Lightning protection design evaluation method and system based on decision tree forest, terminal and medium

Abstract

The application discloses a lightning protection design evaluation method and system based on a decision tree forest, a terminal and a medium, relates to the technical field of lightning protection design analysis, and has the technical scheme as follows: data aggregation is performed on clauses of multiple-source lightning protection technical standards to obtain a structured lightning protection clause dataset of a dynamic hierarchical classification architecture; special decision trees of various lightning protection measure types are generated, and an AHP weight engine and a conflict resolution engine are implanted in nodes of the decision trees; lightning protection design parameters to be evaluated are input into the decision tree forest model to obtain a preliminary compliance identification result; the decision tree forest model is subjected to dynamic branch adjustment processing according to environmental condition parameters, and an environment-adapted identification result is regenerated; and visual report generation processing is performed to obtain a lightning protection design evaluation report. The application solves standard dispersion through the dynamic hierarchical classification architecture, breaks through the limitation of a static model through the double-engine decision tree forest, and realizes accurate protection in a complex scene through environmental risk quantification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lightning protection design analysis, more specifically, it relates to a lightning protection design evaluation method, system, terminal and medium based on a decision tree forest. BACKGROUND

[0002] Lightning is a kind of transient large current, high voltage and strong electromagnetic radiation weather phenomenon occurring in the atmosphere, and lightning disaster is listed as "one of the ten most serious natural disasters" by relevant departments. With the improvement of facility lightning protection safety requirements, the coexistence of multi-source lightning protection technical standards leads to redundant and conflicting clauses. The existing technology relies on manual analysis of scattered standard files, and it is difficult to realize dynamic environment adaptation and conflict resolution, such as different standards requiring different parameters, and traditional evaluation methods using fixed threshold judgment, which cannot respond to complex environment combinations.

[0003] Therefore, how to design a lightning protection design evaluation method, system, terminal and medium based on a decision tree forest to overcome the above defects is a technical problem we need to solve at present. SUMMARY

[0004] To solve the problems in the prior art, the purpose of the present application is to provide a lightning protection design evaluation method, system, terminal and medium based on a decision tree forest. The present application solves the standard dispersion through a dynamic hierarchical classification architecture, breaks through the limitations of static models through a double-engine decision tree forest, and realizes precise protection in complex scenarios through environmental risk quantification.

[0005] The above technical purpose of the present application is realized by the following technical scheme:

[0006] In a first aspect, a lightning protection design evaluation method based on a decision tree forest is provided, comprising the following steps:

[0007] S1: performing data aggregation processing on multi-source lightning protection technical standard clauses to obtain a structured lightning protection clause dataset of a dynamic hierarchical classification architecture;

[0008] S2: generating a special decision tree for each lightning protection measure type according to the structured lightning protection clause dataset, and implanting an AHP weight engine and a conflict resolution engine in the decision tree nodes to obtain a decision tree forest model with weighted branches;

[0009] S3: inputting the lightning protection design parameters to be evaluated into the decision tree forest model for dynamic weight integration and conflict resolution processing to obtain a preliminary compliance identification result;

[0010] S4: performing dynamic branch adjustment processing on the decision tree forest model according to the environmental condition parameters, and regenerating an environment-adaptive identification result based on the preliminary compliance identification result;

[0011] S5: inputting the environment-adaptive identification result into a meteorological disaster risk management system integration gateway to generate a visual report, and obtaining a lightning protection design evaluation report.

[0012] Further, the dynamic hierarchical classification architecture of the structured lightning protection clause dataset comprises:

[0013] a measure-oriented layer, configured to establish a plurality of main classes according to the lightning protection measure types;

[0014] an object scenario layer, configured to establish sub-classes under each main class according to applicable objects and application scenarios;

[0015] a condition constraint layer, configured to construct a tree structure according to parameter logical relationships within the sub-classes;

[0016] wherein the root node of the tree structure is a protection target, the branch node is a condition judgment item, and the leaf node is a specific measure requirement.

[0017] Further, the data aggregation processing of the multi-source lightning protection technical standard clauses comprises:

[0018] adopting a hierarchical clustering algorithm based on Levenshtein distance and semantic similarity weighting to dynamically group atomic clause units according to similarity into the dynamic hierarchical classification architecture;

[0019] extracting a keyword TF-IDF vector and a classification path code for each atomic clause unit to generate a classification feature vector that can be recognized and dynamically matched by a machine.

[0020] Further, the decision tree nodes implanted by the AHP weight engine are root nodes and branch nodes, and the functional time sequence position of the AHP weight engine is after condition judgment and before branch selection.

[0021] wherein the input of the AHP weight engine is a technical element set, a judgment matrix, and a consistency threshold, and the output is an element weight vector and a consistency check state.

[0022] Further, the conflict resolution engine is configured to:

[0023] read the element weight vector output by the AHP weight engine and the satisfaction degree of a single parameter;

[0024] determine the conflict mark between two parameters;

[0025] retrieve the corresponding priority from the rule base according to the conflict mark to obtain a rule weight;

[0026] multiply the AHP weight in the element weight vector, the satisfaction degree, and the rule weight to obtain the priority value of the corresponding parameter.

[0027] The parameter corresponding to the clause with the greater priority value is executed.

[0028] Further, the process of the dynamic branch adjustment processing is specifically:

[0029] An environmental condition parameter including at least one of a geographic parameter, a meteorological parameter and a geological parameter is acquired;

[0030] An environmental risk value is determined according to an absolute value of a difference between a parameter value in the environmental condition parameter and a corresponding parameter threshold value, and the greater the absolute value of the difference, the greater the corresponding environmental risk value;

[0031] A maximum value in all the environmental risk values is selected as an execution risk value;

[0032] When the execution risk value is greater than a first threshold value: an alternative branch library matching a current environmental parameter is retrieved from the decision tree forest model; and an optimal branch is selected by using a satisfaction degree weighted voting mechanism to replace a corresponding branch of the original decision tree;

[0033] When the execution risk value is greater than a second threshold value and less than or equal to the first threshold value: an AHP weight of a corresponding node is proportionally increased according to the execution risk value.

[0034] Further, the AHP weight of the corresponding node is proportionally increased according to the execution risk value, specifically:

[0035] An increased AHP weight increment is calculated by multiplying an industry risk coefficient and the execution risk value;

[0036] If a sum of the AHP weight increment and an original AHP weight is less than 1, the sum is taken as an updated AHP weight;

[0037] If the updated AHP weight is equal to 1, the AHP weight is re-normalized.

[0038] In a second aspect, a lightning protection design evaluation system based on a decision tree forest is provided, comprising:

[0039] A data processing module is configured to perform data aggregation processing on multiple-source lightning protection technical standard clauses to obtain a structured lightning protection clause dataset of a dynamic hierarchical classification architecture;

[0040] A model construction module is configured to generate a special decision tree for each lightning protection measure type according to the structured lightning protection clause dataset, and implant an AHP weight engine and a conflict resolution engine in a decision tree node to obtain a decision tree forest model with weighted branches;

[0041] A preliminary identification module is configured to input the to-be-evaluated lightning protection design parameters into the decision tree forest model for dynamic weight integration and conflict resolution processing, and obtain a preliminary compliance identification result;

[0042] An optimization identification module is configured to perform dynamic branch adjustment processing on the decision tree forest model according to environmental condition parameters, and regenerate an environment-adapted identification result based on the preliminary compliance identification result;

[0043] A report generation module is configured to input the environment-adapted identification result into a meteorological disaster risk management system integration gateway for visual report generation processing, and obtain a lightning protection design evaluation report.

[0044] In a third aspect, a computer terminal is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the decision tree forest-based lightning protection design evaluation method according to any one of the first aspect when executing the computer program.

[0045] In a fourth aspect, a computer readable medium is provided, which stores a computer program, and the computer program is executable by a processor to implement the decision tree forest-based lightning protection design evaluation method according to any one of the first aspect.

[0046] Compared with the prior art, the present application has the following beneficial effects:

[0047] 1. The decision tree forest-based lightning protection design evaluation method provided by the present application solves the standard dispersion through a dynamic hierarchical classification architecture, breaks through the limitations of static models through a dual-engine decision tree forest, and realizes precise protection in complex scenarios through environmental risk quantification, thereby comprehensively overcoming the bottlenecks of the prior art, achieving a 96.3% identification accuracy rate through verification of 26 engineering projects, and improving the comprehensive efficiency by 48 hours, thereby providing the first solution of "multi-source integration-dynamic adaptation-intelligent adjudication" for the field of lightning protection design evaluation.

[0048] 2. The present application constructs a three-layer tree structure composed of a measure-oriented layer, an object scene layer and a condition constraint layer, forms a machine-analyzable hierarchical logic chain with a root node (protection target), a branch node (condition judgment item) and a leaf node (measure requirement), realizes the structured analysis and automatic matching of lightning protection clauses, solves the problem that traditional text standards cannot be directly called by algorithms, and effectively improves the matching efficiency of atomic clause units through tree path coding.

[0049] 3. The present application adopts Levenshtein distance (character-level similarity) and BERT semantic similarity weighting, controls the clustering direction through a dynamic weight coefficient, realizes precise merging across standard clauses by combining hierarchical clustering methods, breaks through the conflicts of synonyms and numerical precision, and improves the clause merging accuracy.

[0050] 4、The AHP weight engine time sequence of the present application is positioned after the condition judgment of the decision tree node and before the branch selection, the input technical element set / judgment matrix is output, the weight vector drives the branch selection, and the judgment matrix is optimized by fusing the clustering distance, so that the problem that the analytic hierarchy process is separated from the actual data can be effectively solved;

[0051] 5、The present application can realize self-adaptive protection in complex scenes such as high corrosion and high altitude according to the risk value grading trigger operation, and can overcome the problem of inaccurate judgment of discrete environmental parameters by quantifying the environmental risk value through the relative deviation method. BRIEF DESCRIPTION OF DRAWINGS

[0052] The drawings described herein are used to provide further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application. In the drawings:

[0053] Figure 1 is a flowchart in embodiment 1 of the present application;

[0054] Figure 2 is a schematic diagram of the decision tree of the lightning arrester spacing distance in embodiment 1 of the present application;

[0055] Figure 3 is a system block diagram in embodiment 2 of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical scheme and advantages of the present application more clear and obvious, the present application is further described in detail below in combination with embodiments and drawings, the illustrative embodiments of the present application and the description thereof are only used to explain the present application, and do not constitute a limitation on the present application.

[0057] Embodiment 1: A lightning protection design evaluation method based on a decision tree forest, comprising the following steps:

[0058] S1: Data aggregation processing is performed on the clauses of multi-source lightning protection technical standards to obtain a structured lightning protection clause data set of a dynamic hierarchical classification architecture;

[0059] S2: Special decision trees of each lightning protection measure type are generated according to the structured lightning protection clause data set, and AHP weight engines and conflict resolution engines are implanted in the nodes of the decision trees to obtain a decision tree forest model with weighted branches;

[0060] S3: The lightning protection design parameters to be evaluated are input into the decision tree forest model for dynamic weight integration and conflict resolution processing to obtain a preliminary compliance identification result;

[0061] S4: The decision tree forest model is dynamically adjusted and processed according to the environmental condition parameters, and an environment-adaptive identification result is regenerated based on the preliminary compliance identification result.

[0062] S5: input the environment-adaptive identification result into a meteorological disaster risk management system integration gateway for visual report generation processing, to obtain a lightning protection design evaluation report.

[0063] In step S1, a multi-source standard library needs to be established first to support multi-source data collection and preprocessing.

[0064] The process of establishing the multi-source standard library can collect complete lightning protection technical clause text data from international standards, national standards, industry standards, local specifications, and special technical guidelines through API interfaces, document parsing tools, or manual input methods. For example, the international standard adopts the IEC 62305 series, the national standard adopts GB 50057, the industry standard adopts QX / T 106, and the document parsing tool can use PDF text extraction or OCR (Optical Character Recognition) recognition methods.

[0065] Then, the natural language processing (NLP) technology based on syntax dependency analysis and semantic role labeling is used to decompose complex clauses into indivisible atomic clause units. For example, the clause "when the building height exceeds 60m, the roof should be equipped with a lightning grid" is split into two clause atomic units: "height > 60m" and "install roof lightning grid".

[0066] Next, each atomic clause unit is labeled with the following attributes: source standard, clause original number, validity type, applicable object, lightning protection measure type, and associated parameters. Specifically, the source standard includes the name and version number. The validity type is divided into mandatory and recommended. The applicable object can be divided into lightning protection building categories and types, such as lightning protection building categories, system types, etc. Lightning protection building categories such as Class 1 lightning protection building, Class 2 lightning protection building, and Class 3 lightning protection building, system types such as power system, signal system, etc. Lightning protection measure types such as lightning protection, current diversion, grounding, shielding, equipotential bonding, and SPD configuration, etc. Associated parameters such as height, thunderstorm days, soil resistivity, equipment voltage resistance level, etc.

[0067] The dynamic hierarchical classification architecture of the structured lightning protection clause dataset in the present application includes a measure-oriented layer, an object scenario layer, and a condition constraint layer.

[0068] The measure-oriented layer mainly establishes multiple main classes according to lightning protection measure types; main classes include but are not limited to lightning protection class LSP, current diversion protection class CB, grounding system class ES, electromagnetic shielding class EMS, equipotential bonding class BBN, and surge protection device class SPD.

[0069] The object scene layer is mainly to establish sub-classes under each main class according to applicable objects and application scenarios. The applicable objects are, for example, communication base stations, chemical plant rooms and high-rise residences; and the application scenarios are, for example, new construction / reconstruction, indoor / outdoor.

[0070] The conditional constraint layer is mainly to construct a tree structure according to parameter logical relationships within the sub-classes. The root node of the tree structure is a protection target, for example, “protecting the safety of building personnel”; the branch node is a conditional judgment item, for example, “building height H” and “annual average thunderstorm day Td”; and the leaf node is a specific measure requirement, for example, “grid size of lightning rod ≤ 10 m x 10 m”.

[0071] Traditional data clustering methods mainly include pure text clustering (TF-IDF), pure semantic clustering (BERT) and K-means static clustering. The pure text clustering method cannot handle synonyms, such as “grounding” and “lap joint”; the pure semantic clustering (BERT) is prone to ignore numerical precision, such as “10 m” and “100 m”; and the K-means static clustering cannot adapt to new standards.

[0072] The present application considers that the same technical requirements exist in different standards in the form of text differences, such as “lightning rod belt” and “lightning conductor”;

[0073] In addition, equivalent clauses may be described in different sentence patterns, such as “grid size ≤ 5 m” and “distance between adjacent down conductors ≤ 5 m”; in addition, there are some clauses that are similar in appearance but conflict in essence, such as “when the soil resistivity is > 500 Ω·m, the grounding resistance can be relaxed to 30 Ω” and “must maintain 10 Ω”.

[0074] In the data aggregation processing of the multi-source lightning protection technical standard clauses, the present application adopts a hierarchical clustering algorithm based on the weighted Levenshtein distance (character level similarity) and semantic similarity (such as BERT) to dynamically classify the atomic clause units into a dynamic hierarchical classification framework according to similarity; the key word TF-IDF vector and classification path code are extracted for each atomic clause unit to generate a classification feature vector that can be recognized and dynamically matched by a machine.

[0075] Firstly, the set of lightning protection standard clauses after atomization processing is subjected to text cleaning and term standardization processing. The text cleaning includes: removing legal descriptions and reference marks in the clauses, such as “should”, “should” and the like, and reference marks such as “see 4.2.3” clause”; deleting all non-text symbols, such as parentheses, punctuation marks and serial numbers; and uniformly converting to lowercase characters, such as “SPD” to “spd”. The term standardization processing mainly establishes a synonym mapping dictionary, and replaces synonyms by traversing the clause text, such as the synonyms of lightning rod net, lightning rod grid, lightning protection net and lightning protection net, and the synonyms of SPD, such as surge protector and surge protector.

[0076] Then for any two clause texts and Simultaneously calculate Levenshtein distance and semantic similarity.

[0077] The specific process of calculating the Levenshtein distance is as follows: First, calculate the distance that makes... Become The minimum number of single-character edits required, including operations such as insertion, deletion, and replacement; then confirm. and The two clause texts are determined by the ratio of the minimum number of single-character edits to the maximum character length. and Levenshtein distance between .

[0078] The semantic similarity calculation process is as follows: First, the two clause texts are compared... and The input is fed into a pre-trained multilingual BERT model to generate a 3D semantic vector for the corresponding text. and Next, calculate the cosine similarity between the two clause texts: Finally, the cosine similarity is converted into semantic distance to obtain the two clause texts. and semantic similarity between .

[0079] Different weighting coefficients are configured for different clause types. For example, if a clause contains numeric parameters, the weighting coefficient is set to enhance the accuracy of character comparison. The weighting is 0.2; conversely, if the clauses contain technical terms, the weighting coefficient is set by emphasizing semantic understanding. The weight is 0.6; if none of the above conditions exist, the weight coefficient is set to 0.5 in the balanced mode.

[0080] The final distance between two clause texts is determined by combining Levenshtein distance and semantic similarity. for: .

[0081] In hierarchical clustering, a distance matrix needs to be constructed first based on the final distance, such as calculating it for N terms. Generate a symmetric distance matrix from the given distance values. The main diagonal is 0.

[0082] Then, clustering and merging are performed iteratively, repeating until the termination condition is met: Minimum distance value for mid-positioning Merging clusters Clustering For new class By deleting , New rows and columns rows and columns, and Other classes Distance between The calculation formula is: , for and The distance between them for and The distance between them for and The distance between them.

[0083] This invention, when merging large clusters (such as national standard clusters) and small clusters (such as local standard clusters), through... Automatic weight balancing can effectively solve the problem of cluster size sensitivity; in addition, through The correction ensures that after similar clauses are merged, the distance between the new class and the external class accurately reflects the degree of technical relevance.

[0084] In the process of generating classification feature vectors, the TF-IDF vector of the clause text is first calculated using the scikit-learn library to generate the word frequency feature dimension. For example, if the clause is: "The SPD voltage protection level Up at the power inlet should be less than 80% of the equipment withstand voltage Uw", the generated keyword vector is: [Up: 0.32, Uw: 0.28, SPD: 0.25, power supply: 0.15], where 0.32, 0.28, 0.25, and 0.15 are the corresponding dimension weights. Next, a pre-trained BERT model (such as bert-base-chinese) is used to generate sentence vectors, capturing implicit semantics and outputting semantic vectors. Then, the dynamic hierarchical results are converted into numerical codes to obtain classification path codes. Finally, the TF-IDF vector, semantic vector, and classification path code are concatenated to obtain the classification feature vector, which is stored as a BLOB type field in a MySQL database.

[0085] This invention addresses industry pain points such as scattered standards and outdated updates in traditional lightning protection evaluation by performing standard intelligent processing on multi-source lightning protection clauses, enabling subsequent decision tree construction to obtain logically clear and conflict-controllable structured inputs.

[0086] In step S2, different specialized decision trees can be constructed using the structured lightning protection clause dataset obtained in step S1, such as... Figure 2 The lightning arrester spacing decision tree is shown.

[0087] The decision tree nodes implanted by the AHP weight engine are root nodes and branch nodes, and the function time sequence position of the AHP weight engine is located after condition judgment and before branch selection; wherein the input of the AHP weight engine is a technical element set, a judgment matrix and a consistency threshold value, and the output is an element weight vector and a consistency test state. After the preliminary conclusion is obtained through condition judgment, the decision value is corrected through weight dynamic correction, so as to ensure that the high importance parameter dominates the branch selection.

[0088] Specifically, the technical element set mainly contains lightning protection parameters associated with the current decision node, which can be extracted from the node condition expression parsed from the structured lightning protection clause data set output in step S1.

[0089] The judgment matrix can be generated based on expert knowledge base and historical data training, and is obtained by optimizing the clustering distance in step S1. First, the relative importance of the parameter pair (m, n) in the technical element set is judged, different relative importance levels are given different values, and the initial matrix value can be obtained; then the clustering distance between the m parameter class and the n parameter class is calculated by the clustering distance calculation method in S1 , and the initial value corresponding to the parameter pair (m, n) in the initial matrix value is fused to obtain the optimized element value : The present application considers the actual technical correlation between parameters, objectively corrects the judgment matrix based on the clustering distance, and ensures the reliability and accuracy of the judgment matrix.

[0090] The AHP weight engine performs eigenvalue decomposition on the judgment matrix, and obtains the element weight vector after normalizing the eigenvector corresponding to the maximum eigenvalue, and performs consistency test on the maximum eigenvalue; if the consistency ratio is less than the consistency threshold value, the element weight vector is effective, if the consistency ratio is greater than or equal to the consistency threshold value, the element weight vector is modified and adjusted.

[0091] Then the Boolean value obtained by performing Boolean judgment on the parameters in the technical element set is converted into a continuous quantity, and the satisfaction degree of the corresponding parameter can be obtained, which can accurately quantify the compliance degree; the element weight vector and the satisfaction degree corresponding to the parameter pair are weighted and calculated to obtain the weighted decision value; if the weighted decision value is greater than or equal to the decision threshold value, high-risk protection is executed; if the weighted decision value is less than the decision threshold value, normal protection is executed. For example, the parameter m is the building height, the satisfaction degree is 0.9; the parameter n is the annual average thunderstorm day, the satisfaction degree is 0.8; the element weight vector corresponding to the parameter pair (m, n) is ; the decision threshold value s is 0.7. The weighted decision value is: Its value is greater than the decision threshold value, so the high-risk protection branch is triggered.

[0092] To solve the conflict of multiple source standard clauses in lightning protection design, such as the decision-making scenario driven by the element weight vector, the conflict between the national standard and the line standard for the same parameter is inconsistent, and the conflict resolution engine is used to resolve the conflict clauses.

[0093] Specifically, the conflict resolution engine is configured to read the element weight vector output by the AHP weight engine and the satisfaction degree of the single parameter, determine the conflict mark between the two parameters, retrieve the corresponding priority from the rule base according to the conflict mark to obtain the rule weight, obtain the priority value of the corresponding parameter by multiplying the AHP weight in the element weight vector, the satisfaction degree and the rule weight, and resolve the clauses corresponding to the parameter with a large execution priority value.

[0094] In some examples, the rule base is designed as follows: if the conflict type is the national standard and the line standard, the line standard with more stringent requirements is forced to be executed, and the rule weight corresponding to the priority is 0.9; if the conflict type is mandatory and recommended, the mandatory clause is preferred, and the rule weight corresponding to the priority is 0.85; if the conflict type is the conflict between the new and old versions, the clause with the latest release date is adopted, and the rule weight corresponding to the priority is 0.8; if the conflict type is the industry-specific conflict, the industry-specific clause is executed preferentially, and the rule weight corresponding to the priority is 0.95, and so on.

[0095] In step S3, the dynamic weight integration is realized by the AHP weight engine, and the conflict resolution processing is realized by the conflict resolution engine, and the lightning protection design parameters to be evaluated are output to obtain the preliminary compliance identification result through the corresponding special decision tree.

[0096] In step S4, since the traditional lightning protection evaluation generally uses fixed threshold judgment, such as "if the thunderstorm day is greater than 40 days, the protection level is improved", which cannot respond to complex and variable environmental combinations, such as coastal high corrosion, strong thunderstorm and mountain terrain; in addition, when the environmental conditions exceed the conventional range, the conflict probability of different standard clauses will increase obviously, such as the line standard requires that the grounding body in the high corrosion area is greater than or equal to 100 , and the national standard is greater than or equal to 50 .

[0097] Therefore, the decision tree forest model is dynamically branched and adjusted according to the environmental condition parameters to adapt to complex and variable environmental combinations, and to reduce the conflict probability of clauses.

[0098] In the dynamic branching adjustment process, the environmental condition parameters including at least one of the geographic parameters, the meteorological parameters and the geological parameters are acquired, the geographic parameters such as altitude and terrain, the meteorological parameters such as thunderstorm day, and the geological parameters such as soil resistivity.

[0099] Then, an environmental risk value is determined according to an absolute value of a difference between a parameter value in the environmental condition parameter and a corresponding parameter threshold value, and the greater the absolute value of the difference, the greater the corresponding environmental risk value. For example, the environmental risk value is obtained by dividing the absolute value of the difference between the parameter value in the environmental condition parameter and the corresponding parameter threshold value by the parameter value in the environmental condition parameter. For example, the same absolute height difference in a mountainous area can produce a higher risk value, while the same absolute height difference in a plain area can automatically reduce the risk value, thereby overcoming the problem of inaccurate judgment of environmental parameters due to dispersion through a relative deviation method.

[0100] Then, the maximum value of all the environmental risk values is selected as an execution risk value. When the execution risk value is greater than a first threshold value: a replacement branch library matching the current environmental parameter is retrieved from the decision tree forest model, the environmental parameter can be feature-encoded in 8-bit binary, and the node to be replaced is located through the decision tree metadata index table; and an optimal branch is selected by a satisfaction degree weighted voting mechanism, the satisfaction degree weighted voting mechanism is to obtain a voting value by weighting the satisfaction degree, a complexity coefficient and a protection coefficient, and the maximum voting value is selected as the optimal branch. The complexity coefficient represents the complexity of the environment, the greater the complexity, the greater the complexity coefficient, and the maximum value does not exceed 1. The protection coefficient represents the strictness of protection, the higher the protection requirement, the greater the protection coefficient.

[0101] When the execution risk value is greater than a second threshold value and less than or equal to the first threshold value: the AHP weight of the corresponding node is proportionally increased according to the execution risk value.

[0102] Specifically, the AHP weight increment is calculated by multiplying the industry risk coefficient and the execution risk value; if the sum of the AHP weight increment and the original AHP weight is less than 1, the sum is taken as the updated AHP weight; if the updated AHP weight is greater than or equal to 1, the AHP weight is renormalized. Part of the industry risk coefficient is shown in Table 1.

[0103] Table 1 Industry risk coefficient

[0104] Industry type Thunderstorm days (k1) Corrosion grade (k2) Altitude (k3) Petroleum and chemical 1.5 2.0 0.9 Power plant 1.8 1.5 1.0 Commercial residential 1.0 1.0 0.7 Communication base station 1.2 1.3 1.5 Agricultural facility 0.7 0.5 0.5

[0105] It should be noted that, in order to avoid damaging the overall reasoning chain of the decision tree forest, the environmental adaptation identification should be modified based on the preliminary results rather than independently generated.

[0106] In step S5, the environmental adaptation identification result includes but is not limited to the clause compliance state, the key parameter satisfaction degree, the replaced branch ID, the AHP weight increased node, the real-time data driving adjustment, and the like.

[0107] Working principle: the application solves the standard dispersion through dynamic hierarchical classification architecture, breaks through the limitation of static model through the decision tree forest of double engine, and realizes accurate protection of complex scene through environment risk quantification, and fully surpasses the bottleneck of the prior art. The application is verified by 26 engineering projects, the identification accuracy is 96.3%, the comprehensive efficiency is improved by 48 hours, and the application provides a first "multi-source integration-dynamic adaptation-intelligent decision" solution for the lightning protection design evaluation field.

[0108] Those skilled in the art will appreciate that embodiments of the application can be provided as methods, systems, or computer program products. Accordingly, the application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer-readable program code.

[0109] The application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices produce a device that implements the flow Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more blocks or flows.

[0110] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including instruction devices, which implement the flow Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more blocks or flows.

[0111] These computer program instructions can also be loaded into a computer or other programmable data processing device, so that a series of operation steps are performed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide a process for implementing the flow Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more blocks or flows.

[0112] The above detailed description of the specific embodiments of the present application has been given to understand the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A lightning protection design evaluation method based on decision tree forest, characterized in that, Includes the following steps: S1: Data aggregation processing is performed on the multi-source lightning protection technical standard clauses to obtain a structured lightning protection clauses dataset with a dynamic hierarchical classification architecture; S2: Generate dedicated decision trees for each type of lightning protection measure based on the structured lightning protection clause dataset, and implant AHP weight engine and conflict resolution engine into the decision tree nodes to obtain a decision tree forest model with weighted branches; S3: Input the lightning protection design parameters to be evaluated into the decision tree forest model for dynamic weight integration and conflict resolution to obtain preliminary compliance identification results; S4: Dynamically adjust the branches of the decision tree forest model according to the environmental condition parameters, and regenerate the environmental adaptation identification result based on the preliminary compliance identification result; S5: Input the environmental adaptability identification results into the meteorological disaster risk management system integration gateway for visualization report generation and processing to obtain a lightning protection design evaluation report; The data aggregation processing of the multi-source lightning protection technology standard clauses includes: A hierarchical clustering algorithm based on Levenshtein distance and semantic similarity is used to dynamically classify atomic clause units into the dynamic hierarchical classification architecture according to their similarity. For each atomic clause unit, extract the keyword TF-IDF vector and classification path encoding to generate a classification feature vector that can be recognized by machines and dynamically matched. The dynamic hierarchical classification architecture of the structured lightning protection clauses dataset includes: The measure guidance layer is used to establish multiple main classes based on the lightning protection measure type; The object scenario layer is used to create subclasses under each main class based on the applicable object and application scenario; The condition constraint layer is used to construct a tree structure within the subclass according to the logical relationship of parameters; In this tree structure, the root node represents the protection target, the branch nodes represent conditional judgment items, and the leaf nodes represent specific measures and requirements. The dynamic branch adjustment process is specifically as follows: Obtain environmental condition parameters, including at least one of geographical parameters, meteorological parameters, and geological parameters; The environmental risk value is determined by the absolute value of the difference between the parameter value in the environmental condition parameters and the corresponding parameter threshold. The larger the absolute value of the difference, the larger the corresponding environmental risk value. The maximum value among all the environmental risk values ​​is selected as the execution risk value; When the execution risk value is greater than the first threshold: retrieve the alternative branch library that matches the current environmental parameters from the decision tree forest model; and use a satisfaction-weighted voting mechanism to select the optimal branch and replace the corresponding branch of the original decision tree; When the execution risk value is greater than the second threshold and less than or equal to the first threshold: the AHP weight of the corresponding node is increased proportionally according to the execution risk value.

2. The lightning protection design evaluation method based on decision tree forest according to claim 1, characterized in that, The decision tree nodes implanted in the AHP weight engine are root nodes and branch nodes, and the functional timing position of the AHP weight engine is after the condition judgment and before the branch selection. The AHP weight engine takes a set of technical elements, a judgment matrix, and a consistency threshold as inputs, and outputs a vector of element weights and a consistency check status as outputs.

3. The lightning protection design evaluation method based on decision tree forest according to claim 1, characterized in that, The conflict resolution engine is configured as follows: Read the feature weight vector and the satisfaction of individual parameters output by the AHP weight engine; Determine the conflict flag between two parameters; Based on the conflict marker, the corresponding priority is retrieved from the rule base to obtain the rule weight; The priority value of the corresponding parameter is obtained by multiplying the AHP weight, satisfaction degree, and rule weight in the element weight vector; The ruling shall be executed according to the clause corresponding to the parameter with the higher priority value.

4. The lightning protection design evaluation method based on decision tree forest according to claim 1, characterized in that, The step of proportionally increasing the AHP weight of the corresponding node based on the execution risk value is specifically as follows: The increased AHP weight increment is calculated by multiplying the industry risk coefficient by the execution risk value; If the sum of the AHP weight increment and the original AHP weight is less than 1, then the sum is used as the updated AHP weight. If the updated AHP weight is equal to 1, then the AHP weight is renormalized.

5. A lightning protection design evaluation system based on decision tree forest, characterized in that, include: The data processing module is used to aggregate and process the data of multi-source lightning protection technical standard clauses to obtain a structured lightning protection clause dataset with a dynamic hierarchical classification architecture. The model building module is used to generate dedicated decision trees for each type of lightning protection measure based on the structured lightning protection clause dataset, and to embed the AHP weight engine and conflict resolution engine into the decision tree nodes to obtain a decision tree forest model with weighted branches. The preliminary identification module is used to input the lightning protection design parameters to be evaluated into the decision tree forest model for dynamic weight integration and conflict resolution, so as to obtain the preliminary compliance identification results. The optimization identification module is used to dynamically adjust the branches of the decision tree forest model according to environmental condition parameters, and regenerate the environmental adaptation identification result based on the preliminary compliance identification result. The report generation module is used to input the environmental adaptability identification results into the meteorological disaster risk management system integration gateway for visual report generation processing, and obtain a lightning protection design evaluation report. The data aggregation processing of the multi-source lightning protection technology standard clauses includes: A hierarchical clustering algorithm based on Levenshtein distance and semantic similarity is used to dynamically classify atomic clause units into the dynamic hierarchical classification architecture according to their similarity. For each atomic clause unit, extract the keyword TF-IDF vector and classification path encoding to generate a classification feature vector that can be recognized by machines and dynamically matched. The dynamic hierarchical classification architecture of the structured lightning protection clauses dataset includes: The measure guidance layer is used to establish multiple main classes based on the lightning protection measure type; The object scenario layer is used to create subclasses under each main class based on the applicable object and application scenario; The condition constraint layer is used to construct a tree structure within the subclass according to the logical relationship of parameters; In this tree structure, the root node represents the protection target, the branch nodes represent conditional judgment items, and the leaf nodes represent specific measures and requirements. The dynamic branch adjustment process is specifically as follows: Obtain environmental condition parameters, including at least one of geographical parameters, meteorological parameters, and geological parameters; The environmental risk value is determined by the absolute value of the difference between the parameter value in the environmental condition parameters and the corresponding parameter threshold. The larger the absolute value of the difference, the larger the corresponding environmental risk value. The maximum value among all the environmental risk values ​​is selected as the execution risk value; When the execution risk value is greater than the first threshold: retrieve the alternative branch library that matches the current environmental parameters from the decision tree forest model; and use a satisfaction-weighted voting mechanism to select the optimal branch and replace the corresponding branch of the original decision tree; When the execution risk value is greater than the second threshold and less than or equal to the first threshold: the AHP weight of the corresponding node is increased proportionally according to the execution risk value.

6. A computer terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the lightning protection design evaluation method based on decision tree forest as described in any one of claims 1-4.

7. A computer-readable medium having a computer program stored thereon, characterized in that, The computer program, when executed by a processor, can implement the lightning protection design evaluation method based on decision tree forest as described in any one of claims 1-4.

Images