to be found between porosity and thermal conductivity, for example, by determining the appropriate porosity range for each typical operating condition through experimental and simulation analysis. For instance, for operating conditions with high heat dissipation requirements, materials with lower porosity are selected to improve thermal conductivity efficiency; while for weight-sensitive operating conditions, the porosity is appropriately increased to reduce the material weight.

[0055] Simultaneously, the thermal conductivity of the materials is precisely matched. Thermal conductivity is a key indicator for measuring a material's heat dissipation capability. Through experimental testing and material database queries, a range of thermal conductivity that can meet the heat dissipation requirements under each typical operating condition is selected. For example, under high-load conditions, carbon mesh materials with higher thermal conductivity are selected to ensure that heat can be quickly conducted to the surrounding environment; while under low-load conditions, materials with moderate thermal conductivity can be selected to balance cost and performance.

[0056] In addition, the mechanical properties of the material must be considered. High thermal conductivity carbon mesh materials need sufficient mechanical strength to withstand various mechanical stresses during cable operation. Mechanical properties include tensile strength, compressive strength, and modulus of elasticity. Based on mechanical stress analysis under typical operating conditions, the required mechanical performance indicators of the material are determined. For example, in cases where the spacing between cable ducts is small, the material needs to have high compressive strength to prevent deformation or damage during the filling process.

[0057] Finally, the mixing ratio of the filler slurry was optimized. The mixing ratio of the filler slurry directly affects the fluidity and curing characteristics of the material. Through experimental research on the fluidity, curing time, and post-curing strength of the slurry under different mixing ratios, the appropriate slurry mixing ratio for each typical working condition was determined. For example, in conditions requiring rapid filling, a slurry mixing ratio with better fluidity was selected; while in conditions requiring high-strength curing, the mixing ratio was adjusted to improve the mechanical properties after curing.

[0058] Through the comprehensive analysis and screening process described above, a suitable combination of thermally conductive materials was determined for each typical operating condition. These combinations include optimized porosity, thermal conductivity, mechanical properties, and filler slurry ratio parameters, ensuring optimal heat dissipation and mechanical stability under different operating conditions.

[0059] P40: Based on the multiple typical cable heat generation and heat transfer models and multiple matching thermal conductive material combinations, full-size pipe heat generation and heat transfer simulation is performed. Multiple thermal conductive material filling strategies are constructed, and the multiple typical working conditions and the multiple thermal conductive material filling strategies are mapped and associated to construct an automated filling strategy library. Each thermal conductive material filling strategy includes filling injection flow rate, injection pressure, filling sequence and allowable solidification time window.

[0060] It should be understood that, in order to determine the optimal thermal conductivity material filling strategy under different operating conditions through simulation analysis, the following steps are taken: First, finite element analysis (FEA) software or other suitable simulation tools are used to model the cable and its surrounding environment within a full-size conduit. During modeling, the cable's geometric parameters need to be set in detail, including its diameter, length, and conduit spacing. Simultaneously, based on the previously matched thermal conductivity material combination, the properties of the filling material, such as thermal conductivity and porosity, need to be set. Furthermore, corresponding operating conditions, such as current load and ambient temperature, need to be set according to typical operating conditions. The previously constructed typical cable heat generation and transfer model is then integrated into the simulation environment to ensure that the model accurately reflects the thermal behavior of the cable under different operating conditions.

[0061] Next, full-scale heat transfer simulations of the pipe network were conducted for each typical operating condition and its corresponding matching thermally conductive material combination. During the simulation, the focus was on the heat transfer path and temperature distribution within the cable, filler material, and surrounding environment. The simulation results were optimized by adjusting the filler material parameters to meet the heat dissipation requirements. Simultaneously, key parameters during the simulation, such as peak temperature and heat distribution uniformity, were recorded; these parameters will serve as the basis for subsequent filler strategy development.

[0062] After completing the simulation analysis, the next step is to construct multiple thermally conductive material filling strategies based on the simulation results. This filling strategy refers to the scheme of efficiently and uniformly injecting the thermally conductive material into the cable conduit to maximize its heat dissipation effect. Specific filling strategies include several key parameters such as injection flow rate, injection pressure, filling sequence, and allowable solidification time window. First, based on the heat distribution and flow characteristics of the filling material in the simulation, a suitable injection flow rate is determined. The injection flow rate needs to ensure uniform distribution of the filling material while avoiding material waste or uneven filling due to excessive flow. Therefore, the optimal injection flow rate range for each typical operating condition can be determined through simulation analysis results. Second, the injection pressure affects the fluidity and filling efficiency of the filling material. During the simulation, by comparing the flow path and filling effect of the filling material under different pressures, the optimal pressure range that ensures smooth material filling without damaging the cable and conduit is determined. Simultaneously, considering equipment limitations and safety requirements in actual operation, the injection pressure is ensured to be within a reasonable range. Third, the filling sequence determines the distribution of the filling material within the conduit. By comparing simulation analysis results, the optimal filling sequence can be determined to ensure efficient heat conduction. For example, in some operating conditions, it may be necessary to fill the portion near the cable first to quickly establish a heat dissipation channel; while in other conditions, it may be necessary to start filling from both ends of the conduit to avoid material accumulation in the middle area. Based on simulation results, the filling sequence for each typical operating condition is determined. Finally, the solidification time of the filler material directly affects the filling effect and subsequent operations. Therefore, the solidification characteristics of filler materials with different ratios can be studied experimentally, and combined with simulation results, the allowable solidification time window for each typical operating condition can be determined. This time window needs to ensure that the material can solidify within a suitable time after filling, while avoiding uneven filling or structural instability caused by solidification that is too fast or too slow.

[0063] Finally, multiple typical operating conditions are mapped and associated with corresponding thermal conductive material filling strategies to construct an automated filling strategy library. This library contains filling strategies for different operating conditions, allowing for the rapid retrieval of suitable filling schemes based on actual needs. In practical applications, the automated system retrieves and loads strategies from the library, automatically selecting the most suitable filling strategy for the current operating conditions, thus ensuring that the thermal conductive material filling effect is always optimal. The application of the automated filling strategy library can significantly improve the efficiency and accuracy of the thermal conductive material filling process in cable ducts.

[0064] P50: Collect real-time running status characteristic parameters, search the automated filling strategy library, load the real-time automated filling strategy, start the automated injection operation and perform synchronous observation to obtain the synchronous observation sequence.

[0065] Optionally, in order to realize the real-time operation and monitoring of the automated filling system, it is necessary to collect real-time operating status characteristic parameters, and retrieve and load the corresponding real-time automated filling strategy from the automated filling strategy library based on these parameters.

[0066] First, a sensor network installed in the cable duct system collects operational status parameters in real time. These parameters include, but are not limited to, real-time current and voltage values ​​of the cables, ambient temperature, humidity, and airflow velocity within the duct. These sensors continuously monitor and report changes in the cable's status during actual operation, ensuring the system can acquire the latest operational data.

[0067] Subsequently, based on the collected real-time operating status characteristic parameters, the automated filling strategy library is searched. The strategy library stores multiple typical operating conditions and their corresponding thermally conductive material filling strategies. Each strategy includes parameters such as filling injection flow rate, injection pressure, filling sequence, and allowable solidification time window. The search process matches the real-time operating status characteristic parameters with the typical operating conditions in the strategy library to find the operating condition closest to the current operating conditions and loads the corresponding filling strategy. This process can be implemented through machine learning or algorithmic models, making the selection of filling strategies both accurate and efficient.

[0068] After the real-time automated filling strategy is loaded, the central control system sends a command to the automated injection equipment to initiate the automated injection operation. Based on the loaded strategy, the injection equipment precisely controls the injection flow rate and pressure of the thermally conductive material, filling it according to a predetermined filling sequence. Simultaneously, a synchronous observation program is initiated, using a sensor network to monitor key parameters during the filling process in real time, such as changes in the temperature and pressure of the filling material, and the filling position, forming a synchronous observation sequence.

[0069] For example, the acquisition of synchronous observation sequences is achieved by continuously collecting data during the filling process. A sensor network collects data at set time intervals, such as per second or per minute, and transmits this data to a central control system in real time. The central control system analyzes the collected data to assess whether the filling operation is proceeding smoothly according to the predetermined strategy. For instance, by monitoring temperature changes in the filler material, it can be determined whether the material has solidified normally within the allowable solidification time window; by monitoring pressure changes, it can be ensured that the injection pressure is within a safe range and meets the requirements of the filling strategy. These synchronous observation sequences serve as the basis for cable performance analysis, providing crucial feedback for long-term cable operation and helping to continuously optimize cable heat dissipation management strategies and filler material selection.

[0070] P60: The synchronous observation sequence is identified using an observation offset detector and a control action uncertainty detector, and the real-time automated filling strategy is calibrated based on the identification results.

[0071] Furthermore, step P60 in this embodiment of the application also includes:

[0072] P61: Input the current observation, which is at the end of the synchronous observation sequence, into the observation offset detector to obtain the observation offset, wherein the observation offset detector includes a target network layer with fixed parameters and a trainable prediction network layer; P62: Transmit the synchronous observation sequence to the control action uncertainty detector to obtain the control action uncertainty coefficient; P63: Perform weighted analysis on the observation offset and the control action uncertainty coefficient to obtain the identification result; P64: When the identification result indicates the presence of an anomaly, trigger an automated safety intervention mechanism to calibrate the real-time automated filling strategy.

[0073] The automated safety intervention mechanism includes at least one or a combination of the following: reducing the injection pressure / flow rate to a safe level, switching to pulse injection and extending the venting time, isolating the current pipe section in segments, performing local back-pulling / re-injection, pausing filling and prompting for manual inspection.

[0074] It should be understood that, in order to ensure the accuracy and safety of the automated filling process, this application uses an observation offset detector and a control action uncertainty detector to identify the synchronous observation sequence, and calibrates the real-time automated filling strategy based on the identification results.

[0075] During automated filling, the real-time acquired synchronous observation sequence contains several key parameters, including the injection pressure of the thermally conductive material, the injection flow rate, the pressure drop within the pipe, the slurry return state or bubble characteristics at the venting end, and the operating status parameters of the injection equipment. These observation sequences reflect the real-time status of the filling process and are an important basis for assessing whether the filling operation is proceeding normally.

[0076] First, the current observation, located at the end of the synchronous observation sequence, is input into the observation offset detector. The observation offset detector calculates the observation offset by comparing the current observation with a preset distribution of successful fill states. This observation offset characterizes the degree of deviation of the current observation state from the distribution of successful fill states. The observation offset detector consists of a target network layer with fixed parameters and a trainable prediction network layer. The target network layer is trained based on a large amount of data from successful fill cases to determine the feature distribution of successful fill states; the prediction network layer predicts the degree of deviation of the current observation from the target distribution in real time, thus obtaining the observation offset. Specifically, the target network layer extracts the feature vector of successful fill states using deep learning algorithms, such as convolutional neural networks, while the prediction network layer adopts a similar network structure, but its weights are continuously updated through online learning to adapt to changes in real-time data.

[0077] Next, the entire synchronous observation sequence is transmitted to the control action uncertainty detector. The time-series characteristics of the observation sequence are analyzed to assess the uncertainty of the control actions during the filling process, generating a control action uncertainty coefficient. The control action uncertainty coefficient measures the magnitude of the uncertainty of control parameters (such as injection pressure and flow rate) during the filling process. High control action uncertainty indicates that the system may experience significant fluctuations or instability during the filling operation, which can affect the filling effect and material uniformity. Therefore, the control action uncertainty detector needs to analyze the changes in control parameters during the filling process based on the input synchronous observations and output an uncertainty coefficient. For example, this coefficient is calculated based on the statistical characteristics of the observation sequence, such as variance and standard deviation, as well as the dynamic trend of the time series. The control action uncertainty detector uses recurrent neural network structures such as Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU) to capture long-term dependencies in the time-series data, thereby accurately assessing the uncertainty of the control actions.

[0078] Subsequently, a weighted analysis is performed on the observation offset and the uncertainty coefficient of the control action to obtain a comprehensive identification result. The weighted analysis first considers the degree of influence of different parameters on the filling process, assigning different weights to the observation offset and the uncertainty coefficient of the control action. For example, the observation offset may have a greater impact on the filling quality, so it is given a higher weight; while the uncertainty coefficient of the control action reflects the stability of the operation more, so it is given an appropriate weight. For example, the weight allocation can be determined through experimental and simulation analysis to ensure that it accurately reflects the abnormal state of the filling process under different operating conditions. Next, the observation offset and the uncertainty coefficient of the control action are weighted according to the assigned weights to comprehensively evaluate the state value of the current filling process, and the comprehensive state value is compared with a preset state threshold to determine whether an anomaly exists.

[0079] When an anomaly is detected, an automated safety intervention mechanism is triggered to calibrate the real-time automated filling strategy. This mechanism responds promptly to anomalies during the filling process, preventing potential filling problems and ensuring cable heat dissipation and operational stability. Specific safety intervention measures include, but are not limited to: reducing injection pressure or flow to a safe level to avoid uneven filling or equipment damage caused by excessive pressure or flow; switching to pulse injection and extending venting time to improve filler material distribution and reduce air bubble formation; isolating the current pipe section in segments, performing local back-pulling and re-injection to resolve localized uneven filling; pausing filling and prompting for manual inspection to facilitate intervention when necessary. These measures can be selected and combined based on the specific type and severity of the anomaly, ensuring the filling process can quickly return to normal. For example, when the observed offset exceeds a preset threshold, the system automatically reduces the injection pressure to a safe level; when the uncertainty coefficient of the control action is high, it switches to pulse injection mode; when two anomalies occur simultaneously, segmented isolation and local back-pulling operations are performed. This process not only improves the accuracy and safety of the filling operation but also ensures that measures can be taken quickly when anomalies occur through the automated safety intervention mechanism, avoiding potential risks and losses.

[0080] Furthermore, step P61 in the embodiments of this application also includes:

[0081] P61-1: Obtain the historical synchronous observation sequence that was successfully filled as training data; P61-2: Train the convolutional neural network with randomly initialized parameters using the training data, and maintain fixed parameters during training to perform feature mapping only on the input observations to obtain the target network layer; P61-3: Input the training data into the target network layer and the prediction network layer with trainable parameters in sequence, perform joint training fitting to obtain the trained prediction network layer, and integrate the target network layer and the prediction network layer to obtain the observation offset detector; P61-4: Input the current observation into the target network layer and the prediction network layer to obtain the output of the target network layer and the output of the prediction network layer; P61-5: Calculate the difference between the output of the target network layer and the output of the prediction network layer to obtain the observation offset metric.

[0082] Optionally, the observation offset detector is constructed to monitor the sequence of synchronous observations during the automated filling process in real time and to assess its deviation from the successful filling state.

[0083] First, historical synchronous observation sequences under successful filling conditions are collected as training data. This data includes injection pressure, injection flow rate, pressure drop within the manifold, slurry return status or bubble characteristics at the exhaust end, and operating parameters of the injection equipment. This historical data provides realistic and representative input for training the neural network.

[0084] Next, this training data is used to train a convolutional neural network with randomly initialized parameters. During training, a portion of the network's parameters are fixed; this portion is called the target network layer. The role of the target network layer is to define a stable reference mapping. It performs feature mapping on the input observations, providing an invariant benchmark for subsequent comparisons and analysis. The training of the target network layer is conducted within a supervised learning framework, where the network learns to map the input data to a feature space that captures the key features of successfully filled states.

[0085] Then, the training data is sequentially fed into a pre-trained target network layer and a prediction network layer with trainable parameters. The target network layer defines the successfully filled state using fixed mapping rules, while the prediction network layer learns features from the training data and mimics the mapping of the target network layer. The learning objective of the prediction network layer is to reproduce the output of the target network layer, rather than directly learning the data itself. Therefore, the task of the prediction network layer is not to learn the specific content of the data, but to learn how to reproduce the mapping of the target network. This process involves a contrastive learning strategy, where the output of the prediction network layer is compared with the output of the target network layer, and the difference between the two is quantified by a loss function. After training, the target network layer and the prediction network layer are integrated into an observation offset detector.

[0086] After the observation offset detector is constructed, the current observations are input into the target network layer and the prediction network layer to obtain the outputs of the target network layer and the prediction network layer, respectively. The target network layer serves as a reference standard to compare with the output of the prediction network layer, ensuring that the output of the prediction network layer is as close as possible to the mapping of the target network.

[0087] Finally, the difference between the target network layer output and the predicted network layer output is calculated; this difference is called the observation offset metric. The observation offset metric characterizes the degree to which the current infill process's observed state deviates from the distribution of successfully infilled states. If the observation offset metric exceeds a preset threshold, it indicates that the current infill process may differ significantly from the successfully infilled state, and adjustments or intervention may be necessary.

[0088] In this step, because the predictive network layer learns to mimic the mapping of the target network layer, rather than the data itself, it can detect out-of-distribution data even when trained using only successful data. This capability makes the observation offset detector highly valuable in practical applications, as it can identify potential problems in the filling process even without explicitly labeled anomalous data. In this way, the system can be trained using only successful data to identify any anomalies that deviate from the normal state.

[0089] Furthermore, step P61-3 in the embodiments of this application also includes:

[0090] P61-31: The output of the target network layer is used as the learning target of the prediction network layer to be trained, which is a trainable parameter; P61-32: The parameters of the prediction network layer are iteratively trained by minimizing the difference between the output of the prediction network layer and the output of the target network layer until convergence, and the trained prediction network layer is obtained.

[0091] Specifically, the training process of the prediction network layer in the observation offset detector can be further refined to ensure that the prediction network layer can accurately reproduce the feature mapping of the target network layer.

[0092] First, the output of the target network layer is used as the learning objective of the prediction network layer. The target network layer performs feature mapping on the input observations using its fixed parameters, generating a stable output that represents the feature representation of the successfully filled state. The task of the prediction network layer is to learn how to generate feature mappings similar to the output of the target network layer. This process uses the output of the target network layer as a supervisory signal to guide the training of the prediction network layer.

[0093] Next, to ensure the prediction network layer accurately reproduces the feature mappings of the target network layer, its parameters are iteratively trained by minimizing the difference between their outputs. This process is achieved by calculating a loss function that measures the degree of difference between the prediction and target network layer outputs. Common loss functions include mean squared error (MSE) or mean absolute error (MAE). During training, the parameters of the prediction network layer are adjusted using backpropagation algorithms and optimizers, such as gradient descent or its variants, to reduce the value of the loss function. Specifically, the network iteratively optimizes its parameters to reduce the gap between the two until the loss function converges to a small, stable value, indicating that the prediction network layer has successfully learned how to reproduce the feature mappings of the target network layer. At this point, the training of the prediction network layer is complete, and it can be used together with the target network layer to form an observation offset detector.

[0094] Through this training method, the prediction network layer can gradually learn how to reproduce the output of the target network layer and has strong generalization ability after training. This design ensures that the prediction network layer can accurately mimic the ideal output of a successful filling state and effectively monitor and detect any anomalies or deviations during future filling processes.

[0095] Furthermore, step P62 in this embodiment of the application also includes:

[0096] P62-1: Use a control action uncertainty detector to identify the similarity of adjacent elements in the synchronous observation sequence to obtain an adjacent observation similarity matrix sequence; P62-2: Perform multi-scale fluctuation analysis on the adjacent observation similarity matrix sequence to obtain a multi-scale fluctuation feature set; P62-3: Perform weighted fusion of the multi-scale fluctuation feature set to obtain multi-scale fused fluctuation features, and perform uncertainty analysis based on the multi-scale fused fluctuation features to obtain the control action uncertainty coefficient.

[0097] In one possible embodiment of this application, the analysis process for the synchronous observation sequence is further optimized. First, for each pair of adjacent observations in the synchronous observation sequence, a similarity is calculated. Specifically, adjacent data points are sequentially selected from the synchronous observation sequence, and the similarity between these adjacent observations is calculated. For example, for each observation in the sequence, a similarity calculation function can be used to determine its similarity to the next observation. This function can be a statistically based measure, such as the Pearson correlation coefficient, or a machine learning-based measure, such as cosine similarity. After calculation, the similarity values ​​between each pair of observations are arranged into a matrix, called the adjacent observation similarity matrix. This process is repeated to generate a similarity matrix for each position in the sequence, thus forming a sequence of adjacent observation similarity matrices.

[0098] Next, multi-scale fluctuation analysis is performed on the adjacent observation similarity matrix sequence generated in the previous step. Multi-scale analysis techniques, such as wavelet transform, can be applied to identify the fluctuation characteristics of the sequence at different time scales. For example, firstly, the observation sequence is divided according to different time or spatial scales, such as dividing the data into time windows of 1 second, 10 seconds, and 1 minute. Then, at each scale, the system analyzes the fluctuation of the similarity matrix, including calculating statistical data such as standard deviation and volatility, and assessing the changing trends at different scales. These fluctuation characteristics are extracted and recorded, forming a multi-scale fluctuation feature set. These features can reflect the fluctuation performance of the filling process at different scales, thereby revealing subtle changes in the filling process.

[0099] Finally, a weighted fusion of the multi-scale fluctuation feature set is performed. By assigning different weights to fluctuation features at different scales, a weighted average or weighted sum of the fluctuation features at all scales is obtained to get a comprehensive fluctuation feature value. Specifically, the importance of each feature is first determined through cross-validation or based on prior knowledge. Then, each feature is multiplied by its corresponding weight, and the weighted features are summed to obtain a comprehensive multi-scale fused fluctuation feature. This fused feature integrates fluctuation information at different time scales and can more comprehensively reflect the uncertainty of control actions.

[0100] Next, this fused feature is used to perform uncertainty analysis and calculate the uncertainty coefficient of the control action. For example, the fused feature can be compared with a preset threshold to determine the fluctuation range, and then the uncertainty coefficient can be calculated based on the fluctuation range. This coefficient is a value between 0 and 1, where a higher value indicates higher uncertainty. Through this process, a quantified uncertainty coefficient of the control action is obtained, which can be used for subsequent automated filling strategy calibration. If this coefficient exceeds the preset threshold, the system will trigger an automated safety intervention mechanism to make necessary adjustments to the real-time automated filling strategy, thereby improving the stability and reliability of the filling operation.

[0101] In summary, the embodiments of this application have at least the following technical effects:

[0102] This application automates the filling process, precisely adjusting the injection flow rate, injection pressure, and filling sequence to avoid human error and ensure uniform and efficient filling of the thermally conductive material. Based on different operating conditions and the cable's heat dissipation requirements, it automatically selects the optimal combination of thermally conductive materials, improving the cable's thermal management flexibility. By monitoring the cable's operating status in real time, it automatically retrieves and loads the most suitable filling strategy from the filling strategy library, achieving automation of the filling process, reducing manual intervention, improving operational consistency and efficiency, and enhancing system adaptability. Utilizing an observation offset detector and a control action uncertainty detector to identify and calibrate the filling strategy in real time ensures the stability and reliability of the filling process.

[0103] This technology achieves the effect of flexibly matching material filling strategies based on the real-time operating status of cable ducts and performing automated filling, thereby improving filling accuracy and efficiency.

[0104] Example 2, based on the same inventive concept as the automated filling method for filling thermally conductive material into cable ducts in the foregoing examples, such as... Figure 2 As shown, this application provides an automated filling system for filling cable conduits with thermally conductive material. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0105] The operating status parameter acquisition module 11 is used to collect a set of operating status parameters for high-voltage duct-laid cables, using load conditions, laying environment, and duct structure dimensions as indexes.

[0106] The typical operating condition determination module 12 is used to perform unsupervised multi-point iterative search based on the set of operating state parameters to determine multiple typical operating conditions and multiple typical cable heat generation and heat transfer models.

[0107] The thermal conductive material combination matching module 13 is used to match and screen the porosity, thermal conductivity, mechanical properties and filling slurry ratio parameters of the high thermal conductivity carbon mesh material according to the multiple typical working conditions, and determine multiple matching thermal conductive material combinations, wherein each typical working condition corresponds to one matching thermal conductive material combination.

[0108] The filling strategy library construction module 14 is used to perform full-size pipe heat generation and heat transfer simulation based on the multiple typical cable heat generation and heat transfer models and multiple matching thermal conductive material combinations, construct multiple thermal conductive material filling strategies, and map and associate the multiple typical working conditions and the multiple thermal conductive material filling strategies to construct an automated filling strategy library. Each thermal conductive material filling strategy includes filling injection flow rate, injection pressure, filling sequence and allowable solidification time window.

[0109] The automated filling observation module 15 is used to collect real-time running status feature parameters, search the automated filling strategy library, load real-time automated filling strategies, start automated injection operations and perform synchronous observations to obtain a synchronous observation sequence.

[0110] The strategy calibration module 16 is used to identify the synchronous observation sequence using an observation offset detector and a control action uncertainty detector, and to calibrate the real-time automated filling strategy based on the identification results.

[0111] Furthermore, the typical operating condition determination module 12 is also used to perform the following steps:

[0112] Each operating state parameter in the set of operating state parameters is mapped to a D-dimensional space to construct an operating state parameter space, where D is the dimension of the operating state parameter and is a positive integer. Multiple initial search points are selected from the operating state parameter space through random sampling. The multiple initial search points are iteratively jointly verified. When the verification is successful, the multiple initial search points are traversed to perform an unsupervised multi-point iterative search in the operating state parameter space to determine multiple typical operating conditions. Multiple cable heat generation models and multiple cable heat transfer models are constructed based on the multiple typical operating conditions. The multiple cable heat generation models and multiple cable heat transfer models are mapped and coupled to obtain the multiple typical cable heat generation and heat transfer models.

[0113] Furthermore, the typical operating condition determination module 12 is also used to perform the following steps:

[0114] The initial search points are enumerated pairwise to obtain an initial search point enumeration combination set; similarity identification is performed on the initial search point enumeration combination set to obtain an enumeration combination similarity set; the number of enumeration combination similarities in the enumeration combination similarity set that is greater than or equal to a preset similarity threshold is counted, and it is determined whether the statistical result is greater than or equal to the preset threshold. If it is, the verification fails; if not, the verification passes.

[0115] Furthermore, the typical operating condition determination module 12 is also used to perform the following steps:

[0116] Iterate through each of the multiple initial search points and perform an iterative search process within the running state parameter space; wherein the iterative search process includes:

[0117] Based on the distribution of operating state parameters around the initial search point, the local state density is calculated and the search update direction is determined. The initial search point is updated along the direction of increasing state density, so that the initial search point gradually moves towards the high-density area of ​​operating state parameters to obtain iterative search points. The search update and movement iteration process is repeated for iterative search points until the position change of the search point is less than a preset threshold or the maximum number of iterations is reached to obtain multiple target search points. The operating state parameters corresponding to the multiple target search points are used as multiple typical operating conditions.

[0118] Furthermore, the strategy calibration module 16 is also used to perform the following steps:

[0119] The current observation, located at the end of the synchronous observation sequence, is input into the observation offset detector to obtain the observation offset. The observation offset detector includes a target network layer with fixed parameters and a trainable prediction network layer. The synchronous observation sequence is transmitted to the control action uncertainty detector to obtain the control action uncertainty coefficient. A weighted analysis is performed on the observation offset and the control action uncertainty coefficient to obtain the identification result. When the identification result indicates the presence of an anomaly, an automated safety intervention mechanism is triggered to calibrate the real-time automated filling strategy.

[0120] Furthermore, the strategy calibration module 16 is also used to perform the following steps:

[0121] The successful filling of historical synchronous observation sequences is obtained as training data. A convolutional neural network with randomly initialized parameters is trained using this training data, maintaining fixed parameters during training and performing feature mapping only on the input observations to obtain the target network layer. The training data is then sequentially input into the target network layer and a prediction network layer with trainable parameters to be trained, performing joint training and fitting to obtain the trained prediction network layer. The target network layer and the prediction network layer are integrated to obtain the observation offset detector. The current observation is input into the target network layer and the prediction network layer to obtain the outputs of the target network layer and the prediction network layer. The difference between the outputs of the target network layer and the prediction network layer is calculated to obtain the observation offset metric.

[0122] Furthermore, the strategy calibration module 16 is also used to perform the following steps:

[0123] The output of the target network layer is used as the learning target of the prediction network layer to be trained, which is a trainable parameter. The parameters of the prediction network layer are iteratively trained by minimizing the difference between the output of the prediction network layer and the output of the target network layer until convergence, and the trained prediction network layer is obtained.

[0124] Furthermore, the strategy calibration module 16 is also used to perform the following steps:

[0125] The synchronous observation sequence is subjected to adjacent similarity identification of similar elements using a control action uncertainty detector to obtain an adjacent observation similarity matrix sequence; multi-scale fluctuation analysis is performed on the adjacent observation similarity matrix sequence to obtain a multi-scale fluctuation feature set; weighted fusion of the multi-scale fluctuation feature set is performed to obtain multi-scale fused fluctuation features, and uncertainty analysis is performed based on the multi-scale fused fluctuation features to obtain the uncertainty coefficient of the control action.

[0126] Furthermore, in the strategy calibration module 16:

[0127] The automated safety intervention mechanism includes at least one or a combination of the following: reducing the injection pressure / flow rate to a safe level, switching to pulse injection and extending the venting time, isolating the current pipe segment in sections, performing local back-pull / re-injection, pausing filling and prompting for manual inspection.

[0128] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0129] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0130] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. An automated filling method for filling cable conduits with thermally conductive material, characterized in that, The method includes: Using load conditions, laying environment, and duct structure dimensions as indexes, a set of operating status parameters for high-voltage duct-laid cables is collected. Based on the set of operating state parameters, an unsupervised multi-point iterative search is performed to determine multiple typical operating conditions and multiple typical cable heat generation and heat transfer models. Based on the aforementioned multiple typical working conditions, the porosity, thermal conductivity, mechanical properties, and filling slurry ratio parameters of the high thermal conductivity carbon mesh material are matched and screened to determine multiple matching thermal conductivity material combinations. Each typical working condition corresponds to one matching thermal conductivity material combination. Based on the aforementioned typical cable heat generation and heat transfer models and multiple matching thermal conductive material combinations, full-size pipe heat generation and heat transfer simulations were performed. Multiple thermal conductive material filling strategies were constructed, and the aforementioned typical operating conditions and multiple thermal conductive material filling strategies were mapped and associated to construct an automated filling strategy library. Each thermal conductive material filling strategy includes filling injection flow rate, injection pressure, filling sequence, and allowable solidification time window. Collect real-time running status characteristic parameters, search the automated filling strategy library, load the real-time automated filling strategy, start the automated injection operation and perform synchronous observation to obtain the synchronous observation sequence. The synchronous observation sequence is identified using an observation offset detector and a control action uncertainty detector, and the real-time automated filling strategy is calibrated based on the identification results.

2. The automated filling method for filling thermally conductive material into cable conduits as described in claim 1, characterized in that, Based on the aforementioned set of operating state parameters, an unsupervised multi-point iterative search is performed to determine multiple typical operating conditions and multiple typical cable heat generation and transfer models, including: Each running state parameter in the set of running state parameters is mapped to a D-dimensional space to construct a running state parameter space, where D is the dimension of the running state parameter and is a positive integer; Multiple initial search points are selected from the operating state parameter space by random sampling; The multiple initial search points are subjected to iterative joint verification. When the verification is passed, the multiple initial search points are traversed in the running state parameter space to perform unsupervised multi-point iterative search to determine multiple typical working conditions. Based on the aforementioned typical operating conditions, multiple cable heat generation models and multiple cable heat transfer models are constructed. By mapping and coupling the multiple cable heat generation models and multiple cable heat transfer models, the multiple typical cable heat generation and heat transfer models are obtained.

3. The automated filling method for filling thermally conductive material into cable conduits as described in claim 2, characterized in that, Iterative joint verification of the multiple initial search points includes: Enumerate each of the multiple initial search points in pairs to obtain a set of enumerated combinations of initial search points; The initial search point enumeration combination set is subjected to similarity identification to obtain the enumeration combination similarity set; The number of similarities in the enumerated combination set that are greater than or equal to a preset similarity threshold is counted, and it is determined whether the statistical result is greater than or equal to the preset threshold. If so, the verification fails. If not, then the verification passes.

4. The automated filling method for filling thermally conductive material into cable conduits as described in claim 3, characterized in that, Upon successful verification, an unsupervised multi-point iterative search is performed within the operating state parameter space, traversing multiple initial search points, to determine several typical operating conditions, including: Traverse each of the multiple initial search points and perform an iterative search process within the running state parameter space; The iterative search process includes: Based on the distribution of operating state parameters around the initial search point, calculate the local state density and determine the search update direction; The initial search point is updated along the direction of increasing state density, so that the initial search point gradually moves towards the high-density region of running state parameters to obtain the iterative search point; The iterative search, update, and move iteration process is repeated for the search point until the position change of the search point is less than a preset threshold or the maximum number of iterations is reached, thus obtaining multiple target search points. The operating status parameters corresponding to the multiple target search points are used as multiple typical operating conditions.

5. An automated filling method for filling thermally conductive material into cable conduits as described in claim 1, characterized in that, The synchronous observation sequence is identified using an observation offset detector and a control action uncertainty detector, and the real-time automated filling strategy is calibrated based on the identification results, including: The current observation, which is located at the end of the synchronous observation sequence, is input into the observation offset detector to obtain the observation offset. The observation offset detector includes a target network layer with fixed parameters and a trainable prediction network layer. The synchronized observation sequence is transmitted to the control action uncertainty detector to obtain the control action uncertainty coefficient; A weighted analysis is performed on the observed offset and the uncertainty coefficient of the control action to obtain the identification result; When the identification result indicates an anomaly, an automated security intervention mechanism is triggered to calibrate the real-time automated filling strategy.

6. An automated filling method for filling thermally conductive material into cable conduits as described in claim 5, characterized in that, The current observation, located at the end of the synchronized observation sequence, is input into the observation offset detector to obtain the observation offset. The observation offset detector comprises a target network layer with fixed parameters and a trainable prediction network layer, including: Obtain the historical synchronous observation sequence that was successfully filled as training data; The target network layer is obtained by training a convolutional neural network with randomly initialized parameters using training data, while keeping the parameters fixed during training and performing feature mapping only on the input observations. The training data is sequentially input into the target network layer and the prediction network layer with trainable parameters, and joint training fitting is performed to obtain the trained prediction network layer. The target network layer and the prediction network layer are then integrated to obtain the observation offset detector. The current observation is input into the target network layer and the prediction network layer to obtain the output of the target network layer and the output of the prediction network layer. The difference between the output of the target network layer and the output of the predicted network layer is calculated to obtain the observation offset metric.

7. An automated filling method for filling thermally conductive material into cable conduits as described in claim 6, characterized in that, The training data is sequentially input into the target network layer and the prediction network layer with trainable parameters, and joint training and fitting are performed to obtain the trained prediction network layer, including: The output of the target network layer is used as the learning target of the prediction network layer to be trained, which is a trainable parameter. By minimizing the difference between the output of the prediction network layer and the output of the target network layer, the parameters of the prediction network layer are iteratively trained until convergence, thus obtaining the trained prediction network layer.

8. An automated filling method for filling thermally conductive material into cable conduits as described in claim 6, characterized in that, The synchronized observation sequence is transmitted to the control action uncertainty detector to obtain the control action uncertainty coefficient, including: The similarity of adjacent elements in the same category is identified by using a control action uncertainty detector to obtain an adjacent observation similarity matrix sequence; Multi-scale fluctuation analysis is performed on the adjacent observation similarity matrix sequence to obtain a multi-scale fluctuation feature set; A weighted fusion of the multi-scale fluctuation feature set is performed to obtain multi-scale fused fluctuation features, and uncertainty analysis is performed based on the multi-scale fused fluctuation features to obtain the uncertainty coefficient of the control action.

9. An automated filling method for filling thermally conductive material into cable conduits as described in claim 5, characterized in that, The automated safety intervention mechanism includes at least one or a combination of the following: reducing the injection pressure / flow rate to a safe level, switching to pulse injection and extending the venting time, isolating the current pipe segment in sections, performing local back-pull / re-injection, pausing filling and prompting for manual inspection.

10. An automated filling system for filling cable conduits with thermally conductive material, characterized in that, The system includes: The operating status parameter acquisition module is used to collect a set of operating status parameters for high-voltage duct-laid cables, indexed by load conditions, laying environment, and duct structure dimensions. The typical operating condition determination module is used to perform unsupervised multi-point iterative search based on the set of operating state parameters to determine multiple typical operating conditions and multiple typical cable heat generation and heat transfer models. The thermally conductive material combination matching module is used to match and screen the porosity, thermal conductivity, mechanical properties and filling slurry ratio parameters of the high thermal conductivity carbon mesh material according to the multiple typical working conditions, and determine multiple matching thermally conductive material combinations, wherein each typical working condition corresponds to one matching thermally conductive material combination; The filling strategy library construction module is used to perform full-size pipe heat generation and heat transfer simulation based on the multiple typical cable heat generation and heat transfer models and multiple matching thermal conductive material combinations, construct multiple thermal conductive material filling strategies, and map and associate the multiple typical working conditions and the multiple thermal conductive material filling strategies to build an automated filling strategy library. Each thermal conductive material filling strategy includes filling injection flow rate, injection pressure, filling sequence and allowable solidification time window. The automated filling observation module is used to collect real-time running status feature parameters, search the automated filling strategy library, load real-time automated filling strategies, start automated injection operations and perform synchronous observations to obtain a synchronous observation sequence. The strategy calibration module is used to identify the synchronous observation sequence using an observation offset detector and a control action uncertainty detector, and to calibrate the real-time automated filling strategy based on the identification results.

Images