Intelligent modeling method and system for electric power survey design based on multi-source data fusion

Abstract

The application relates to the technical field of electric power survey design intelligent modeling, and discloses a power survey design intelligent modeling method and system based on multi-source data fusion, which comprises the following steps: acquiring and structurally processing multi-source data such as historical design, surveying and mapping and manual collection, constructing a multi-dimensional reliability vector based on data engineering attributes, initializing a hierarchical modeling constraint system containing a hard constraint layer, a semi-hard constraint layer and a soft constraint layer, dynamically scheduling the constraint levels through a rule-constrained machine learning model, determining a constraint transmission coefficient, and finally adjusting the modeling object parameters and updating the model under the joint control of the two. The application realizes accurate quantification of data reliability and intelligent scheduling of constraints, suppresses the chain amplification of data conflicts, reduces manual intervention, and improves the engineering implementability and efficiency of the modeling results.

Description

Technical Field

[0001] This application relates to the field of intelligent modeling technology for power survey and design, specifically to an intelligent modeling method and system for power survey and design based on multi-source data fusion. Background Technology

[0002] In the process of power grid survey and design, modeling work needs to integrate multi-source data, including historical design data, survey and mapping data, and manually collected data. These data vary significantly in terms of collection time, method, accuracy level, and engineering background, leading to inherent inconsistencies in their engineering reliability.

[0003] Existing modeling methods generally treat all types of input data as equivalent and reliable, and perform fusion processing only through simple weighting or priority sorting. When different data have spatial conflicts, contradictory parameters, or logical inconsistencies, they often rely on manual experience for correction or directly use the latest data to cover the data, which can easily introduce systematic errors.

[0004] Furthermore, existing technologies still have many unresolved key issues. For example, data credibility is a dynamic attribute related to the modeling stage and the modeling object, while existing technologies can only handle it statically. Credibility differences not only affect parameter values ​​but also whether constraints take effect, and existing technologies cannot achieve fine-grained control. Conflicts between multiple data sources are easily amplified along the model structure, and deviations at a single location can lead to overall deformation. There is a lack of a mechanism to transform data credibility into internal constraint control variables of the model, making it difficult to avoid the problem of data conflict amplification, which seriously affects the engineering feasibility of the modeling results. Summary of the Invention

[0005] The purpose of this application is to provide an intelligent modeling method and system for power survey and design based on multi-source data fusion, so as to solve the problems mentioned in the background art.

[0006] According to the first aspect of this application, a smart modeling method for power survey and design based on multi-source data fusion is provided, comprising the following steps:

[0007] S1. Acquire and structure multi-source survey and design data, including historical design data, survey and mapping data, and manually collected data;

[0008] S2. Based on the engineering attributes of the multi-source survey and design data, construct a multi-dimensional credibility vector for each type or group of data. The multi-dimensional credibility vector is used to quantitatively characterize the credibility of the data in different dimensions.

[0009] S3. Based on the preset constraint level definition, initialize a hierarchical modeling constraint system containing at least two different constraint levels. The constraint level includes at least a hard constraint layer, a semi-hard constraint layer, and a soft constraint layer, and initially classify different design constraints to the corresponding constraint levels.

[0010] S4. Using the multidimensional confidence vector as input, a rule-based machine learning model is used to dynamically schedule at least some constraints in the hierarchical modeling constraint system, outputting the probability distribution of the at least some constraints belonging to each constraint level, and determining the constraint level corresponding to the maximum probability as the adjusted effective constraint level. The machine learning model learns the mapping relationship between the confidence vector and the constraint level adjustment based on pre-trained rules and samples.

[0011] S5. Based on the effective constraint hierarchy and the multidimensional confidence vector, determine the constraint transfer coefficient associated with each constraint. The constraint transfer coefficient is used to quantify and control the strength of the corresponding constraint when it is transferred between adjacent modeling objects in the model structure.

[0012] S6. Under the joint control of the effective constraint level and the constraint transfer coefficient, the parameters of the modeling object are adjusted and the model is updated to generate a power survey and design model.

[0013] Preferably, constructing a multidimensional credibility vector specifically includes extracting engineering attribute features from at least two of the multi-source survey and design data in the time dimension, spatial dimension, specification dimension, and structural dimension; quantifying the engineering attribute features into numerical features; and combining them in a preset order to form the multidimensional credibility vector.

[0014] Preferably, the engineering attribute characteristics of the time dimension include the time difference between the data acquisition time and the current design stage; the engineering attribute characteristics of the spatial dimension include the degree of consistency between the data and other survey or design data in the same area within a local region; the engineering attribute characteristics of the specification dimension include the difference between the design specification version followed by the data and the current design specification; and the engineering attribute characteristics of the structural dimension include the adaptability of the modeling object corresponding to the data to the overall model topology.

[0015] Preferably, the constraint hierarchy further includes a semi-hard constraint layer located between the hard constraint layer and the soft constraint layer; the hard constraint layer consists of engineering safety and structural continuity constraints that must be met, the semi-hard constraint layer consists of general design rule constraints that can be adjusted within a finite range, and the soft constraint layer consists of empirical or historical constraints that can be modified.

[0016] Preferably, the input to the rule-constrained machine learning model further includes the object type identifier of the current modeling object and the current design stage identifier.

[0017] Preferably, the machine learning model embeds rule features in the decision layer. These rule features are derived from a predefined engineering rule library, which contains quantized rule vectors that associate confidence vectors, object types, and constraint levels.

[0018] Preferably, the constraint transfer coefficient is determined based on a weighted sum of a first contribution coefficient and a second contribution coefficient, wherein the first contribution coefficient is determined by the multidimensional confidence vector and the second contribution coefficient is determined by the effective constraint level; wherein the second contribution coefficient corresponding to the effective constraint level being a hard constraint level is higher than the second contribution coefficient corresponding to the effective constraint level being a soft constraint level.

[0019] The constraint transfer coefficients are also determined based on the third contribution coefficient and the fourth contribution coefficient. The third contribution coefficient is determined by the object type of the modeling object, and the fourth contribution coefficient is determined by the current design stage. The constraint transfer coefficients are the weighted sum of the first contribution coefficient, the second contribution coefficient, the third contribution coefficient, and the fourth contribution coefficient.

[0020] Preferably, adjusting the parameters of the modeling objects specifically includes multiplying the parameter adjustment amount of the upstream modeling object by the constraint transfer coefficient according to the structural relationship between the modeling objects, and using this as the influence amount on the parameter adjustment of the downstream adjacent modeling objects.

[0021] A second aspect of this application also provides an intelligent modeling system for power survey and design based on multi-source data fusion, comprising:

[0022] The data acquisition and structuring module is used to acquire and structure multi-source survey and design data, which includes historical design data, survey and mapping data, and manually collected data.

[0023] The multidimensional credibility vector construction module is used to construct a multidimensional credibility vector for each type or group of data based on the engineering attributes of the multi-source survey and design data. The multidimensional credibility vector is used to quantitatively characterize the credibility of the data in different dimensions.

[0024] The hierarchical modeling constraint system initialization module is used to initialize a hierarchical modeling constraint system containing at least two different constraint levels based on a preset constraint level definition. The constraint levels include at least a hard constraint layer, a semi-hard constraint layer, and a soft constraint layer, and initially classify different design constraints to the corresponding constraint levels.

[0025] The constraint hierarchy dynamic scheduling module is used to take the multidimensional confidence vector as input, and through a rule-based constraint-based machine learning model, to perform hierarchical dynamic scheduling of at least some constraints in the hierarchical modeling constraint system, output the probability distribution of the at least some constraints belonging to each constraint hierarchy, and determine the constraint hierarchy corresponding to the maximum probability as the adjusted effective constraint hierarchy. The machine learning model learns the mapping relationship between the confidence vector and the constraint hierarchy adjustment based on pre-trained rules and samples.

[0026] The constraint transfer coefficient determination module is used to determine the constraint transfer coefficient associated with each constraint based on the effective constraint level and the multidimensional confidence vector. The constraint transfer coefficient is used to quantify and control the strength of the effect of the corresponding constraint when it is transferred between adjacent modeling objects in the model structure.

[0027] The design model generation module is used to adjust the parameters and update the model of the modeling object under the joint control of the effective constraint level and the constraint transfer coefficient, so as to generate a power survey and design model.

[0028] This application effectively solves the technical problems in existing power survey and design modeling, such as static data credibility processing, fixed constraint forms, reliance on manual conflict handling, and chain-like error amplification, through the synergistic effect of multidimensional credibility vector construction, hierarchical modeling constraint system, rule-based constraint machine learning model, and constraint propagation coefficient adjustment. The multidimensional credibility vector enables accurate characterization of data credibility status, making credibility differences across different dimensions calculable; the hierarchical constraint system and dynamic scheduling mechanism ensure that constraint effectiveness matches data credibility, avoiding low-credibility data triggering strong constraints; the rule-based constraint machine learning model achieves intelligent and automated constraint scheduling, improving modeling efficiency and accuracy; and the constraint propagation coefficient enables proactive error attenuation, suppressing chain-like amplification effects. The final output model meets the requirements of structural continuity, safety and reliability, and engineering feasibility, significantly improving the efficiency and quality of power survey and design. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0030] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0031] Figure 1A schematic diagram of an intelligent modeling method for power survey and design based on multi-source data fusion provided in an embodiment of this application;

[0032] Figure 2 A schematic diagram illustrating the construction process of the multidimensional credibility vector provided in the embodiments of this application;

[0033] Figure 3 A schematic diagram of the initialization process of the hierarchical modeling constraint system provided in the embodiments of this application;

[0034] Figure 4 This is a schematic diagram of the constraint hierarchy dynamic scheduling process provided in the embodiments of this application;

[0035] Figure 5 This is a schematic diagram of the deep learning model structure provided in the embodiments of this application;

[0036] Figure 6 A schematic diagram illustrating the process for determining constraint transfer coefficients provided in an embodiment of this application;

[0037] Figure 7 This is a schematic diagram of the structure of an intelligent modeling system for power survey and design based on multi-source data fusion, provided in an embodiment of this application. Detailed Implementation

[0038] 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 some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0039] It should be noted that all user information (including but not limited to user device information, user personal information, object information corresponding to device usage data, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, device usage data, etc.) involved in all embodiments of this application are information and data authorized by the user or fully authorized by all parties.

[0040] This method is applicable to the construction, renovation, or relocation of transmission and distribution lines, and is particularly suitable for power engineering design scenarios that require the reuse of historical design results and existing line models, as well as projects where multiple batches of survey data and multiple versions of design results are used concurrently. This method can cover the modeling needs of the entire design phase, including feasibility studies, preliminary design, and construction drawing design. The implementing entity can be a power survey and design modeling system deployed on a server or the design unit's internal computing platform. This system can interface with existing power design software or data management systems, meeting the operational environment requirements of conventional computer-aided design and ensuring that the modeling process adapts to the actual engineering situation where the reliability of multi-source data varies.

[0041] The following detailed description, in conjunction with specific embodiments, illustrates the implementation process of the intelligent modeling method for power survey and design based on multi-source data fusion described in this application. It should be noted that these embodiments are merely for explaining this application and not for limiting its scope of protection. Any conventional adjustments or substitutions made by those skilled in the art to the steps without departing from the concept of this application should be included within the scope of protection of this application.

[0042] like Figure 1 As shown in the figure, this application discloses a schematic diagram of an intelligent modeling method for power survey and design based on multi-source data fusion, including the following method steps:

[0043] S1. Acquire and structure multi-source survey and design data, including historical design data, survey and mapping data, and manually collected data;

[0044] S2. Based on the engineering attributes of the multi-source survey and design data, construct a multi-dimensional credibility vector for each type or group of data. The multi-dimensional credibility vector is used to quantitatively characterize the credibility of the data in different dimensions.

[0045] S3. Based on the preset constraint level definition, initialize a hierarchical modeling constraint system containing at least two different constraint levels. The constraint level includes at least a hard constraint layer and a soft constraint layer, and initially classify different design constraints to the corresponding constraint levels.

[0046] S4. Using the multidimensional confidence vector as input, a rule-based machine learning model is used to dynamically schedule at least some constraints in the hierarchical modeling constraint system, outputting the probability distribution of the at least some constraints belonging to each constraint level, and determining the constraint level corresponding to the maximum probability as the adjusted effective constraint level. The machine learning model learns the mapping relationship between the confidence vector and the constraint level adjustment based on pre-trained rules and samples.

[0047] S5. Based on the effective constraint hierarchy and the multidimensional confidence vector, determine the constraint transfer coefficient associated with each constraint. The constraint transfer coefficient is used to quantify and control the strength of the corresponding constraint when it is transferred between adjacent modeling objects in the model structure.

[0048] S6. Under the joint control of the effective constraint level and the constraint transfer coefficient, the parameters of the modeling object are adjusted and the model is updated to generate a power survey and design model.

[0049] In some embodiments, step S1 involves the acquisition and structured processing of multi-source survey and design data. This step provides a unified data input for subsequent modeling processes, and its core is to achieve standardized integration of multi-source data.

[0050] In practice, historical design data for the target route or engineering area can be extracted from the project's data management system, including historical design drawings, route parameter tables, and tower layout records. Topographic and feature data, such as elevation data, contour data, and feature outline data, can be obtained from surveying and mapping results. Simultaneously, manually collected key point information and auxiliary explanatory data are also collected. For the aforementioned multi-source data, format and coordinate benchmark unification processes are implemented. Format unification involves converting heterogeneous data such as CAD, GIS, and tabular formats into a system-compatible standardized format using a data conversion interface. Coordinate benchmark unification is based on the project's pre-set national Cartesian coordinate system, calibrating and correcting the coordinates of various data types to eliminate coordinate deviations from different data sources. In the final structured data set, each data item is associated with a clear data type and source identifier, ensuring accurate identification of data attributes in subsequent steps and providing fundamental data support for credibility assessment and constraint scheduling.

[0051] Before initializing the hierarchical modeling constraint system, the modeling objects and their relationships should be clearly defined to provide a structural foundation for constraint layering and transfer. In practice, based on engineering design requirements, pole / tower objects, conductor objects, foundation objects, and auxiliary objects are identified from a structured dataset using pre-defined object recognition rules. Each object is associated with a corresponding data source and key parameters. Based on the characteristics of the power line topology, relationships between objects are constructed, including the connection points between poles and conductors, the spatial relationship between conductors and terrain, the support relationship between foundations and poles, and the spacing between adjacent poles. Finally, a graph-structured data format is used to record these objects and relationships, where nodes represent modeling objects, edges represent relationships, and edge attributes record the relationship type. This ensures that subsequent steps can quickly invoke these relationships, achieving precise control over constraint transfer.

[0052] In some embodiments, step S2 aims to address the technical problem in the prior art that data credibility can only be described by static weights and cannot reflect the actual credibility status of the project from multiple dimensions. The principle is to extract multi-dimensional engineering attributes from the data and construct a structured credibility vector, thereby achieving a dynamic and accurate characterization of data credibility. This provides a quantitative basis for subsequent constraint scheduling, accurately distinguishing the credibility status of different data in dimensions such as time, space, specifications, and structure, and avoiding the problem of unreasonable constraint effectiveness due to inaccurate credibility descriptions.

[0053] Please see Figure 2 , Figure 2 This is a schematic diagram illustrating the construction process of the multidimensional credibility vector provided in this application embodiment. In specific implementation, this step can be broken down into feature extraction, feature quantization, and credibility vector combination, as follows:

[0054] In S201, engineering attribute feature extraction involves extracting features from at least two of the following dimensions—time, space, specification, and structure—based on the engineering attributes of each data category or group within the structured dataset, thus avoiding interference from subjective evaluations.

[0055] Among them, the engineering attribute characteristics of the time dimension mainly include the data collection timestamp, the historical design stage corresponding to the data, and the interval between the current design stage and the data collection stage. The core feature is the time difference between the data collection time and the current design stage. For example, if some historical design data was collected several years ago, corresponding to the preliminary design stage, and the current stage is the construction drawing design stage of the same line renovation project, its core feature of the time dimension is the time interval between the two.

[0056] The spatial dimension of engineering attributes includes the latitude and longitude spatial range of the data, the sampling density within the local space, the degree of deviation from the same source data within the same spatial range, and the degree of matching with different source data. The core feature is the consistency of the data with other data within a local area. For example, a batch of terrain elevation data covers a specific length of mountainous section, with a sampling density of one sampling point every several meters. The elevation deviation from adjacent batches of data in the same area has a maximum value and an average value. These are all core features of the spatial dimension.

[0057] The engineering attribute characteristics of the specification dimension include the design specification version followed during data collection, the specification version adopted in the current design, the specification basis for key data parameters, and the differences in requirements for these parameters between the old and new specifications. The core characteristic is the difference between the design specification version followed by the data and the current design specification. For example, if the safety distance parameter of some historical data is based on an old specification, while the current specification adopts a new version with higher requirements, the difference in specification version and the difference in parameter requirements are the core characteristics of the specification dimension.

[0058] The structural dimension of engineering attributes includes the topological connections between the modeling object corresponding to the data and other objects, the adaptability of object parameters to the overall model topology, and the coherence between local structures and the overall engineering structure. The core feature is the adaptability of the modeling object corresponding to the data to the overall model topology. For example, the consistency between the coordinate data of a certain tower and the overall route topology, and the rationality of the spacing between adjacent towers, are core features of the structural dimension.

[0059] In S202, feature quantization is performed. The aforementioned non-numerical engineering attribute features are transformed into computable numerical features. The feature quantization methods for each dimension are as follows:

[0060] Time dimension feature quantization is calculated using a time decay coefficient, the formula of which is:

[0061]

[0062] Where T is the time dimension quantification value, with a value range of [0,1]; Δt is the time interval between data acquisition and the current design stage; k is the time decay coefficient, which can be configured to a suitable value for transmission line projects and another suitable value for distribution lines; St is the stage matching coefficient, which is 1.0 in the same stage. The data from the preliminary design stage is applied to the construction drawing design stage, which takes a suitable value. The data from the feasibility study stage is applied to the construction drawing design stage, which takes another suitable value. The data from the subsequent design stage is applied to the early design stage, which takes yet another suitable value.

[0063] Spatial dimensional feature quantification is calculated using the spatial consistency coefficient, with the following formula:

[0064]

[0065] Where S is the spatial dimension quantization value, with a value range of [0,1]; α is the weight coefficient, which can be configured to 0.5; This represents the number of effective sampling points in the local space. The number of sampling points is determined according to the survey specifications and is the theoretically necessary number. This represents the average deviation of this data from other data in the same region. This is the maximum allowable deviation threshold for the project.

[0066] Canonical dimensional feature quantization is calculated using the canonical fit coefficient, with the following formula:

[0067]

[0068] Where C is the quantified value of the specification dimension, with a value range of [0,1]; n is the number of key specification parameters; The weight of the i-th parameter is: the safety distance parameter can be configured with a suitable weight, the load-bearing limit parameter can be configured with another suitable weight, and the insulation class parameter can be configured with yet another suitable weight. This represents the allowed value for the i-th parameter according to the current design specifications. The allowed value for the i-th parameter according to the old specifications followed during data acquisition.

[0069] Structural dimension feature quantization is calculated using topological matching coefficients, as shown in the formula:

[0070]

[0071] Where Str is the quantized value of the structural dimension, with a value range of [0,1]; β is the weight coefficient, which can be configured to 0.6; This represents the number of topological matching connections between the modeling object corresponding to the data and the overall model. ΔL represents the theoretical number of connections between the modeling object and the overall model; ΔL represents the length deviation between the local structure and the overall engineering structure. This is the maximum allowable length deviation threshold for this type of structure.

[0072] In S203, the credibility vector is combined. The quantized values ​​of the above dimensions are combined in a preset order to form a multi-dimensional credibility vector. If features of four dimensions—time, space, specification, and structure—are extracted, the credibility vector is a four-dimensional vector V=[T,S,C,Str], where T corresponds to time credibility, S to space credibility, C to specification credibility, and Str to structure credibility. If only two or three dimensions of features are extracted, the quantized values ​​of the corresponding dimensions are combined to form a two-dimensional or three-dimensional credibility vector. The values ​​of this vector are all between [0,1], and the closer the value is to 1, the higher the credibility of that dimension. For example, if the quantized value of the time dimension of a certain data is 0.4, the quantized value of the space dimension is 0.9, the quantized value of the specification dimension is 0.3, and the quantized value of the structure dimension is 0.8, then its credibility vector is [0.4,0.9,0.3,0.8]. This indicates that the time and specification credibility of the data are low, while the space and structure credibility are high. When modeling such data, spatial constraints can be prioritized, and specification constraints can be weakened to accurately adapt to the actual engineering conditions.

[0073] In some embodiments, step S3 aims to address the technical problem in the prior art where the constraint form is fixed and the level and intensity cannot be adjusted according to data credibility. The principle is to predefine a hierarchical constraint system, dividing design constraints into different levels according to engineering importance and flexibility, providing a framework for subsequent dynamic scheduling. The constraint system is flexible and adaptable, ensuring that constraints corresponding to high-credibility data take priority and constraints corresponding to low-credibility data are appropriately weakened.

[0074] Please see Figure 3 , Figure 3This is a schematic diagram of the initialization process of the hierarchical modeling constraint system provided in an embodiment of this application. In specific implementation, this step can be broken down into constraint hierarchy definition, initial constraint classification, and constraint-object association, as follows:

[0075] In S301, constraint hierarchy is defined. Based on power engineering design specifications and engineering practice experience, at least two different constraint levels can be preset, including a hard constraint level and a soft constraint level. Preferably, a semi-hard constraint level located between the hard constraint level and the soft constraint level can also be set. The definition standards, engineering significance and constraint types of each constraint level are clearly defined to ensure that the hierarchy division conforms to the power engineering design logic.

[0076] The hard constraint layer consists of mandatory engineering safety and structural continuity constraints that cannot be violated; breaching these constraints will result in safety hazards or structural discontinuities. The hard constraint layer includes safety distance constraints, such as the minimum safe distance between conductors and the ground or buildings; structural continuity constraints, such as smooth transitions in conductor sag and foundation bearing capacity matching tower loads; and mandatory code constraints, such as the minimum cross-sectional area of ​​conductors and the minimum load-bearing limit of towers. All of these are determined based on mandatory provisions and core safety requirements of design codes.

[0077] The semi-rigid constraint layer comprises general design rules that allow adjustments within a limited range. Exceeding this adjustment range may affect the project's economic efficiency or rationality, but it does not directly lead to safety hazards. The semi-rigid constraint layer includes tower spacing constraints, such as the conventional spacing range for lines of a specific voltage level on flat terrain; conductor type selection constraints, such as the range of conductor types suitable for the voltage level; and conventional foundation size constraints, such as the foundation depth and bottom surface dimensions corresponding to a certain type of tower, which are determined based on recommended provisions in the specifications and conventional design rules.

[0078] The soft constraint layer consists of empirical or historical constraints that can be corrected by highly reliable data and are intended only as a design reference. The soft constraint layer includes historical engineering parameter constraints, such as commonly used tower types and foundation forms in a particular region; design experience constraints, such as empirical values ​​for sag adjustment in similar projects; and non-mandatory reference standard constraints, such as parameter ranges in industry-recommended technical guidelines, determined based on historical experience and non-mandatory standards.

[0079] In S302, initial constraint classification is performed. All constraints in power engineering design are initially classified according to the aforementioned constraint hierarchy definition standards. First, all constraints from design specifications, engineering manuals, and historical experience are comprehensively collected, clarifying the specific content, applicable scenarios, and importance level of each constraint. Second, constraints are classified into corresponding constraint levels according to their safety importance, mandatory nature, and adjustment flexibility, in accordance with the hierarchy definitions. Finally, a constraint hierarchy list is established, clarifying the constraint name, content, parameters, and adjustment rules for each level. For example, constraints requiring a minimum safe distance between conductors and the ground of a certain value are classified as hard constraints; constraints requiring tower spacing within a certain range on flat terrain for specific voltage levels are classified as semi-hard constraints; and constraints requiring a specific type of tower commonly used in a certain area are classified as soft constraints. For constraints with ambiguous boundaries and difficult to classify directly, an importance scoring threshold is used: those with a score of a certain value are hard constraints, those with a score within a certain range are semi-hard constraints, and those with a score below a certain value are soft constraints, ensuring the consistency and rationality of the classification results.

[0080] In S303, constraints are associated with objects. Establishing the association between constraints and modeled objects clarifies the applicable objects for each type of constraint, providing a basis for subsequent precise scheduling. The association is determined based on the constraint's scope and object attributes. For example, safety distance constraints are primarily associated with conductors and tower objects, tower spacing constraints are only associated with tower objects, foundation size constraints are associated with both foundation and tower objects, and historical engineering parameter constraints can be associated with tower and foundation objects, etc. These associations are recorded in a structured table format, containing fields such as constraint identifier, constraint name, constraint level, applicable object type, and associated parameter name. This ensures that when the reliability of the data source corresponding to an object changes, the type and level of its associated constraints can be quickly queried, providing data support for dynamic scheduling based on constraint levels.

[0081] In some embodiments, step S4 aims to address the technical problems in the prior art where multi-source data conflicts rely on manual or simple overwrite processing and lack intelligent constraint scheduling. The principle is to construct a hybrid model combining rule constraints and machine learning, using a confidence vector as the core input, learning constraint hierarchy adjustment rules to achieve automatic and precise scheduling of constraint levels, efficiently handling constraint conflicts, ensuring that high-confidence data constraints take precedence, and reasonably weakening low-confidence data constraints, thus avoiding systematic errors.

[0082] Please see Figure 4 , Figure 4 This is a schematic diagram of the constraint hierarchy dynamic scheduling process provided in an embodiment of this application. In specific implementation, this step can be broken down into model structure design, training data preparation, model training, and constraint scheduling execution, as follows:

[0083] In S401, the machine model structure is designed. This model is a rule-guided decision-making model that does not generate design results but outputs the effective hierarchy of constraints. It is constructed using a deep learning framework, such as... Figure 5 As shown, Figure 5 The schematic diagram of the deep learning model structure provided in the embodiments of this application is divided into five parts: input layer, feature processing layer, rule embedding layer, core decision layer, and output layer.

[0084] The input layer receives three types of core features, with a total input dimension of 11 dimensions. The first type of feature is a multi-dimensional credibility vector. If it is a four-dimensional credibility vector, it includes four dimensions: time credibility T, spatial credibility S, specification credibility C, and structural credibility Str. The second type of feature is the object type identifier of the current modeling object, which is represented by one-hot encoding. The modeling objects include four types: towers, conductors, foundations, and auxiliary objects, with corresponding codes [1,0,0,0], [0,1,0,0], [0,0,1,0], and [0,0,0,1]. The third type of feature is the identifier of the current design stage, which is represented by one-hot encoding. The design stages include three types: feasibility study, preliminary design, and construction drawing design, with corresponding codes [1,0,0], [0,1,0], and [0,0,1]. The input vector expression is X=[T,S,C,Str,O1,O2,O3,O4,P1,P2,P3], where O1-O4 are one-hot encoded components of the object type, and P1-P3 are one-hot encoded components of the design phase.

[0085] The feature processing layer includes a batch normalization sublayer and a feature enhancement sublayer. The batch normalization sublayer is used to eliminate dimensional differences between different features, and the formula is:

[0086]

[0087]

[0088] in, Here, x represents the normalized intermediate variable, and x represents the original value of the input feature. The mean of the batch training data. The variance of the batch training data. To prevent the use of tiny constants with zero denominators, the value is set to 1e-5. γ is the scaling factor, with an initial value of 1. β is the offset factor, with an initial value of 0. y is the output feature after batch normalization.

[0089] The feature enhancement sublayer adopts a fully connected layer structure with 64 neurons. The ReLU activation function is used to enhance the expressive power of the features. The calculation formula is as follows:

[0090]

[0091] Where F is the output vector of the feature enhancement sublayer, with a dimension of 64; This is the weight matrix of the fully connected layer, with dimensions 11×64; The output vector of the batch normalized sublayer; This is the bias vector of the fully connected layer, with a dimension of 64.

[0092] The rule embedding layer is used to transform engineering rules into computable features and integrate them into the model's decision-making process. First, a predefined engineering rule base is constructed. This base contains quantized rule vectors that correlate confidence vectors, object types, and constraint levels. For example, the rule "geometric constraints with high spatial confidence can be upgraded to hard constraints" is quantized as a rule vector R=[0,1,0,0,0.8,…], where each component corresponds to the association weight of the input feature. For each input data point, its matching degree with each rule in the engineering rule base is calculated using a dot product operation, with the formula:

[0093]

[0094] in, Let F be the matching degree between the input data and the i-th rule, and let F be the output vector of the feature enhancement sublayer. Let be the quantization vector of the i-th rule, with the superscript T denoteing the vector transpose. Then, the matching scores of all rules are weighted and summed to obtain the rule embedding features, as shown in the formula:

[0095]

[0096] in, is the rule embedding feature, with a dimension of 1; m is the number of rules in the engineering rule base; The weight of the i-th rule is set according to its importance. Finally, the rule embedding feature is concatenated with the output vector of the feature enhancement sublayer to form the output vector of the rule embedding layer. =[F, [, with 65 dimensions.]

[0097] The core decision layer consists of two fully connected layers and one dropout layer, used to implement complex feature mapping and decision logic. The first fully connected layer has 128 neurons, and the ReLU activation function is used, calculated as follows:

[0098]

[0099] in, This is the output vector of the first fully connected layer, with a dimension of 128. This is the weight matrix of the first fully connected layer, with a dimension of 65×128; This is the output vector of the rule embedding layer; This is the bias vector of the first fully connected layer, with a dimension of 128.

[0100] The dropout layer is set with a dropout rate of 0.2, which only takes effect during the model training phase. It is used to prevent the model from overfitting by randomly deactivating some neurons, thereby improving the model's generalization ability.

[0101] The second fully connected layer has 64 neurons, and the ReLU activation function is used. The calculation formula is as follows:

[0102]

[0103] in, This is the output vector of the second fully connected layer, with a dimension of 64. This is the weight matrix of the second fully connected layer, with a dimension of 128×64; This is the output vector of the dropout layer; This is the bias vector of the second fully connected layer, with a dimension of 64.

[0104] The output layer employs a multi-classification structure with three neurons. The activation function chosen is the Softmax function, which is used to calculate the probability distribution of each constraint belonging to its respective constraint level. The formula is as follows:

[0105]

[0106] Where P is the output vector of the output layer, with a dimension of 3, P=[p1,p2,p3], and p1, p2, and p3 represent the probabilities that the constraint belongs to the hard constraint layer, the semi-hard constraint layer, and the soft constraint layer, respectively. This is the weight matrix of the output layer, with dimensions 64×3; This is the output vector of the second fully connected layer; This is the bias vector of the output layer, with a dimension of 3.

[0107] In S402, training data preparation. The training dataset must meet the requirements of realism and coverage, and contain multiple samples to ensure that the model can learn the constraint level adjustment rules under different scenarios.

[0108] The training data is collected from two main sources: first, historical engineering data, which is obtained by retrospectively analyzing completed power survey and design projects and extracting multi-source data credibility vectors, modeling object types, design stages, initial constraint levels, and final effective levels to form historical sample data; second, rule-generated data, which is generated by professional designers based on engineering rules to address scenarios where credibility vector combinations and object type-design stage pairings are not covered in the historical engineering data. Historical engineering data accounts for no less than 70% of the sample data to ensure the authenticity and engineering adaptability of the training data.

[0109] The annotation of training data adopts a combination of rules and expert experience, and a clear annotation rule is formulated: For each sample, the comprehensive score of the confidence vector is first calculated, and the comprehensive score formula is Score=0.3T+0.3S+0.2C+0.2Str; If the comprehensive score is not lower than 0.7 and the initial level of the corresponding constraint is a semi-hard constraint, then the annotation is upgraded by one level, that is, the semi-hard constraint is upgraded to a hard constraint; If the comprehensive score is not higher than 0.3 and the initial level of the constraint is a semi-hard constraint, then the annotation is downgraded by one level, that is, the semi-hard constraint is downgraded to a soft constraint; If the comprehensive score is between 0.3 and 0.7 and the initial level of the constraint is a soft constraint, then the annotation is maintained at the original level; Hard constraints are only marked as invalid in extreme cases, that is, when the confidence vector of the data is not higher than 0.1 in all dimensions and is verified by experts as erroneous data, otherwise the annotation is maintained as a hard constraint.

[0110] Preprocessing of training data includes three stages: data cleaning, normalization, and dataset partitioning. The data cleaning stage removes outlier samples, such as those with a confidence vector dimension exceeding the [0,1] range, or samples with incorrect object type or design phase labeling. The normalization stage uses the min-max normalization method to normalize all input features to the [0,1] interval, as shown in the formula:

[0111]

[0112] in, Here, x represents the normalized eigenvalues, and x represents the original eigenvalues. The minimum value of this feature in the dataset. This represents the maximum value of the feature in the dataset. In the dataset partitioning stage, the preprocessed dataset is divided into training, validation, and test sets in a ratio of 7:2:1 to ensure a balanced distribution of samples of different constraint types, object types, and design stages in each set, thereby avoiding overfitting during model training.

[0113] In S403, machine learning model training takes place. Model training can be implemented using deep learning frameworks such as PyTorch or TensorFlow, and the specific process is as follows:

[0114] The model parameters are initialized using the He normal initialization method, which is suitable for the ReLU activation function and ensures that the variance of the outputs of each layer remains consistent, thus accelerating model convergence. The weight matrices W1, W2, W3, and W4 are all initialized using the He normal initialization method, and the bias vectors b1, b2, b3, and b4 are all initialized to zero. The hyperparameters for model training are set; for example, the initial learning rate can be configured to 0.001, the batch size to 32, and the maximum number of iterations to 200.

[0115] The Adam optimizer was chosen for model training. This optimizer combines the advantages of momentum gradient descent and adaptive learning rate, effectively improving training efficiency. The momentum parameters β1 and β2 of the Adam optimizer were set to 0.9 and 0.999, respectively, and the weight decay coefficient λ was set to 1e-5 to balance the model's fitting ability and generalization ability, preventing overfitting.

[0116] The cross-entropy loss function is chosen for model training. This function is suitable for multi-class classification problems and can effectively measure the difference between the model's output probability distribution and the true labels. The formula is:

[0117]

[0118] Where Loss is the loss value. The true hierarchical one-hot encoding for annotation is as follows: hard constraints correspond to [1,0,0], semi-hard constraints correspond to [0,1,0], and soft constraints correspond to [0,0,1]. This represents the probability of the corresponding level output by the model.

[0119] The model training process can be performed in the following steps: First, the training set is input into the model in batches, and the output probability and loss value of the model are calculated through forward propagation; then, the gradient of the parameters of each layer is calculated through the backpropagation algorithm, and a gradient pruning method is adopted, with a pruning threshold of 1.0 to prevent gradient explosion; next, the Adam optimizer updates the model's weight parameters and bias parameters according to the gradient to minimize the loss value; every 10 epochs, the model's accuracy, precision, recall and other performance indicators are calculated on the validation set. If the accuracy of the validation set does not improve for 20 consecutive epochs, an early stopping strategy is triggered to save the current optimal model parameters and avoid overtraining; after training, the model performance is comprehensively evaluated on the test set, requiring the model accuracy to be no less than 90% and the recall rate of key constraints (related to hard constraints) to be no less than 95%, ensuring that the model can run reliably in real-world engineering scenarios.

[0120] In S404, constraint scheduling is executed. After model training is complete, dynamic constraint scheduling is performed based on the input confidence vector, modeling object type, and design phase. The specific process is as follows:

[0121] The input data is preprocessed by combining the constructed multidimensional confidence vector, the identified modeling object type one-hot encoding, and the current design stage one-hot encoding into an 11-dimensional input vector according to the model input format. The input vector is then normalized to the [0,1] interval using the same min-max normalization method as the training data before being input into the trained model.

[0122] During the model inference phase, the model outputs probability distributions of constraints belonging to hard constraint layers, semi-hard constraint layers, and soft constraint layers. The layer corresponding to the maximum probability is selected as the effective layer of the constraint. If the difference between the maximum probability and the second-highest probability is less than 0.1, it indicates that the model decision has fuzziness. In this case, a secondary judgment needs to be made in conjunction with the engineering rule base. For example, when the spatial confidence S is not less than 0.8, it is preferentially determined to be a higher-level constraint to ensure the uniqueness of the decision of the effective constraint layer.

[0123] During the constraint level adjustment phase, the initial constraint level is adjusted based on the effective level results. For example, if the initial level is a semi-hard constraint, the level is increased if the effective level is a hard constraint, and decreased if the effective level is a soft constraint. If the initial level is a soft constraint, the level is increased if the effective level is a semi-hard constraint, and increased only if the effective level is a hard constraint, provided that the specification credibility C is not lower than 0.7. If the initial level is a hard constraint, it is marked as temporarily invalid only if the effective level is invalid and the comprehensive score of the data credibility vector is not higher than 0.1, and manual verification is required; otherwise, the hard constraint level remains unchanged.

[0124] When multiple constraints target the same modeling object and conflict after adjustment, conflict resolution is achieved using hierarchy priority and credibility weight. Specifically, hard constraint hierarchy has higher priority than semi-hard constraint hierarchy, and semi-hard constraint hierarchy has higher priority than soft constraint hierarchy. If the constraints conflict at the same hierarchy, the comprehensive score of the data source credibility vector corresponding to each constraint is calculated. The constraint with the higher comprehensive score takes effect first, and the constraint with the lower comprehensive score has its effectiveness weakened according to the ratio of the two scores. For example, if the score ratio of two constraints at the same hierarchy is 3:2, the effectiveness of the constraint with the lower score is reduced by 40%, ensuring that conflict handling conforms to engineering logic.

[0125] In some embodiments, step S5 aims to address the technical problem of chain-like amplification of data errors along the model structure in the prior art. The principle is to introduce constraint propagation coefficients to quantify the degree of constraint propagation attenuation in the model structure, and dynamically adjust these coefficients in conjunction with factors such as data reliability and constraint level. This achieves active attenuation of low-reliability data errors and stable propagation of high-reliability data constraints, suppressing the spread of local errors and ensuring the overall rationality of the model structure.

[0126] Please see Figure 6 , Figure 6 This is a schematic diagram illustrating the process of determining constraint transfer coefficients according to an embodiment of this application. In specific implementation, this step can be broken down into defining transfer coefficients, calculating contribution coefficients, and synthesizing transfer coefficients, as follows:

[0127] In S601, the constraint propagation coefficient is defined as follows: The constraint propagation coefficient is a quantitative indicator describing the degree of attenuation or amplification of the constraint's effect as it propagates to adjacent modeling objects in the model structure. Its value ranges from [0, 1.2]. A propagation coefficient of 0 indicates that the constraint is not propagated at all; a coefficient between 0 and 1 indicates that the constraint's effect is attenuated during propagation; a coefficient of 1 indicates that the constraint is propagated without attenuation; and a coefficient between 1 and 1.2 indicates that the constraint's effect is moderately amplified during propagation. This amplification range applies only to high-confidence critical constraints to ensure that their constraint effect is fully propagated in the structural chain.

[0128] In S602, contribution coefficients are calculated. The constraint transfer coefficients are determined based on the weighted sum of the first, second, third, and fourth contribution coefficients. The calculation method for each contribution coefficient is as follows:

[0129] The first contribution coefficient is determined by the multidimensional credibility vector, denoted as... . The confidence vector comprehensive score is calculated using the formula: Score = 0.3T + 0.3S + 0.2C + 0.2Str.

[0130]

[0131] The second contribution coefficient is determined by the effective constraint level, denoted as... . The rule for determining the value is that when the effective constraint level is a hard constraint level, =1.2; When the effective constraint level is a semi-hard constraint level, =0.8; When the effective constraint level is a soft constraint level, =0.3; When the effective constraint level is temporary failure, =0.05. This coefficient reflects the importance of the constraint hierarchy. Hard constraints, as the core of safety, have the highest transmission coefficient, ensuring that their constraint effect is fully transmitted in the structural chain.

[0132] The third contribution coefficient is determined by the object type of the modeling object, and is denoted as... . The rule for determining the value is: when the modeling object is a pole or tower object, =1.0; when it is a wire object, =0.9; when using a base object, =0.8; when it is an auxiliary object, =0.5. As the core of the transmission line structure, the tower has a higher constraint transfer coefficient than other objects to ensure the constraint consistency of the core structure.

[0133] The fourth contribution factor is determined at the current design stage and is denoted as... . The rule for determining the value is: when the current design stage is construction drawing design, =1.0; for the preliminary design, =0.8; for feasibility studies, =0.6. The construction drawing design stage has the highest requirements for modeling accuracy, with higher constraint transfer coefficients to ensure the accuracy of model details.

[0134] In S603, the transfer coefficients are synthesized. The formula for synthesizing the constraint transfer coefficients is:

[0135]

[0136] in, , , , The weighting coefficients for each contribution coefficient satisfy the following conditions: + + + =1, which can be configured based on engineering practice experience.

[0137] In some embodiments, step S6 involves adjusting the parameters of the modeling object and updating the model under the joint control of the effective constraint level and constraint transfer coefficient, ensuring that the model meets engineering requirements. Specifically, firstly, based on the modeling object and its relationships, the basic structure of the model is initialized, including parameters such as the initial position of the tower, the initial sag of the conductor, and the initial dimensions of the foundation; then, the key parameters of the modeling object are iteratively adjusted according to the effective constraint level and constraint transfer coefficient.

[0138] During parameter adjustment, based on the structural relationships between modeling objects, the parameter adjustment amount of the upstream modeling object is multiplied by the constraint transfer coefficient, and this multiplication is used as the influence on the parameter adjustment of the downstream adjacent modeling objects. For example, in a strongly coupled structural chain of "tower location, conductor sag, clearance check, and foundation design" of a power line, the tower location is the upstream object, with a parameter adjustment amount of ΔL and a corresponding constraint transfer coefficient of K1. Then, the conductor sag, as the downstream adjacent object, has a parameter adjustment influence amount of ΔL×K1. The conductor sag parameter adjustment amount is ΔH, with a corresponding constraint transfer coefficient of K2. Then, the clearance check parameter adjustment influence amount is ΔH×K2, and so on, thus achieving quantitative control of the constraint transfer strength.

[0139] The focus of parameter adjustment differs depending on the type of modeling object. For example, for tower objects, the main adjustments are to their coordinates and height parameters to ensure that the hard constraints on safety distance and the semi-hard constraints on tower spacing are met. For conductor objects, the sag parameters are adjusted based on the tower location and transmission coefficient to ensure a smooth sag transition and that the clearance distance meets the standards. For foundation objects, the burial depth and bottom surface dimensions are adjusted according to the tower load and transmission coefficient to ensure that the load-bearing capacity is matched.

[0140] After each parameter adjustment, the model state is updated in real time, recording the specific results of the constraint effect, including the adjusted parameter values ​​and the degree of constraint satisfaction. If the adjusted model does not meet the constraint requirements of the hard constraint layer, such as insufficient safe distance between the conductor and the ground, a backtracking mechanism can be triggered. This can involve pausing the current adjustment process, returning to the constraint level dynamic scheduling step to re-adjust the constraint level, or directly adjusting the constraint transfer coefficients, recalculating the constraint state, and then adjusting the parameters again until the model meets all hard constraint requirements.

[0141] The termination condition for iterative adjustment is that the change in the modeling object parameters in two consecutive iterations is less than a preset threshold, and all hard constraints are satisfied, more than 90% of the semi-hard constraints are satisfied, and the soft constraints are reasonably adapted. At this time, the model reaches a stable state and parameter adjustment stops.

[0142] Once the model stabilizes, the final power survey and design model is output. This model contains complete structured information such as modeling object parameters, object relationships, constraint records, and data reliability vectors, and can be directly imported into existing power design software for subsequent engineering applications. Before output, a final verification is required. Verification includes checking the structural continuity between modeling objects, the fulfillment of key safety constraints, and the control effect of data reliability differences, ensuring the model has no significant engineering defects.

[0143] This method effectively solves the technical problems of static data credibility processing, fixed constraint forms, reliance on manual conflict resolution, and chain-like error amplification in existing power survey and design modeling by synergistically employing multidimensional credibility vector construction, hierarchical modeling constraint system, rule-based constraint machine learning model, and constraint propagation coefficient adjustment. The multidimensional credibility vector enables precise characterization of data credibility status, making credibility differences across different dimensions calculable; the hierarchical constraint system and dynamic scheduling mechanism ensure that constraint effectiveness matches data credibility, avoiding low-credibility data triggering strong constraints; the rule-based constraint machine learning model achieves intelligent and automated constraint scheduling, improving modeling efficiency and accuracy; and the constraint propagation coefficient enables proactive error attenuation, suppressing chain-like amplification effects. The final output model meets the requirements of structural continuity, safety and reliability, and engineering feasibility, significantly improving the efficiency and quality of power survey and design compared to existing technologies.

[0144] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0145] Please see Figure 7 , Figure 7 This application provides a structural block diagram of an intelligent modeling system for power surveying and design based on multi-source data fusion. The system specifically includes:

[0146] The data acquisition and structuring processing module 701 is used to acquire and structure multi-source survey and design data, which includes historical design data, survey and mapping data and manually collected data.

[0147] The multidimensional credibility vector construction module 702 is used to construct a multidimensional credibility vector for each type or group of data based on the engineering attributes of the multi-source survey and design data. The multidimensional credibility vector is used to quantitatively characterize the credibility of the data in different dimensions.

[0148] The hierarchical modeling constraint system initialization module 703 is used to initialize a hierarchical modeling constraint system containing at least two different constraint levels based on a preset constraint level definition. The constraint levels include at least a hard constraint layer, a semi-hard constraint layer, and a soft constraint layer, and initially classify different design constraints to the corresponding constraint levels.

[0149] The constraint hierarchy dynamic scheduling module 704 is used to take the multi-dimensional confidence vector as input, and through a rule-based constraint-based machine learning model, perform hierarchical dynamic scheduling on at least some constraints in the hierarchical modeling constraint system, output the probability distribution of the at least some constraints belonging to each constraint hierarchy, and determine the constraint hierarchy corresponding to the maximum probability as the adjusted effective constraint hierarchy. The machine learning model learns the mapping relationship between the confidence vector and the constraint hierarchy adjustment based on pre-trained rules and samples.

[0150] The constraint transfer coefficient determination module 705 is used to determine the constraint transfer coefficient associated with each constraint based on the effective constraint level and the multidimensional confidence vector. The constraint transfer coefficient is used to quantify and control the strength of the effect of the corresponding constraint when it is transferred between adjacent modeling objects in the model structure.

[0151] The design model generation module 706 is used to adjust the parameters and update the model of the modeling object under the joint control of the effective constraint level and the constraint transfer coefficient, so as to generate a power survey and design model.

[0152] It should be noted that the working process of each module in the intelligent modeling system for power survey and design based on multi-source data fusion described in this embodiment can refer to the working process of the intelligent modeling method for power survey and design based on multi-source data fusion described in the above embodiments. The technical effects achieved are also the same as those of the intelligent modeling method for power survey and design based on multi-source data fusion described in the above embodiments, and will not be repeated here.

[0153] The above description represents the preferred embodiments of the present invention. It should be noted that, for those skilled in the art, various improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A power survey and design intelligent modeling method based on multi-source data fusion, characterized in that, Includes the following steps: S1. Acquire and structure multi-source survey and design data, including historical design data, survey and mapping data, and manually collected data; S2. Based on the engineering attributes of the multi-source survey and design data, construct a multi-dimensional credibility vector for each type or group of data. The multi-dimensional credibility vector is used to quantitatively characterize the credibility of the data in different dimensions. S3. Based on the preset constraint level definition, initialize a hierarchical modeling constraint system containing at least two different constraint levels. The constraint level includes at least a hard constraint layer, a semi-hard constraint layer, and a soft constraint layer, and initially classify different design constraints to the corresponding constraint levels. S4. Using the multidimensional confidence vector as input, a rule-based machine learning model is used to dynamically schedule at least some constraints in the hierarchical modeling constraint system, outputting the probability distribution of the at least some constraints belonging to each constraint level, and determining the constraint level corresponding to the maximum probability as the adjusted effective constraint level. The machine learning model learns the mapping relationship between the confidence vector and the constraint level adjustment based on pre-trained rules and samples. S5. Based on the effective constraint hierarchy and the multidimensional confidence vector, determine the constraint transfer coefficient associated with each constraint. The constraint transfer coefficient is used to quantify and control the strength of the corresponding constraint when it is transferred between adjacent modeling objects in the model structure. S6. Under the joint control of the effective constraint level and the constraint transfer coefficient, the parameters of the modeling object are adjusted and the model is updated to generate a power survey and design model.

2. The intelligent modeling method for power survey and design based on multi-source data fusion according to claim 1, characterized in that, Constructing a multidimensional credibility vector specifically includes extracting engineering attribute features from at least two of the multi-source survey and design data in the time dimension, spatial dimension, specification dimension, and structural dimension; quantifying the engineering attribute features into numerical features; and combining them in a preset order to form the multidimensional credibility vector.

3. The intelligent modeling method for power survey and design based on multi-source data fusion according to claim 2, characterized in that, The engineering attribute characteristics in the time dimension include the time difference between the data acquisition time and the current design stage; the engineering attribute characteristics in the spatial dimension include the degree of consistency between the data and other survey or design data in the same area within a local region; the engineering attribute characteristics in the specification dimension include the difference between the design specification version followed by the data and the current design specification; and the engineering attribute characteristics in the structural dimension include the adaptability of the modeling object corresponding to the data to the overall model topology.

4. The intelligent modeling method for power survey and design based on multi-source data fusion according to claim 3, characterized in that, The constraint hierarchy also includes a semi-hard constraint layer located between the hard constraint layer and the soft constraint layer; the hard constraint layer consists of engineering safety and structural continuity constraints that must be met, the semi-hard constraint layer consists of general design rule constraints that can be adjusted within a finite range, and the soft constraint layer consists of empirical or historical constraints that can be modified.

5. The intelligent modeling method for power survey and design based on multi-source data fusion according to claim 1, characterized in that, The input to the rule-based machine learning model also includes the object type identifier of the current modeling object and the current design stage identifier.

6. The intelligent modeling method for power survey and design based on multi-source data fusion according to claim 5, characterized in that, The machine learning model embeds rule features in the decision layer. These rule features are derived from a predefined engineering rule library, which contains quantized rule vectors that associate confidence vectors, object types, and constraint levels.

7. The intelligent modeling method for power survey and design based on multi-source data fusion according to claim 1, characterized in that, The constraint transfer coefficient is determined based on the weighted sum of the first contribution coefficient and the second contribution coefficient. The first contribution coefficient is determined by the multidimensional confidence vector, and the second contribution coefficient is determined by the effective constraint level. The second contribution coefficient corresponding to the effective constraint level being a hard constraint level is higher than the second contribution coefficient corresponding to the effective constraint level being a soft constraint level.

8. The intelligent modeling method for power survey and design based on multi-source data fusion according to claim 7, characterized in that, The determination of the constraint transfer coefficient is also based on the third contribution coefficient and the fourth contribution coefficient. The third contribution coefficient is determined by the object type of the modeling object, and the fourth contribution coefficient is determined by the current design stage. The constraint transfer coefficient is a weighted sum of the first contribution coefficient, the second contribution coefficient, the third contribution coefficient, and the fourth contribution coefficient.

9. The intelligent modeling method for power survey and design based on multi-source data fusion according to claim 1, characterized in that, The parameter adjustment of the modeling objects specifically includes multiplying the parameter adjustment amount of the upstream modeling object by the constraint transfer coefficient according to the structural relationship between the modeling objects, and using this as the influence amount on the parameter adjustment of the downstream adjacent modeling objects.

10. A power survey and design intelligent modeling system based on multi-source data fusion, characterized in that, include: The data acquisition and structuring module is used to acquire and structure multi-source survey and design data, which includes historical design data, survey and mapping data, and manually collected data. The multidimensional credibility vector construction module is used to construct a multidimensional credibility vector for each type or group of data based on the engineering attributes of the multi-source survey and design data. The multidimensional credibility vector is used to quantitatively characterize the credibility of the data in different dimensions. The hierarchical modeling constraint system initialization module is used to initialize a hierarchical modeling constraint system containing at least two different constraint levels based on a preset constraint level definition. The constraint levels include at least a hard constraint layer, a semi-hard constraint layer, and a soft constraint layer, and initially classify different design constraints to the corresponding constraint levels. The constraint hierarchy dynamic scheduling module is used to take the multidimensional confidence vector as input, and through a rule-based constraint-based machine learning model, to perform hierarchical dynamic scheduling of at least some constraints in the hierarchical modeling constraint system, output the probability distribution of the at least some constraints belonging to each constraint hierarchy, and determine the constraint hierarchy corresponding to the maximum probability as the adjusted effective constraint hierarchy. The machine learning model learns the mapping relationship between the confidence vector and the constraint hierarchy adjustment based on pre-trained rules and samples. The constraint transfer coefficient determination module is used to determine the constraint transfer coefficient associated with each constraint based on the effective constraint level and the multidimensional confidence vector. The constraint transfer coefficient is used to quantify and control the strength of the effect of the corresponding constraint when it is transferred between adjacent modeling objects in the model structure. The design model generation module is used to adjust the parameters and update the model of the modeling object under the joint control of the effective constraint level and the constraint transfer coefficient, so as to generate a power survey and design model.

Images