Large-scale engineering equipment recommendation method based on knowledge graph in complex environment

Abstract

The invention relates to a knowledge graph-based large-scale engineering equipment recommendation method in a complex environment, and the method comprises the following steps: S1, collecting related data selected by a plurality of large-scale engineering equipment in a complex construction environment, extracting a plurality of influence factors selected by each large-scale engineering equipment, and constructing a knowledge graph oriented to the complex construction environment; s2, training a random forest model by using the knowledge graph data, and adjusting hyper-parameters of the random forest model through Bayesian optimization; and S3, obtaining an adaptability score of each kind of large engineering equipment under each influence factor, calculating a total adaptability score, and selecting the large engineering equipment with the highest total adaptability score as a recommendation result. According to the method, the knowledge graph is constructed to integrate multi-source data in a complex construction environment, the equipment adaptability is predicted by using the random forest algorithm, and the model performance is improved through Bayesian optimization, so that accurate and efficient large-scale engineering equipment recommendation is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent recommendation of engineering equipment, and in particular relates to a method for recommending large-scale engineering equipment based on knowledge graphs in complex environments. Background Art

[0002] In modern engineering projects, the rational selection of equipment has become one of the core factors to ensure the smooth progress of construction projects and to meet high-quality engineering standards. Faced with a complex and changeable construction environment, equipment selection is not only related to the economic benefits of the project, but also has a profound impact on the safety, schedule control and quality management of the project. The selection of large-scale engineering equipment in a complex construction environment is affected by multiple factors such as geological conditions, climatic conditions and equipment performance. Therefore, in large-scale engineering projects, the rational selection of construction equipment directly affects construction efficiency, cost control and engineering quality, especially in complex construction environments (such as changeable geological conditions, harsh climate, and high equipment performance requirements). The scientificity and accuracy of equipment selection are crucial. Traditional methods mainly rely on manual experience judgment and make decisions by consulting historical cases, equipment manuals and expert opinions. However, with the expansion of the scale of the project and the complexity of the construction environment, the traditional model has gradually exposed the following problems: Insufficient data integration capabilities; the selection of construction equipment involves multi-source heterogeneous data such as geological parameters, equipment performance, and climate conditions (including structured data such as sensor logs and unstructured data such as construction reports), and traditional methods lack effective data integration methods. The data is scattered and difficult to associate, resulting in the inability to fully evaluate the mutual impact of various factors when making decisions.

[0003] Subjectivity and inefficiency: Manual decision-making relies on expert experience, is highly subjective, and has low analysis efficiency when faced with massive amounts of data. For example, in tunnel engineering, when geological conditions change dynamically, it is difficult for humans to quickly adjust equipment selection strategies, resulting in construction delays or cost overruns.

[0004] Poor model adaptability: Existing algorithm models (such as single regression models or rule engines) have difficulty in handling complex relationships involving multiple factors. For example, traditional machine learning models cannot dynamically integrate real-time changes in equipment performance parameters and environmental conditions, resulting in a large deviation between recommended results and actual needs.

[0005] Limitations of hyperparameter optimization: Although integrated models such as random forests can process high-dimensional data, their hyperparameters (such as the number and depth of trees) rely on manual trial and error adjustment and lack automated optimization mechanisms, making it difficult to achieve optimal model performance.

[0006] In the existing technology, although there have been studies on equipment selection based on methods such as Analytic Hierarchy Process (AHP) and multi-criteria decision-making, these methods have limitations when processing multi-dimensional and complex relationship data, and cannot accurately and efficiently select large-scale engineering equipment in complex construction environments. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a large-scale engineering equipment recommendation method based on knowledge graph in a complex environment. The method integrates multi-source data in a complex construction environment by constructing a knowledge graph, predicts equipment adaptability using a random forest algorithm, and improves model performance through Bayesian optimization, thereby achieving accurate and efficient large-scale engineering equipment recommendation.

[0008] The present invention provides a large-scale engineering equipment recommendation method based on knowledge graph in a complex environment, comprising the following steps: S1: Collect relevant data on the selection of several types of large-scale engineering equipment in complex construction environments, extract several influencing factors for the selection of each type of large-scale engineering equipment, and construct a knowledge graph for complex construction environments; S2: Use knowledge graph data to train the random forest model and adjust the hyperparameters of the random forest model through Bayesian optimization; S3: Obtain the adaptability score of each large-scale engineering equipment under each influencing factor, calculate the total adaptability score, and select the large-scale engineering equipment with the highest total adaptability score as the recommendation result.

[0009] Preferably, the step S1 specifically includes the following steps: S1.1: Collect and extract entities, attributes and relationships related to complex construction environments; collect and extract various structured and unstructured data files such as literature, engineering cases, expert opinions, equipment manuals, construction logs, etc. from leading enterprises in the domestic engineering equipment manufacturing industry; S1.2: Build a knowledge graph and use Neo4j graph database to store and query entities and relationships in the knowledge graph; S1.3: Entities and relations are mapped into low-dimensional vectors through knowledge graph representation learning methods.

[0010] Preferably, the knowledge graph representation learning method in step S1.3 adopts the TransH model, introduces hyperplanes for different relationships, captures the many-to-many relationships between entities, and uses the scoring function Calculate the distance between entities to support the recommendation of large engineering equipment. The scoring function The calculation formula includes the following: , , , Among them, h and t are the vector representations of the head entity and the tail entity respectively, and r is the many-to-many relationship between entities. is the hyperplane normal vector of relation r, T is the transposition sign, and dr is the translation vector of relation r.

[0011] Preferably, the step S2 specifically includes the following steps: S2.1: Extract key features that affect the selection of each large-scale engineering equipment from the knowledge graph as input, including the relationship, distance and topological structure features between nodes; S2.2: Use historical data to train a random forest model to predict the adaptability score of large-scale engineering equipment in a specific complex construction environment; S2.3: Adjust the hyperparameters of random forest through Bayesian optimization to optimize model performance.

[0012] Preferably, the step S2.3 specifically includes the following steps: S2.3.1: Determine the hyperparameter optimization space; define the key hyperparameter range of random forest, including the number of trees, maximum depth and splitting criteria, as the search space for Bayesian optimization; S2.3.2: Initialize the sample; randomly select several hyperparameter combinations and calculate their model performance objective function values ​​as initial training data; S2.3.3: Build a surrogate model; use Gaussian process regression to fit the mapping relationship between hyperparameters and objective function values ​​to generate a posterior distribution; S2.3.4: Optimize the acquisition function; use the expected improvement or upper confidence limit acquisition function to select the next set of hyperparameters to be evaluated from the search space; S2.3.5: Iterative update: add the newly evaluated hyperparameters and their performance results to the training data, update the proxy model, and iterate until convergence to the optimal hyperparameter combination.

[0013] Preferably, the Bayesian optimization fits the posterior distribution of the objective function through Gaussian process regression, and uses an acquisition function to select the optimal hyperparameter combination.

[0014] Preferably, the step S3 specifically includes the following steps: S3.1: Combined with the features extracted from the knowledge graph, the optimized random forest model is used to predict the predicted adaptability score of each large-scale engineering equipment under several influencing factors; S3.2: The total adaptability score of each large-scale engineering equipment is calculated by a weighted sum formula, and the total adaptability score includes the following calculation formula: , in, is the weight of the i-th influencing factor, Score the predicted adaptability of the equipment under the i-th influencing factor; S3.3: Compare the overall adaptability scores of various types of large-scale engineering equipment, and select the large-scale engineering equipment with the highest overall adaptability score as the recommendation result.

[0015] The method is suitable for equipment selection in diverse and complex construction environments such as plateau railways, complex geological conditions or severe climates.

[0016] The present invention has the following technical effects: 1. By building a knowledge graph to integrate multi-source data in complex construction environments, using the random forest algorithm for prediction, and improving model performance through Bayesian optimization, we can achieve intelligent recommendations for large-scale engineering equipment.

[0017] 2. Through the knowledge graph, various influencing factors such as geological conditions and climatic conditions are effectively integrated, improving data utilization efficiency.

[0018] 3. Combining the random forest algorithm and Bayesian optimization, the accuracy and adaptability of the recommendation model are significantly improved.

[0019] 4. Reduces reliance on human intervention and subjective judgment, and reduces decision-making risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Constructing a flow chart for the knowledge graph of the method of the present invention; Figure 2 It is a schematic diagram of the TransH model; Figure 3 Schematic diagram of the random forest-Bayesian optimization model of the present invention; Figure 4 is an example of a node in the Neo4j graph database of the present invention; Figure 5 This is a graph showing the importance ratio of the influencing factors of the present invention; Figure 6 This is a performance comparison chart of the algorithm of the present invention. DETAILED DESCRIPTION

[0021] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings.

[0022] Example A large-scale engineering equipment recommendation method based on knowledge graph in a complex environment includes the following steps: S1: Collect relevant data on the selection of several types of large-scale engineering equipment in complex construction environments, extract several influencing factors for the selection of each type of large-scale engineering equipment, and construct a knowledge graph for complex construction environments; S1.1: Collect and extract entities, attributes, and relationships related to complex construction environments; collect and extract various structured and unstructured data files such as literature materials, engineering cases, expert opinions, equipment manuals, and construction logs from leading domestic engineering equipment manufacturing enterprises; Perform text cleaning on unstructured data such as construction logs to remove irrelevant words (such as "of", "is"); use named entity recognition (NER) technology to extract key variables, including "geological hardness", "low-temperature environment", "roadheader efficiency"; identify the relationships between variables through relation extraction technology, such as "altitude > 3000m reduces roadheader efficiency"; Output result: Screen out 20 main influencing factors, including geological conditions, climate conditions, etc., and the weight distribution is shown in Table 1 and Figure 5 as follows.

[0023] Table 1

[0024] S1.2: Construct a knowledge graph, and use the Neo4j graph database to store and query entities and relationships in the knowledge graph; as Figure 4 shown; S1.3: Through the knowledge graph representation learning method, map entities and relationships into low-dimensional vectors.

[0025] In the knowledge graph representation learning method in step S1.3, the TransH model is adopted, as Figure 2 shown. Introduce a hyperplane for different relationships to capture the many-to-many relationships between entities, and calculate the distance between entities through a scoring function to support the recommendation of large engineering equipment; among them, the optimized embedding dimension is 100, the learning rate is 0.01, and the scoring function includes the following calculation formula: , , , where h and t are the vector representations of the head entity and the tail entity respectively, r is the many-to-many relationship between entities, is the normal vector of the hyperplane of relationship r, T is the transpose symbol, and dr is the translation vector of relationship r.

[0026] Calculate the distance between the equipment and the influencing factors on the hyperplane through this function to support subsequent recommendations. The output result is a knowledge graph containing 20 influencing factors and 3 types of equipment (roadheaders, rock drilling jumbo, wet shotcreting machines), as Figure 1 shown.

[0027] S2: Use knowledge graph data to train the random forest model and adjust the hyperparameters of the random forest model through Bayesian optimization; Figure 3 As shown; S2.1: Extract key features that affect the selection of each large-scale engineering equipment from the knowledge graph as input, including the relationship, distance and topological structure features between nodes; S2.2: Use historical data to train a random forest model to predict the adaptability score of large-scale engineering equipment in a specific complex construction environment. Use historical data (including three types of equipment: tunnel boring machine, rock drilling rig, and wet spraying rig) to train a random forest model, use random forest to predict the adaptability score of the equipment under 20 factors, evaluate the model performance through 10-fold cross validation, and generate an initial equipment adaptability score prediction model.

[0028] S2.3: Adjust the hyperparameters of random forest through Bayesian optimization to optimize model performance.

[0029] S2.3.1: Determine the hyperparameter optimization space; define the key hyperparameter range of random forest, including the number of trees, maximum depth and splitting criterion, as the search space for Bayesian optimization; S2.3.2: Initialize the sample; randomly select several hyperparameter combinations and calculate their model performance objective function values ​​as the initial training data; S2.3.3: Build a proxy model; fit the mapping relationship between hyperparameters and objective function values ​​through Gaussian process regression to generate the posterior distribution; S2.3.4: Optimize the acquisition function; use the expected improvement or upper confidence limit acquisition function to select the next set of hyperparameters to be evaluated from the search space; S2.3.5: Iterative update; add the newly evaluated hyperparameters and their performance results to the training data, update the proxy model, and iterate until convergence to the optimal hyperparameter combination.

[0030] The Bayesian optimization fits the posterior distribution of the objective function through Gaussian process regression and selects the optimal hyperparameter combination using an acquisition function.

[0031] The hyperparameters ranged from 100 to 500 trees, 10 to 50 maximum depth, and 2 to 10 minimum number of splits. Gaussian process regression was used to fit the posterior distribution of the performance function. The hyperparameters were optimized using the expected improvement (EI) acquisition function with 20 iterations.

[0032] S3: Obtain the adaptability score of each large-scale engineering equipment under each influencing factor, calculate the total adaptability score, and select the large-scale engineering equipment with the highest total adaptability score as the recommendation result; S3.1: Combined with the features extracted from the knowledge graph, the optimized random forest model is used to predict the adaptability score of each large-scale engineering equipment under several influencing factors; The plateau railway project data (geological hardness 40 MPa, altitude 3500m, temperature -10℃) was input; the predicted adaptability scores under 20 factors were output using the optimized random forest model. The predicted adaptability scores of three types of large-scale engineering equipment are shown in Table 2.

[0033] Table 2

[0034] S3.2: The total adaptability score of each large-scale engineering equipment is calculated by a weighted sum formula, and the total adaptability score includes the following calculation formula: , in, is the weight of the i-th influencing factor (from Table 1), The predicted adaptability score (from Table 2) of the equipment under the i-th influencing factor should be equal to 1.

[0035] S3.3: Compare the overall adaptability scores of various types of large-scale engineering equipment, and select the large-scale engineering equipment with the highest overall adaptability score as the recommendation result.

[0036] The calculation results of the above three large-scale engineering equipment are as follows: Total score of roadheader: 0.82 Total score of drilling rig: 0.67 Total score of wet spraying trolley: 0.59 From the above, we can see that the roadheader has the highest score and is recommended as the optimal equipment.

[0037] Test example 1. Performance test: The evaluation indicators commonly used for recommendation tasks in recommendation systems are: AUC, Recall, Precision, F1, and Accuracy.

[0038] Accuracy represents the accuracy, which is calculated as follows: , Among them, TP represents the number of samples correctly predicted as positive by the model; FP represents the number of samples incorrectly predicted as positive by the model; FN represents the number of samples incorrectly predicted as negative by the model; TN represents the number of samples correctly predicted as negative by the model.

[0039] Recall represents the recall rate, which is calculated as follows: , Among them, TP represents the number of samples correctly predicted as positive by the model; FP represents the number of samples incorrectly predicted as positive by the model; FN represents the number of samples incorrectly predicted as negative by the model; TN represents the number of samples correctly predicted as negative by the model.

[0040] Precision represents the precision rate, which is calculated as follows: , Among them, TP represents the number of samples correctly predicted as positive by the model; FP represents the number of samples incorrectly predicted as positive by the model; FN represents the number of samples incorrectly predicted as negative by the model; TN represents the number of samples correctly predicted as negative by the model.

[0041] The F1 value combines the Recall and Precision evaluation indicators. The larger the F1 value, the larger the Recall and Precision values. The calculation method is as follows: , Among them, TP represents the number of samples correctly predicted as positive by the model; FP represents the number of samples incorrectly predicted as positive by the model; FN represents the number of samples incorrectly predicted as negative by the model; TN represents the number of samples correctly predicted as negative by the model.

[0042] AUC is the area under the ROC curve. AUC = 0.5 is defined as random guessing. When AUC is greater than 0.5, the larger the classification accuracy is. The meanings of the parameters in the above formula are shown in Table 3.

[0043] Table 3

[0044] 2. Determination of comparison method: Compare the recommendation performance based on this embodiment (hereinafter referred to as this embodiment RF (Knowledge Graph + BO)) and the other four algorithms. Four classification methods in machine learning, namely, random forest algorithm (RF (Standard)), random forest algorithm based on knowledge graph (RF (Knowledge Graph)), gradient boosting decision tree algorithm (GBDT), and K nearest neighbor algorithm (KNN), are selected as comparative experimental methods.

[0045] 3. Record evaluation indicators: The recommended algorithms of this embodiment RF (Knowledge Graph+BO), random forest algorithm (RF (Standard)), random forest algorithm based on knowledge graph (RF (Knowledge Graph)), gradient boosting decision tree algorithm (GBDT), and K nearest neighbor algorithm (KNN) are recorded, and their performance is shown in Table 4 and Figure 6 shown.

[0046] Table 4

[0047] As shown in Table 4, by comparing the RF (Knowledge Graph+BO) algorithm of this embodiment with the random forest algorithm (RF (Standard)), the random forest algorithm based on the knowledge graph (RF (Knowledge Graph)), the gradient boosting decision tree algorithm (GBDT), and the K nearest neighbor algorithm (KNN), it is shown that the method described in the present invention is significantly higher than the existing algorithms in accuracy, recall rate and precision rate, and the prediction effect of positive examples is significantly improved. At the same time, it can reduce the occurrence of negative examples predicted as positive to a certain extent, and improve the recommendation effect of large-scale engineering equipment selection under complex construction environments. The present invention integrates multi-source data under complex construction environments by constructing a knowledge graph, uses the random forest algorithm for prediction, and improves the model performance through Bayesian optimization, thereby realizing intelligent recommendation for large-scale engineering equipment. The knowledge graph effectively integrates various influencing factors such as geological conditions and climatic conditions, thereby improving data utilization efficiency. Combining the random forest algorithm and Bayesian optimization, the accuracy and adaptability of the recommendation model are significantly improved. The dependence on manual intervention and subjective judgment is reduced, and the decision-making risk is reduced.

[0048] The present invention is suitable for equipment selection in diverse and complex construction environments such as plateau railways, complex geological conditions or severe climates.

[0049] The embodiments described above are only descriptions of the preferred modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A large-scale engineering equipment recommendation method based on knowledge graph in a complex environment, characterized by: The steps include: S1: Collect relevant data on the selection of several types of large-scale engineering equipment in complex construction environments, extract several influencing factors for the selection of each type of large-scale engineering equipment, and construct a knowledge graph for complex construction environments; S2: Use knowledge graph data to train the random forest model and adjust the hyperparameters of the random forest model through Bayesian optimization; S3: Obtain the adaptability score of each large-scale engineering equipment under each influencing factor, calculate the total adaptability score, and select the large-scale engineering equipment with the highest total adaptability score as the recommendation result.

2. The large-scale engineering equipment recommendation method based on knowledge graph in a complex environment according to claim 1 is characterized by: The step S1 specifically includes the following steps: S1.1: Collect and extract entities, attributes and relationships related to complex construction environments; S1.2: Build a knowledge graph and use Neo4j graph database to store and query entities and relationships in the knowledge graph; S1.3: Entities and relations are mapped into low-dimensional vectors through knowledge graph representation learning methods.

3. The large-scale engineering equipment recommendation method based on knowledge graph in complex environment according to claim 2 is characterized by: The knowledge graph representation learning method in step S1.3 adopts the TransH model, introduces hyperplanes for different relationships, captures the many-to-many relationships between entities, and uses the scoring function Calculate the distance between entities to support the recommendation of large engineering equipment. The scoring function The calculation formula includes the following: ， ， ， Among them, h and t are the vector representations of the head entity and the tail entity respectively, and r is the many-to-many relationship between entities. is the hyperplane normal vector of relation r, T is the transposition sign, and dr is the translation vector of relation r.

4. The large-scale engineering equipment recommendation method based on knowledge graph in complex environment according to claim 3 is characterized by: The step S2 specifically includes the following steps: S2.1: Extract key features that affect the selection of each large-scale engineering equipment from the knowledge graph as input, including the relationship, distance and topological structure features between nodes; S2.2: Use historical data to train a random forest model to predict the adaptability score of large-scale engineering equipment in a specific complex construction environment; S2.3: Adjust the hyperparameters of random forest through Bayesian optimization to optimize model performance.

5. The large-scale engineering equipment recommendation method based on knowledge graph in complex environment according to claim 4 is characterized by: The step S2.3 specifically includes the following steps: S2.3.1: Determine the hyperparameter optimization space; define the key hyperparameter range of random forest, including the number of trees, maximum depth and splitting criterion, as the search space for Bayesian optimization; S2.3.2: Initialize the sample; randomly select several hyperparameter combinations and calculate their model performance objective function values ​​as the initial training data; S2.3.3: Build a proxy model; fit the mapping relationship between hyperparameters and objective function values ​​through Gaussian process regression to generate the posterior distribution; S2.3.4: Optimize the acquisition function; use the expected improvement or upper confidence limit acquisition function to select the next set of hyperparameters to be evaluated from the search space; S2.3.5: Iterative update; add the newly evaluated hyperparameters and their performance results to the training data, update the proxy model, and iterate until convergence to the optimal hyperparameter combination.

6. The large-scale engineering equipment recommendation method based on knowledge graph in complex environment according to claim 5 is characterized in that: The Bayesian optimization fits the posterior distribution of the objective function through Gaussian process regression and selects the optimal hyperparameter combination using an acquisition function.

7. The large-scale engineering equipment recommendation method based on knowledge graph in complex environment according to claim 6 is characterized by: The step S3 specifically comprises the following steps: S3.1: Combined with the features extracted from the knowledge graph, the optimized random forest model is used to predict the predicted adaptability score of each large-scale engineering equipment under several influencing factors; S3.2: The total adaptability score of each large-scale engineering equipment is calculated by a weighted sum formula, and the total adaptability score includes the following calculation formula: ， in, is the weight of the i-th influencing factor, Score the predicted adaptability of the equipment under the i-th influencing factor; S3.3: Compare the overall adaptability scores of various types of large-scale engineering equipment, and select the large-scale engineering equipment with the highest overall adaptability score as the recommendation result.

Images