A method for evaluating energy transmission efficiency of vibration compaction forming

Abstract

The present application relates to the technical field of data processing, in particular to a vibration compaction forming energy transmission efficiency evaluation method, comprising: firstly, integrating multi-source data, generating semantic enhanced data set through adaptive semantic labeling algorithm; then, based on the semantic enhanced data set, node relationship extraction is carried out, and a first graph algorithm is used to construct and dynamically maintain a knowledge graph model, which is used to structurally represent the nodes and relationships of the compaction process; based on the knowledge graph model, the second graph algorithm is used to evaluate the parameter influence result, and the time series prediction algorithm is used to generate the risk prediction result; then, the reward function is defined according to the parameter influence result and the risk prediction result, and the optimal vibration compaction parameter combination is determined; finally, the parameter recommendation report is generated, and the user feedback is collected to iteratively optimize the knowledge graph. The present application can improve the accuracy of vibration compaction parameter selection, optimize the energy transmission efficiency and engineering quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a vibration compaction forming energy transmission efficiency evaluation method. BACKGROUND

[0002] The final quality and energy efficiency of vibration compaction engineering are highly nonlinearly coupled with the physical and mechanical properties of the fillers, the performance parameters of the compaction equipment, and the fine control strategy of the compaction process. In order to ensure the long-term stability and use performance of, for example, high-grade highway subgrade, high-speed railway bed, or key infrastructure backfill area, the engineering field puts forward increasingly strict requirements for the matching of filler gradation, maximum particle size, water content, and compaction equipment selection and operation parameters. The traditional vibration compaction parameter selection is mostly dependent on the experience of engineers, local standards, or verification based on a small amount of field tests, which is often difficult to fully consider the variable working conditions and material differences, and is prone to cause low compaction energy utilization rate and insufficient compaction uniformity, or local overpressure and underpressure problems.

[0003] Reinforcement learning, as a machine learning method that can learn the optimal strategy through interaction with the environment, provides a new idea for solving the above problems. By constructing a reinforcement learning model suitable for the field of vibration compaction, the scattered geological data, material constitutive model parameters, equipment performance curves, construction specification requirements, historical project compaction schemes and their effect data, and even a large amount of CAE simulation data set can be used as part of the training data or environmental state for structured utilization. However, merely introducing the reinforcement learning framework does not mean that the strategy optimization has been achieved. The current core technical challenge is that there is a lack of effective means to recommend the compaction parameter sequence suitable for the current working condition through reinforcement learning intelligence, to achieve strategy optimization, and to dynamically optimize the energy distribution of the entire compaction process, avoiding problems such as "vibration jump", excessive energy consumption, reduced equipment life, or excessive material crushing caused by improper excitation energy selection. Therefore, a vibration compaction forming energy transmission efficiency evaluation method is proposed. SUMMARY

[0004] The application aims to provide a vibration compaction forming energy transmission efficiency evaluation method, aiming to improve the accuracy of vibration compaction parameter selection, optimize energy transmission efficiency and engineering quality. First, the multi-source data is integrated, and a semantic enhancement dataset is generated by an adaptive semantic labeling algorithm based on machine learning and graph clustering; then, a GNN algorithm is applied for node relationship extraction, and a first graph algorithm is used to build and dynamically maintain a knowledge graph model for structured representation of node relationships in the compaction process; based on the knowledge graph model, a second graph algorithm is used to evaluate parameter influence results, and a time series prediction algorithm is used to generate risk prediction results; then, a reward function is defined according to the parameter influence results and the risk prediction results, and a deep reinforcement learning algorithm is used to determine the optimal vibration compaction parameter combination to realize strategy optimization; finally, a parameter recommendation report is generated and user feedback is collected to iteratively optimize the knowledge graph.

[0005] To achieve the above-mentioned purpose, the application provides the following technical scheme:

[0006] A vibration compaction forming energy transmission efficiency evaluation method comprises:

[0007] Integrate multi-source heterogeneous vibration compaction data, and generate a semantic enhancement dataset by an adaptive semantic labeling algorithm based on a machine learning algorithm and a graph clustering algorithm;

[0008] Based on the semantic enhancement dataset, a domain-enhanced GNN algorithm is applied for node relationship extraction, and a first graph algorithm is used for incremental update to build and dynamically maintain a knowledge graph model; the knowledge graph model is used to structure the nodes and relationships in the vibration compaction process;

[0009] Based on the knowledge graph model, a second graph algorithm is used to evaluate parameter influence results; based on the node embedding features of the knowledge graph model, a time series prediction algorithm is used to generate risk prediction results; a reward function is defined according to the parameter influence results and the risk prediction results, and a deep reinforcement learning algorithm is used to determine the optimal vibration compaction parameter combination;

[0010] Based on the optimal vibration compaction parameter combination and the knowledge graph path, a parameter recommendation report is generated, and user feedback is collected to iteratively optimize the knowledge graph model.

[0011] Further, the implementation process of the adaptive semantic labeling algorithm comprises:

[0012] A machine learning algorithm fine-tuned on a target engineering field dataset is used to generate a field embedding vector for each original data field in the multi-source heterogeneous vibration compaction data;

[0013] embedding the fields into a vector input graph clustering algorithm, which clusters the original data fields according to the similarity degree of the embedding vectors and establishes a dynamic annotation rule for each cluster;

[0014] applying the dynamic annotation rule to map the original data fields to standard concepts and attributes in a predefined vibration compaction domain ontology, performing consistency verification on the semantic labels formed by the mapping through reference to the vibration compaction domain ontology, and generating the semantic enhanced data set.

[0015] Further, the process of node relationship extraction includes:

[0016] Converting the semantic enhanced data set into an initial graph structure, which is used to represent the key data elements identified in the semantic enhanced data set as nodes and the established semantic classification information and preliminary structural links between key data elements as relationships;

[0017] Using a GNN algorithm pre-trained on a target engineering field data set to perform distributed representation learning on the nodes in the initial graph structure to identify key engineering entities; and performing relationship classification based on the key engineering entities to form structured knowledge triples;

[0018] Performing effectiveness verification on the structured knowledge triples and a predefined vibration compaction domain ontology, and outputting the verified structured knowledge triples.

[0019] Further, the process of using the first graph algorithm to perform incremental updates to build and dynamically maintain the knowledge graph model includes:

[0020] When receiving newly added semantic enhanced data, using the first graph algorithm to generate a time sequence embedding representation for the entity interaction events in the newly added semantic enhanced data;

[0021] Extracting new structured knowledge triples from the time sequence embedding representation in combination with a lightweight graph neural network to form a local new graph;

[0022] Through a graph difference detection algorithm, identifying the change part of integrating the local new graph into the knowledge graph model, and performing an update operation on the corresponding change part of the knowledge graph model; the update operation includes adding, modifying and deleting nodes and relationships, and simultaneously performing effectiveness verification based on a predefined vibration compaction domain ontology.

[0023] Further, the process of evaluating parameter influence results using the second graph algorithm comprises: extracting a dynamic context subgraph from the knowledge graph model according to the current compaction working condition and the candidate vibration compaction parameter by using a graph segmentation algorithm; inputting the dynamic context subgraph into a pre-trained second graph algorithm; the second graph algorithm is used to capture the spatio-temporal dependency and node interaction mode in the dynamic context subgraph, reason and output the parameter influence result, and the parameter influence result is a quantitative representation of the candidate vibration compaction parameter on the pre-defined compaction effect index.

[0024] Further, the process of generating a risk prediction result comprises: extracting a dynamic context subgraph from the knowledge graph model according to the current compaction working condition and the candidate vibration compaction parameter by using a graph segmentation algorithm; applying a node embedding algorithm to generate a node embedding vector representation for the nodes in the dynamic context subgraph, and inputting the node embedding vector representation as a node embedding feature into a pre-trained time series prediction algorithm to output the risk prediction result corresponding to the candidate vibration compaction parameter.

[0025] Further, the process of determining the optimal vibration compaction parameter combination using a deep reinforcement learning algorithm comprises:

[0026] The deep reinforcement learning environment is constructed, including: defining the agent state, including the current compaction working condition features, material properties, device state, parameter influence result and risk prediction result extracted from the knowledge graph model; defining the agent action, including selecting the next set of vibration compaction parameters; defining the agent reward function, including: the expected compaction quality improvement amount evaluated according to the parameter influence result, the expected energy consumption and device wear and tear inferred from the knowledge graph model, and the risk exceeding the safety threshold according to the risk prediction result;

[0027] Based on the deep reinforcement learning environment, a deep reinforcement learning algorithm is used to train a parameter selection strategy that can maximize the long-term cumulative expected reward, and the output of the parameter selection strategy is the optimal vibration compaction parameter combination.

[0028] Compared with the prior art, the present application has the following beneficial effects:

[0029] 1、The present application integrates multi-source heterogeneous vibration compaction data such as computer-aided engineering export data by using an adaptive semantic labeling algorithm. On this basis, a structured vibration compaction field knowledge graph model is constructed and continuously dynamically maintained by using a field-enhanced GNN algorithm and advanced first graph algorithm technology, which provides a unified, comprehensive and real-time mapping of working condition changes for the subsequent energy transmission efficiency evaluation, parameter influence analysis, risk prediction and intelligent optimization decision-making.

[0030] 2、The application accurately evaluates the parameter influence by constructing a dynamic knowledge graph model combined with the second graph algorithm. Meanwhile, the node embedding features of the knowledge graph and the time series prediction algorithm are integrated to effectively predict the operational risk. This multi-dimensional and intelligent analysis and prediction capability reveals the dynamic influence of various parameters on the compaction effect indicators in the complex vibration compaction process, identifies potential risks in a forward-looking manner, not only improves the understanding depth and prediction accuracy of the vibration compaction process, but also provides a solid foundation for realizing fine and adaptive process control.

[0031] 3、Based on the comprehensive evaluation of parameter influence and risk, the application combines the reward function defined by the knowledge graph model attribute and the deep reinforcement learning algorithm integrated with risk constraints to achieve multi-objective optimization of the vibration compaction process parameter combination. This method can accurately determine the optimal parameter combination that ensures compaction quality and avoids operational risks, generates an interpretable parameter recommendation report with the help of the knowledge graph path, and assists engineers in making scientific decisions. In addition, a user feedback mechanism is introduced to continuously iterate and optimize the knowledge graph model, forming a learning and improvement closed loop. This effectively improves the vibration compaction forming energy transmission efficiency, construction quality and equipment utilization, and optimizes the energy transmission efficiency and engineering quality. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 A flowchart of a vibration compaction forming energy transmission efficiency evaluation method is provided for the application;

[0033] Figure 2 A flowchart of the process of entity relationship extraction is provided for the application;

[0034] Figure 3 A flowchart of using a deep reinforcement learning algorithm to determine the optimal vibration compaction parameter combination is provided for the application. DETAILED DESCRIPTION

[0035] The technical solutions in the embodiments of the application will be described in detail below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only a part of the embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0036] Please refer to Figures 1 to 3 The application provides a vibration compaction forming energy transmission efficiency evaluation method, and the technical solutions are as follows:

[0037] Embodiment one:

[0038] Vibration compaction is a key process in infrastructure construction such as road and foundation, and its compaction quality and energy utilization efficiency directly affect the economy and durability of the project. However, the process involves dynamic interactions between equipment characteristics, complex multi-source material data, and numerous process parameters, making it extremely challenging to achieve optimal control and energy efficiency evaluation. Traditional methods rely heavily on experience or isolated analysis, making it difficult to integrate heterogeneous data for comprehensive and dynamic parameter optimization and efficiency evaluation, often resulting in low energy efficiency and quality fluctuations.

[0039] To overcome the above limitations and improve the intelligent level and energy utilization efficiency of vibration compaction, this embodiment takes the vibration compaction process of a certain highway roadbed as an example and elaborates a vibration compaction energy transfer efficiency evaluation method. As shown in Figure 1

[0040] Integrate multi-source heterogeneous vibration compaction data and generate a semantic enhanced dataset through an adaptive semantic labeling algorithm based on machine learning algorithm and graph clustering algorithm.

[0041] The multi-source heterogeneous vibration compaction data includes but is not limited to: computer-aided engineering (CAE) simulation analysis output files, such as CSV or text format output containing stress field, strain field, and density distribution information; real-time sensor data collected in the compaction field, including but not limited to vibration frequency, amplitude, rolling speed, GPS positioning information, recorded in log or streaming data form; database records of historical engineering projects, including construction process parameters, material mechanical performance indicators, and final compaction quality data; standardized material physical and mechanical performance test reports, such as PDF or electronic form test data.

[0042] Further, the multi-source heterogeneous vibration compaction data has problems such as format diversification and semantic expression heterogenization, affecting the efficiency and accuracy of data fusion and semantic unified processing. To address these challenges, the implementation process of the adaptive semantic labeling algorithm includes:

[0043] First, use a machine learning algorithm fine-tuned on a target engineering domain dataset to generate a field embedding vector for each original data field in the multi-source heterogeneous vibration compaction data.

[0044] The original data field refers to the original name or identifier of a specific information segment in the data source, such as the column name "Freq_Hz" in the CSV file and the key name "VibrationFrequency" in the sensor log. In this embodiment, a BERT model based on the Transformer architecture is pre-trained on the relevant corpus of a certain highway to extract context-sensitive semantic vectors, and each field is converted into a field embedding vector. ​

[0045] Subsequently, the fields are embedded into vector input graph clustering algorithms (such as Louvain community detection algorithm or K-Means clustering, etc.), which cluster the original data fields according to the similarity degree of the embedding vectors (such as cosine similarity or Euclidean distance). Even if their original names are different, such as "Freq_Hz" and "VibrationFrequency", they will be divided into the same cluster because their embedding vectors are close in vector space.

[0046] Subsequently, for each formed field cluster, dynamic labeling rules are automatically or with minimal human intervention generated, which define how to map all original data fields within the cluster to a unified semantic meaning, laying the foundation for subsequent automated semantic alignment and unification.

[0047] Finally, the dynamic labeling rules are applied to map the original data fields to standard concepts and attributes in the predefined vibration compaction domain ontology. The vibration compaction domain ontology is a shared knowledge framework pre-built for vibration compaction professionals, and standard concepts such as "compaction machinery type", "soil material", "vibration parameter, etc. Key categories or things. For the standard concept of "vibration parameter category", its attributes may include "parameter name", "parameter value", "standard unit", etc. When processing an "original data field" like "Freq_Hz", it corresponds to the "working frequency" attribute under the "vibration parameter category" standard concept in the ontology, and its unit should be hertz and its data type should be numerical.

[0048] During the mapping process, consistency checks are performed on the results of semantic labeling according to the attribute constraints defined in the ontology, including data type restrictions, numerical range, hierarchical relationships, and mutual exclusion relationships, etc. Only when all labeling results meet the consistency and logicality requirements, they are included in the final output semantic enhanced dataset.

[0049] The dataset generated after verification contains original data and standardized semantic labels, providing high-quality input for subsequent knowledge graph model construction. Using a fine-tuned pre-trained model to generate field embedding, combined with a graph clustering algorithm to establish dynamic labeling rules, adaptive labeling of multi-source heterogeneous data is achieved. Compared with the template matching method, this method can better handle data diversity, non-standard naming, and term complexity issues, improve the degree of automation and accuracy of labeling, and help build an accurate and reliable vibration compaction knowledge graph model, and improve the intelligent level of evaluation and optimization methods.

[0050] Based on the semantic enhanced dataset, a domain enhanced GNN algorithm is applied for node relation extraction, and a first graph algorithm is used for incremental updating to build and dynamically maintain a knowledge graph model; the knowledge graph model is used for structured representation of nodes and relations in the vibration compaction process.

[0051] Further, as shown in Figure 2 The process of entity relation extraction includes:

[0052] First, the semantic enhanced dataset is converted into an initial graph structure. The initial graph structure includes:

[0053] The nodes are used to represent the key data elements identified in the semantic enhanced dataset. The nodes can represent: a specific type of compaction equipment (such as "20-ton single steel wheel vibration roller of model XXX"); construction parameter values (such as "vibration frequency 35Hz", "rolling speed 2.5km / h").

[0054] And the relationships represent the established semantic classification information and preliminary structural links between key data elements. The relationships can represent: preliminary semantic relationships formed based on semantic labels in the semantic enhanced data, such as the "XP303 roller" node pointing to the class node in the ontology through the relationship; and the preliminary structural links of the data source itself, for example, the inherent data association between the geometric model, material constitutive parameters and stress-strain calculation results in CAE simulation data.

[0055] Next, a GNN algorithm (such as GraphSAGE or GAT) pre-trained on a dataset in the target engineering field (such as road engineering and geotechnical engineering) is used to process the initial graph structure. This includes: distributed representation learning of the nodes in the initial graph structure to obtain their node feature vectors, which integrate the attributes of the nodes themselves and the semantic information of their structural neighborhoods. Based on these learned node feature vectors, key engineering entities (such as roadbed structural units, compaction equipment and key construction parameters) that are important for vibration compaction process analysis are further identified; then, based on the key engineering entities, the relationship classification is performed, and the relationship types are mapped to the pre-defined semantic relationship types in the domain ontology, for example, the roadbed level contains the filler type, the compaction equipment uses the vibration parameter, and the structured knowledge triple is formed, each triple is in the form of [head entity, relationship type, tail entity];

[0056] Finally, the structured knowledge triplets are checked for validity against a predefined ontology of the vibration compaction domain, which defines constraints for each entity type and relation type, e.g., a certain relation type can only be established between specific types of entities. Only when the triplets satisfy the above semantic reasonableness and structural consistency requirements, the checked structured knowledge triplets are output for subsequent use in incremental updating or initial construction of the knowledge graph model.

[0057] By converting the semantic enhanced dataset into an initial graph structure, the entity representation learning and relation classification are carried out using the domain-enhanced GNN algorithm model, which accurately extracts deep entities and complex semantic relations from semi-structured or structured data. After pre-training and domain ontology verification, the accuracy of the extraction results is further guaranteed, providing technical support for constructing a high-quality vibration compaction domain knowledge graph model.

[0058] Further, the process of constructing and dynamically maintaining the knowledge graph model using the first graph algorithm for incremental updating includes:

[0059] First, when receiving newly added semantic enhanced data, a first graph algorithm, i.e., a Temporal Graph Network (TGN) model, is used to generate temporal embedding representations for entity interaction events in the newly added semantic enhanced data. For example, the event of a specific type of compaction equipment, "vibratory roller", performing its third rolling operation on the A2 layer roadbed filler of the K10+200 to K10+300 section of the high-grade highway, will be encoded as an embedding representation with temporal context. This representation not only reflects the core features of the event itself, but also integrates the temporal information of the event occurrence, enhancing the model's understanding of dynamic evolution processes.

[0060] Subsequently, in combination with a lightweight graph neural network (Lightweight GNN), new structured knowledge triplets are extracted from the temporal embedding representations. These triplets represent new knowledge fragments extracted from the latest semantic enhanced data that need to be integrated into the main knowledge graph model, logically forming a local incremental graph, i.e., a small temporary graph structure containing the latest dynamic information.

[0061] Finally, the change part of integrating the local new graph into the knowledge graph model is identified by a graph difference detection algorithm (such as Graph Edit Distance), focusing on comparing the subgraph area highly related to the new information topic. To ensure that the new triplets can be integrated without conflict with the existing graph structure, update operations are performed on the corresponding change part of the knowledge graph model; the update operations include adding, modifying and deleting nodes and relationships, while performing effectiveness verification based on the predefined vibration compaction domain ontology, ensuring that all integrated information is semantically consistent with the domain knowledge specification and the overall structure is reasonable and usable.

[0062] By using the first graph algorithm to process new dynamic data, using a lightweight graph neural network to extract new knowledge, and using a graph difference detection algorithm for targeted incremental update, efficient dynamic maintenance of the vibration compaction domain knowledge graph model is achieved. Only the change part of the model is updated, reducing the computational overhead and improving the update efficiency. At the same time, the domain ontology is continuously verified during the update process to ensure the quality and consistency of the model in dynamic evolution, providing a reliable and up-to-date knowledge base for upper-layer applications that rely on real-time analysis and decision-making based on the model.

[0063] Based on the knowledge graph model, a second graph algorithm is used to evaluate the parameter influence result; and based on the node embedding features of the knowledge graph model, a time series prediction algorithm is used to generate a risk prediction result; according to the parameter influence result and the risk prediction result, a reward function is defined, and a deep reinforcement learning algorithm is used to determine the optimal vibration compaction parameter combination.

[0064] Further, a second graph algorithm is used, also a Temporal Graph Network (TGN) model, to predict the future impact of a specific input (candidate parameters). The process of evaluating the parameter influence result includes:

[0065] First, the current compaction working condition information is obtained, including the material type used in the construction area, the moisture content of the material, the number of compaction passes already completed, and the environmental temperature on site, etc. These working condition data can be extracted from the existing knowledge graph model or input by the user on site.

[0066] Under the premise of knowing the current compaction working condition, a plurality of candidate vibration compaction parameter combinations are constructed, and the parameter content usually includes different vibration frequencies, amplitudes, and rolling speeds, etc. Subsequently, information related to the above working conditions and candidate parameters is extracted from the knowledge graph model. In order to extract effective knowledge structure fragments, a graph segmentation algorithm (such as Louvain community discovery algorithm) is used to divide the knowledge graph model, and a dynamic context subgraph is extracted from it. This dynamic context subgraph not only contains entity information directly related to the current working condition and candidate parameters, such as the type and properties of specific materials and compaction equipment, but also contains the evolution state and historical interaction process of these entities in the time dimension, reflecting their behavior characteristics over time.

[0067] Then, the dynamic context subgraph is input into a second pre-trained graph algorithm. The network has the ability to capture the spatio-temporal dependence and interaction mode of nodes and relationships in the graph, and can identify the complex influence mechanism of parameter changes on the material compaction process. Through reasoning and outputting parameter influence results, the parameter influence results are the quantitative representation of the candidate vibration compaction parameters on the pre-defined compaction effect indicators under the current compaction working condition.

[0068] These compaction effect indicators include but are not limited to: the next pass density increment, the uniformity coefficient of stress distribution, the effective utilization rate of energy, and the time required to reach the target density, etc. By calculating the parameter influence results, key reference basis is provided for the subsequent parameter selection and optimization process.

[0069] According to the working condition, a dynamic context subgraph is constructed, and a second graph algorithm is combined to deeply reason the parameter utility, achieving dynamic prediction of the consequences of vibration compaction parameters. It can fully utilize the context semantic and time evolution information in the knowledge graph, improve the prediction fidelity and applicability, and provide high-quality input and state transition prediction basis for vibration compaction parameter optimization relying on reinforcement learning and other means, thereby improving the accuracy of vibration compaction parameter selection.

[0070] Further, the process of generating a risk prediction result includes:

[0071] First, according to the current compaction working condition and the candidate vibration compaction parameters, relevant information is obtained. The current compaction working condition usually includes the type of material in the construction area, the moisture content, the compaction pass number, the environmental temperature, etc.; the candidate vibration compaction parameters include different combinations of vibration frequencies, amplitudes, and rolling speeds, etc. Based on the above input information, graph structure information related to risk prediction is extracted from the dynamically maintained knowledge graph model.

[0072] To achieve this goal, a dynamic context subgraph is extracted from the knowledge graph model using a graph partitioning algorithm. This subgraph contains entities and their attributes related to risk events, such as typical working condition instances of historical risk occurrence, material sensitivity parameters, and equipment limit state records, providing a structured foundation for subsequent risk analysis.

[0073] Then, a node embedding algorithm is applied to generate node embedding vector representations for the nodes in the dynamic context subgraph. The node embedding algorithm used can be Node2Vec or GraphSAGE, which maps each node in the subgraph to a low-dimensional real vector. Through the analysis of node neighborhood structure, connection patterns, and semantic information, the vector embedding effectively captures the position of the node in the graph and its potential association with other entities, forming discriminative node embedding features.

[0074] Subsequently, the node embedding vector representations are input as node embedding features into a pre-trained time series prediction algorithm, and the risk prediction results corresponding to the candidate vibration compaction parameters are output. The model can be a classification model or a regression model, including but not limited to logistic regression, support vector machine, or gradient boosting tree (such as XGBoost). The model aims to predict potential risks during the compaction process, and can identify operation risks such as "jumping" risk, "particle breakage" level, probability or level of "equipment load abnormality" occurrence, etc. After receiving the input node embedding features, the model performs inference and outputs risk prediction results corresponding to each group of candidate vibration compaction parameters. The output risk prediction results can be probability values, risk scores, or risk level labels, reflecting the potential operation risk level of different parameter combinations under the current working condition.

[0075] By combining the semantic expression ability of the knowledge graph, the structure perception ability of the node embedding algorithm, and the prediction ability of the time series prediction algorithm, the potential operation risks in complex compaction working conditions can be accurately identified, providing support for subsequent risk avoidance and intelligent optimization decisions of vibration compaction parameters, thereby improving the accuracy of vibration compaction parameter selection and optimizing energy transfer efficiency and engineering quality.

[0076] Further, as shown in Figure 3 the process of determining the optimal vibration compaction parameter combination using a deep reinforcement learning (DRL) algorithm includes:

[0077] First, a deep reinforcement learning environment is constructed, which includes the definition of the following core elements:

[0078] Define the state of the agent, which is the information observed by the agent about the environment at each decision-making moment. The state contains multiple dimensions of data input, including: current compaction working condition features extracted from the knowledge graph model, such as construction area number, current compaction layer thickness, completed pass number, environmental temperature and humidity, etc.; material properties, such as soil type, particle size distribution, water content, target dry density, etc.; equipment status, such as roller model, current vibration frequency, amplitude, travel speed, etc.; parameter influence results, which are the predicted values of compaction quality (such as density, uniformity) based on the aforementioned parameters; risk prediction results, which are the probabilities or risk levels of candidate parameter combinations causing operation risks such as vibration jump, particle breakage, and equipment overload under the current working condition.

[0079] These multi-dimensional information together constitute the basis for DRL agent decision-making.

[0080] Define the action of the agent, which defines the actions that the agent can execute, mainly including selecting the next step of vibration compaction parameter combination. The parameter combination can include vibration frequency, amplitude, rolling speed, whether to continue compaction, etc. The action space can be set to discrete (select from pre-set parameter combinations) or continuous (dynamically adjust parameter values within a specified range).

[0081] Define the reward function of the agent, build a reward function to guide learning behavior, which considers factors such as compaction quality, energy consumption, equipment wear and tear, and operation risk, including: compaction quality improvement benefits, such as density increase value, uniformity improvement, etc., giving positive rewards; energy consumption and equipment wear and tear penalties, based on the estimated fuel or power consumption of the selected parameters, as well as equipment wear and tear, life reduction, etc., applying moderate negative penalties; risk factor penalties, for behaviors that may cause operation risk indicators (such as vibration jump probability and particle breakage rate) to exceed safety thresholds, applying negative penalties. The reward function can use weighted summation or other function forms to integrate the above indicators, ultimately generating a scalar reward value reflecting the overall merits and demerits of the current action.

[0082] After the environment is built, apply specific deep reinforcement learning algorithms for policy training, such as Proximal Policy Optimization (PPO), Deep Q Network (DQN), and their improved variants. Through a large number of "state-action-reward" cycle interactions in the simulation environment, the agent continuously updates its policy function, gradually learning to obtain a policy model that can output the optimal vibration compaction parameter combination under different compaction working conditions.

[0083] By modeling the vibration compaction parameter optimization problem as a deep reinforcement learning task, the adaptive and globally optimal determination mechanism of vibration compaction parameters is realized by comprehensively utilizing the parameter influence evaluation and risk prediction information provided by the knowledge graph model in the constructed multi-dimensional fusion environment. The optimal strategy selection can be achieved in multi-objective trade-off, improving the accuracy of vibration compaction parameter selection and optimizing the energy transfer efficiency and engineering quality.

[0084] Based on the optimal vibration compaction parameter combination and the knowledge graph path generation parameter recommendation report, the system collects the adoption of the recommended scheme by engineers, actual construction effect and other feedback information, which is used for continuous iterative optimization of the content of the knowledge graph model, forming a closed-loop learning improvement mechanism.

[0085] By constructing and dynamically maintaining a vibration compaction field knowledge graph model integrating multi-source heterogeneous data including computer-aided engineering analysis, and using neural networks, first graph algorithms and deep reinforcement learning algorithms, the limitations of traditional vibration compaction methods in data integration, process understanding, parameter optimization and risk control are overcome. Not only the precise evaluation and prediction of dynamic influence and potential construction risks of compaction parameters are realized, but also the multi-objective optimal process parameter combination considering energy transfer efficiency, compaction quality and operation safety is automatically optimized. Finally, by providing interpretable decision support and establishing a continuous learning feedback mechanism, the accuracy of vibration compaction parameter selection is improved, and the energy transfer efficiency and engineering quality are optimized.

[0086] Embodiment two:

[0087] Embodiment two is based on embodiment one, further showing the application performance of the present application in engineering scenarios, ensuring that high energy transfer efficiency is still achieved under complex working conditions. A vibration compaction forming energy transfer efficiency evaluation method, comprising:

[0088] Integrating multi-source heterogeneous vibration compaction data, generating a semantic enhanced dataset through an adaptive semantic labeling algorithm based on machine learning algorithm and graph clustering algorithm;

[0089] Based on the semantic enhanced dataset, applying a domain enhanced GNN algorithm for node relationship extraction, and using a first graph algorithm for incremental update to build and dynamically maintain a knowledge graph model; the knowledge graph model is used to structurally represent nodes and relationships in the vibration compaction process;

[0090] Based on the knowledge graph model, using a second graph algorithm to evaluate parameter influence results; and based on the node embedding features of the knowledge graph model, using a time series prediction algorithm to generate risk prediction results; defining a reward function according to the parameter influence results and the risk prediction results, and using a deep reinforcement learning algorithm to determine the optimal vibration compaction parameter combination;

[0091] Based on the optimal vibration compaction parameter combination and the knowledge graph path, a parameter recommendation report is generated, and user feedback is collected to iteratively optimize the knowledge graph model.

[0092] For comparison, this embodiment considers a common type of subgrade filler (Group A graded gravel) and sets a clear compaction target (average compaction degree reaches 95%, and compaction degree variation coefficient is less than 4%). The method of the present application is compared with two baseline methods:

[0093] Baseline method one: traditional empirical parameter method. According to typical construction manuals and operator experience, fixed vibration frequency, amplitude and rolling speed are set.

[0094] Baseline method two: optimization algorithm based on simplified model (standard genetic algorithm). This method is based on a simplified filler compaction model (which does not include complex knowledge graph and real-time fusion of multi-source data), and uses genetic algorithm to optimize compaction parameters offline.

[0095] All methods are run under the same simulation conditions and data input conditions to ensure fairness of comparison. As shown in Table 1, the performance indicators and overall efficiency of the vibration compaction process are shown. Among them, the average energy transfer efficiency directly reflects the energy utilization effectiveness, and the method of the present application is expected to achieve a significantly higher percentage, which indicates that it can more effectively use the energy output by the compaction equipment to improve material density through intelligent parameter optimization, thereby reducing energy waste; at the same time, the final average compaction degree is used to measure the compaction quality, and the present application not only more easily stabilizes and moderately exceeds the target value of engineering design; the compaction degree standard deviation reflects the compaction uniformity, and the smaller the value, the better the uniformity, the present application can improve the uniformity of compaction work and avoid engineering quality defects caused by local overpressure or underpressure, thereby fully guaranteeing the overall performance and long-term stability of the compaction structure. In addition, thanks to higher energy transfer efficiency and more precise parameter control strategy, the present application also shows advantages in terms of average operation time to reach the target compaction degree, and can complete the compaction task more quickly, thereby improving overall construction efficiency and economic benefits.

[0096] Table 1 Comparison of compaction performance and efficiency

[0097] Performance indicator Baseline method one Baseline method two Inventive method Average energy transfer efficiency (%) 62 70 83 Final average compaction (%) 94.2 95.1 95.8 Compaction standard deviation 1.5 1.1 0.7 Average job length to achieve target compaction (hours / 100m 2 )]]> 5.5 4.8 3.9

[0098] To further verify the effect of the present application, a specific working condition is simulated, the type of roadbed material is set to medium-grained graded gravel, the initial moisture content is 8%, the thickness of the compacted layer is 30 cm, and the target compaction degree reaches 95%. As shown in Table 2, the compaction performance and efficiency are compared. It can be seen that the method of the present application can recommend slightly different frequencies and amplitudes through its intelligent optimization decision according to the current working condition (material, target, etc., which has been integrated into the knowledge graph model), and through a more optimal number of times and speed combination, the expected total energy consumption is also lower than that of other baseline methods while achieving a better compaction effect.

[0099] Table 2 Comparison of compaction parameters and performance results of different methods under specific working conditions

[0100]

[0101] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating energy transmission efficiency of a vibratory compaction molding, characterized by, The method comprises the following steps: Integrating multi-source heterogeneous vibration compaction data, generating a semantic enhanced dataset through an adaptive semantic labeling algorithm based on machine learning algorithm and graph clustering algorithm; Based on the semantic enhanced dataset, applying a domain enhanced GNN algorithm for node relationship extraction, and using a first graph algorithm to perform incremental update to build and dynamically maintain a knowledge graph model; The knowledge graph model is used to structurally represent nodes and relationships in the vibration compaction process; Based on the knowledge graph model, using a second graph algorithm to evaluate parameter influence results, and using a time series prediction algorithm to generate risk prediction results based on the node embedding features of the knowledge graph model; defining a reward function according to the parameter influence results and the risk prediction results, and using a deep reinforcement learning algorithm to determine the optimal vibration compaction parameter combination; Based on the optimal vibration compaction parameter combination and the knowledge graph path, a parameter recommendation report is generated, and user feedback is collected to iteratively optimize the knowledge graph model.

2. The method of claim 1, wherein, The implementation process of the adaptive semantic labeling algorithm comprises: Using a machine learning algorithm fine-tuned on a target engineering field dataset to generate a field embedding vector for each original data field in the multi-source heterogeneous vibration compaction data; Inputting the field embedding vector into a graph clustering algorithm, which clusters the original data fields according to the similarity of the embedding vectors and establishes dynamic labeling rules for each cluster; Applying the dynamic labeling rules to map the original data fields to standard concepts and attributes in the predefined vibration compaction domain ontology, and performing consistency verification on the semantic labels formed by the mapping through the vibration compaction domain ontology to generate the semantic enhanced dataset.

3. The method of claim 1, wherein, The process of node relationship extraction comprises: Converting the semantic enhanced dataset into an initial graph structure, which represents the identified key data elements in the semantic enhanced dataset as nodes and represents the established semantic classification information and preliminary structural links between key data elements as relationships; Using a GNN algorithm pre-trained on a target engineering field dataset to perform distributed representation learning on the nodes in the initial graph structure to identify key engineering entities; based on the key engineering entities, performing relationship classification to form structured knowledge triples; Performing effectiveness verification on the structured knowledge triples and the predefined vibration compaction domain ontology to output the verified structured knowledge triples.

4. The method of claim 1, wherein, The process of using a first graph algorithm to perform incremental update to build and dynamically maintain a knowledge graph model comprises: When receiving newly added semantic enhanced data, using a first graph algorithm to generate a time series embedding representation for entity interaction events in the newly added semantic enhanced data; Combining a lightweight graph neural network to extract newly added structured knowledge triples from the time series embedding representation to form a local newly added graph; By the graph difference detection algorithm, the change part of integrating the local new graph into the knowledge graph model is identified, and an update operation is performed on the corresponding change part of the knowledge graph model; the update operation includes adding, modifying and deleting nodes and relationships, and simultaneously performing validity verification based on a predefined vibration compaction field ontology.

5. The method of claim 1, wherein, The process of evaluating the parameter influence result using the second graph algorithm includes: according to the current compaction working condition and the candidate vibration compaction parameter, extracting a dynamic context subgraph from the knowledge graph model using a graph segmentation algorithm; inputting the dynamic context subgraph into a pre-trained second graph algorithm; the second graph algorithm is used to capture the spatio-temporal dependency and node interaction mode in the dynamic context subgraph, reason and output the parameter influence result, and the parameter influence result is a quantitative representation of the candidate vibration compaction parameter on the pre-defined compaction effect index.

6. The method of claim 1, wherein, The process of generating a risk prediction result includes: according to the current compaction working condition and the candidate vibration compaction parameter, extracting a dynamic context subgraph from the knowledge graph model using a graph segmentation algorithm; applying a node embedding algorithm to generate a node embedding vector representation for the nodes in the dynamic context subgraph, and inputting the node embedding vector representation as a node embedding feature into a pre-trained time series prediction algorithm to output the risk prediction result corresponding to the candidate vibration compaction parameter.

7. The method of claim 1, wherein, The process of determining the optimal vibration compaction parameter combination using a deep reinforcement learning algorithm includes: Constructing a deep reinforcement learning environment, including: defining the state of the agent, including the current compaction working condition features, material properties, device state, parameter influence result and risk prediction result extracted from the knowledge graph model; defining the action of the agent, including selecting the next set of vibration compaction parameters; defining the reward function of the agent, including: the expected compaction quality improvement amount evaluated according to the parameter influence result, the expected energy consumption and device wear inferred from the knowledge graph model, and the risk exceeding the safety threshold according to the risk prediction result; Based on the deep reinforcement learning environment, a deep reinforcement learning algorithm is used to train a parameter selection strategy that can maximize the long-term cumulative expected reward, and the output of the parameter selection strategy is the optimal vibration compaction parameter combination.

Images