A Dynamic Optimization Method for Communication Engineering Construction Based on Multidimensional Perception

Abstract

This invention discloses a dynamic optimization method for communication engineering construction based on multi-dimensional sensing, comprising: real-time acquisition of multi-dimensional sensing data from the construction site through a multi-source heterogeneous sensor network, including environmental parameters, equipment operating status, construction progress, and personnel behavior data; spatiotemporal alignment processing of the multi-dimensional sensing data to construct a spatiotemporally correlated dynamic construction digital twin; establishing a dynamic optimization model based on a reinforcement learning algorithm, with the objective optimization function being maximizing construction efficiency, minimizing resource consumption, and minimizing safety risks, and embedding risk constraints; inputting the dynamic construction digital twin into the dynamic optimization model, and outputting a multi-dimensional parameter optimization instruction set including equipment scheduling parameters, construction path parameters, and resource configuration parameters; and performing real-time performance evaluation of the multi-dimensional parameter optimization instruction set through edge computing nodes, dynamically adjusting the model's weight parameters to form a closed-loop feedback control. This invention achieves the goal of synergistic improvement in construction efficiency, safety, and economy.

Description

Technical Field

[0001] This invention relates to the field of communication engineering technology, and specifically to a dynamic optimization method for communication engineering construction based on multi-dimensional perception. Background Technology

[0002] With the large-scale deployment of 5G networks and communication infrastructure, communication engineering construction faces severe challenges from complex and dynamic environments. Traditional construction optimization methods mainly rely on manual experience and static planning models, which have the following technical bottlenecks:

[0003] Single data perception dimension: Existing technologies mostly use a single type of sensor (such as position or temperature and humidity sensor), lacking the ability to perceive equipment operating status, personnel behavior and environmental parameters in a multi-dimensional real-time manner, resulting in a serious data silo phenomenon;

[0004] Insufficient dynamic response capability: Traditional static planning models cannot adapt to real-time changes at the construction site (such as sudden equipment failures or sudden weather changes), making it difficult to achieve dynamic balance optimization of construction efficiency, resource consumption, and safety risks.

[0005] Risk prediction lag: Existing safety early warning systems are mostly based on threshold triggering mechanisms and lack the ability to model the multi-dimensional risk transmission chain of equipment-personnel-environment, resulting in delayed risk response;

[0006] Lack of collaborative control: The resource scheduling, path planning and safety control modules operate independently, lacking a global collaborative mechanism, resulting in insufficient improvement in construction efficiency.

[0007] While some research has attempted to introduce digital twin technology, it generally suffers from low accuracy in physical simulation and lag in dynamic model updates, making it difficult to meet the high-precision construction requirements of communication engineering. Furthermore, rule-based optimization methods are prone to policy oscillations in complex scenarios, severely impacting construction continuity. Summary of the Invention

[0008] To overcome the shortcomings of existing technologies, this invention provides a dynamic optimization method for communication engineering construction based on multi-dimensional perception, which is used to solve the problems of low efficiency, resource waste and safety hazards caused by complex and dynamic environments in communication engineering construction, thereby achieving the goal of synergistic improvement in construction efficiency, safety and economy.

[0009] To solve the above problems, the technical solution adopted by the present invention is as follows:

[0010] A dynamic optimization method for communication engineering construction based on multi-dimensional sensing includes the following steps:

[0011] Multi-dimensional sensing data of the construction site are collected in real time through a multi-source heterogeneous sensor network, including environmental parameters, equipment operating status, construction progress and personnel behavior data;

[0012] Spatiotemporal alignment processing is performed on multidimensional sensing data to construct a dynamic construction digital twin with spatiotemporal correlation;

[0013] A dynamic optimization model is established based on reinforcement learning algorithm, with the objective optimization function being to maximize construction efficiency, minimize resource consumption and safety risks, and risk constraints are embedded.

[0014] Input the dynamic construction digital twin into the dynamic optimization model and output a multi-dimensional parameter optimization instruction set that includes equipment scheduling parameters, construction path parameters, and resource configuration parameters;

[0015] By using edge computing nodes to perform real-time performance evaluation of the multi-dimensional parameter optimization instruction set, the weight parameters of the dynamic optimization model are dynamically adjusted to form a closed-loop feedback control.

[0016] In a preferred embodiment of the present invention, the multi-source heterogeneous sensor network includes:

[0017] Vibration sensors and GNSS positioning modules deployed on construction machinery;

[0018] Temperature and humidity sensors, noise sensors, and image acquisition devices are installed at the construction site.

[0019] Wearable positioning terminal and physiological parameter monitoring equipment for construction workers;

[0020] RFID tracking system for material transport vehicles.

[0021] As a preferred embodiment of the present invention, constructing a spatiotemporally correlated dynamic construction digital twin includes:

[0022] The multidimensional sensing data is spatiotemporally calibrated by improving the Kalman filter algorithm to eliminate clock deviations and spatial coordinate system differences between devices.

[0023] Establish a multi-dimensional association rule base for equipment, personnel, and environment based on knowledge graphs;

[0024] Generate physically accurate 3D visualizations of construction scenes using a digital twin engine;

[0025] The improved Kalman filter algorithm incorporates an adaptive noise covariance matrix that includes the variances of environmental noise, equipment noise, and human noise.

[0026] As a preferred embodiment of the present invention, the establishment of a multi-dimensional association rule base for equipment-personnel-environment includes:

[0027] Define the entity types and associated attributes of the construction scene based on dynamic ontology modeling technology;

[0028] An improved Apriori algorithm is used to mine frequent association patterns in multi-source heterogeneous data, and the confidence of association rules is dynamically adjusted by a time decay factor.

[0029] A multidimensional association tensor of equipment, personnel, and environment is constructed, implicit association rules are extracted through tensor decomposition algorithm, and the rule weights are dynamically adjusted based on fuzzy logic controller;

[0030] A real-time rule conflict detection mechanism is deployed, and the third-order gradient descent method is used to eliminate logical contradictions between equipment scheduling instructions and environmental constraints, generating a multi-dimensional association rule instance library with spatiotemporal labels.

[0031] In a preferred embodiment of the present invention, generating a three-dimensional visualized construction scene includes:

[0032] A hybrid physics engine based on the coupling of discrete element method and finite element method is constructed to perform multi-field joint simulation of construction machinery motion trajectory, material stress distribution and geological deformation;

[0033] The diameter of the discrete element particle is determined by the maximum speed of the device, the simulation step size, and the number of particle contact pairs per step.

[0034] In a preferred embodiment of the present invention, the dynamic optimization model is established by including:

[0035] Construct a hierarchical reinforcement learning architecture that includes an upper decision-making layer, a lower execution layer, and a risk prediction layer;

[0036] Among them, the upper decision-making layer uses a deep deterministic strategy gradient algorithm to generate a global optimization strategy;

[0037] The lower layer implements the optimal allocation of local resources based on the Q-learning algorithm;

[0038] The risk prediction layer integrates a Bayesian network to assess construction safety risk values ​​in real time.

[0039] The deep deterministic policy gradient algorithm adopts a dual-delay deep deterministic policy gradient architecture, with a target network smoothing coefficient τ=0.005.

[0040] In a preferred embodiment of the present invention, achieving optimal allocation of local resources includes:

[0041] A dynamic multidimensional state space is constructed based on a variational autoencoder, and the vibration spectrum of construction machinery, the topology of material transportation path and fault early warning parameters are mapped as an orthogonal basis for state vectors.

[0042] Design a multi-objective reward function that integrates resource utilization, task delay, and security risks;

[0043] Deploy a dual-memory pool experience playback mechanism: short-term storage of device vibration spectrum characteristics, long-term storage of environmental noise data, and use priority sampling for training;

[0044] A dynamic ε-greedy strategy is introduced to adaptively adjust the exploration probability based on the construction progress.

[0045] By updating the Q-value matrix in real time through edge computing nodes, a spatiotemporal constraint instruction set is generated by triggering Shapley value game equilibrium when there are resource allocation conflicts among multiple devices.

[0046] As a preferred embodiment of the present invention, the real-time assessment of construction safety risk values ​​includes:

[0047] A dynamic topological Bayesian network is constructed to map equipment vibration spectrum, personnel positioning offset and meteorological parameters into multidimensional evidence variables, and a risk transmission chain of equipment-personnel-environment is established through a causal inference engine;

[0048] The conditional probability table is updated in real time based on the sliding time window data stream, and the variational Bayesian inference algorithm is used to dynamically optimize the dependency strength weights between nodes.

[0049] The deviation between the actual construction status and the safety baseline is quantified by KL divergence. When the joint probability of multiple risk nodes exceeds the threshold, emergency plan matching based on convolutional neural network is triggered.

[0050] Generate a spatiotemporal risk heat map, with edge computing nodes providing feedback on the risk gradient descent direction, and adjust the action selection probability distribution of the reinforcement learning policy network in real time.

[0051] In a preferred embodiment of the present invention, the generation of a multidimensional parameter optimization instruction set includes:

[0052] Based on the spatiotemporal correlation of risk heatmaps and the direction of risk gradient descent, a three-dimensional collaborative optimization model of geology, equipment, and personnel is constructed, and the safety risk entropy value is embedded into the reward function correction term of the Q-learning algorithm.

[0053] Spatial safety distance constraints are imposed on equipment scheduling parameters based on the risk transmission chain prediction results of dynamic topology Bayesian networks.

[0054] As a preferred embodiment of the present invention, the real-time performance evaluation includes:

[0055] A dynamic performance simulation sandbox is built based on a digital twin engine, a multi-dimensional parameter optimization instruction set is loaded and real-time perception data stream is injected to generate three-dimensional performance evaluation indicators.

[0056] The performance of the multi-dimensional parameter optimization instruction set is verified by segmenting based on sliding time windows. Within each time window, the deviation between the actual construction trajectory and the predicted trajectory of the digital twin is calculated using grey relational analysis.

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0058] (1) High-precision dynamic perception and modeling

[0059] High-precision spatial positioning and temporal alignment are achieved through a multi-source heterogeneous sensor network (vibration, GNSS, RFID, etc.), which significantly improves accuracy compared to traditional single-source sensing solutions.

[0060] An improved Kalman filter algorithm was developed by introducing an adaptive noise covariance matrix (adjustable from 0.5 to 1.2), achieving an abnormal data identification rate of 94.3% under heavy machinery interference environments, a 7.2% improvement over the fixed parameter scheme.

[0061] The discrete-element-finite-element coupled physics engine supports sub-meter-level simulation of construction machinery trajectories, and dynamic LOD technology significantly improves rendering efficiency.

[0062] (2) Multi-objective collaborative optimization capability

[0063] A hierarchical reinforcement learning architecture (DDPG + Q-learning + Bayesian network) achieves three-dimensional collaborative optimization of construction efficiency, resource consumption, and safety risks.

[0064] Construction efficiency improved by 23.5% (experimental data), and resource consumption decreased by 15.8%.

[0065] Risk transmission chain modeling reduces the early warning time for safety incidents to within 30 seconds, improving the response speed by 6 times compared to traditional methods.

[0066] The dynamic ε-greedy strategy adaptively adjusts the exploration probability according to the construction progress, reducing the number of strategy oscillations to 1.1 times / hour, thus ensuring construction continuity.

[0067] (3) Closed-loop self-correction mechanism

[0068] Real-time performance evaluation of edge computing nodes is combined with a lightweight LSTM model to achieve online iteration of weight parameters (period ≤ 5 seconds) and dynamically adapt to environmental changes.

[0069] Q-learning correction driven by risk heatmap significantly improves the response speed of resource evacuation in high-risk areas and greatly reduces the violation rate of safety distance constraints;

[0070] The digital twin sandbox verification mechanism keeps the deviation of the construction trajectory within a safe threshold, greatly improving the stability of the optimization effect compared to schemes without closed-loop control.

[0071] Experimental comparison data

[0072]

[0073] Note: The experimental data comes from a base station construction project and is the result of a test run of 72 hours.

[0074] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0075] Figure 1 A flowchart illustrating the steps of a dynamic optimization method for communication engineering construction based on multi-dimensional sensing;

[0076] Figure 2 A flowchart for constructing a dynamic construction digital twin with spatiotemporal correlation;

[0077] Figure 3 This is a logic diagram for the dynamic optimization model. Detailed Implementation

[0078] See Figure 1 The dynamic optimization method for communication engineering construction based on multi-dimensional perception provided by this invention includes the following steps:

[0079] Step S1: Collect multi-dimensional sensing data of the construction site in real time through a multi-source heterogeneous sensor network. The multi-dimensional sensing data includes environmental parameter data, equipment operating status data, construction progress data, and personnel behavior data.

[0080] Step S2: Perform spatiotemporal alignment processing on the multidimensional sensing data to construct a spatiotemporally correlated dynamic construction digital twin;

[0081] Step S3: Establish a dynamic optimization model based on reinforcement learning algorithm. The dynamic optimization model takes maximizing construction efficiency, minimizing resource consumption and safety risks as the objective optimization function, and embeds risk constraints.

[0082] Step S4: Input the dynamic construction digital twin into the dynamic optimization model to generate a multi-dimensional parameter optimization instruction set, which includes a collaborative optimization scheme for equipment scheduling parameters, construction path parameters, and resource configuration parameters.

[0083] Step S5: Real-time performance evaluation of the multi-dimensional parameter optimization instruction set is performed through edge computing nodes, and the weight parameters of the dynamic optimization model are dynamically adjusted according to the evaluation results to form a closed-loop feedback control.

[0084] In step S1 above, the multi-source heterogeneous sensor network includes:

[0085] Vibration sensors and GNSS positioning modules deployed on construction machinery;

[0086] Temperature and humidity sensors, noise sensors, and image acquisition devices are installed at the construction site.

[0087] Wearable positioning terminal and physiological parameter monitoring equipment for construction workers;

[0088] RFID tracking system for material transport vehicles.

[0089] Specifically, environmental parameter data refers to physical environmental indicators within the construction site that affect construction safety and efficiency, including meteorological parameters and spatial state parameters.

[0090] The meteorological parameters include: temperature and humidity data (collected by temperature and humidity sensors), noise data (collected by noise sensors), wind speed, and rainfall (collected by meteorological sensors).

[0091] Spatial state parameters include: geological deformation data (such as soil moisture and geological hardness).

[0092] Equipment operating status data refers to the real-time performance indicators of construction machinery and transportation equipment, including: machinery body parameters, transportation status parameters, and fault warning parameters.

[0093] The mechanical body parameters include: vibration spectrum (collected by vibration sensors, including frequency and amplitude), positioning coordinates and motion trajectory (collected by GNSS positioning module).

[0094] Transportation status parameters include: material transportation path topology (RFID tracking system records the real-time location and path planning of vehicles) and maximum equipment speed.

[0095] Fault warning parameters include: abnormal equipment temperature (collected by temperature sensor) and hydraulic system pressure fluctuation (comparison of pressure sensor data with historical baseline).

[0096] Construction progress data refers to quantitative indicators that reflect the deviation between the actual progress of a project and the plan, including: time dimension parameters, resource consumption parameters, and spatial schedule parameters.

[0097] The time dimension parameters include: task completion timestamp (generated by spatiotemporal alignment of BIM model and GNSS positioning data) and process delay duration.

[0098] Resource consumption parameters include: material usage (RFID system tracks the unloading frequency of transport vehicles and material consumption) and cumulative equipment runtime (GNSS positioning data and equipment start-up and shutdown log correlation analysis).

[0099] Spatial progress parameters include: the completed volume of the structure (calculated by fitting discrete element simulation and image point cloud data), and the geological excavation depth (comparison of GNSS elevation data and BIM model).

[0100] Personnel behavior data refers to the work status and safety-related indicators of construction workers, including: physiological parameters, behavioral trajectory parameters, safety status parameters, and interactive behavior parameters.

[0101] Among them, physiological parameters include: heart rate (collected in real time by wearable physiological monitoring devices).

[0102] The behavioral trajectory parameters include: personnel location coordinates, personnel location offset (collected by wearable terminal), and personnel movement speed (obtained through location data).

[0103] Safety status parameters include: safety equipment wearing data (such as the percentage of time without safety equipment) and fatigue index (calculated by combining heart rate variability (HRV) and continuous working time).

[0104] Interactive behavior parameters include: device operation data (such as operation frequency, device operation events and their intervals (such as start, stop, adjustment, etc.)).

[0105] See Figure 2 In step S2 above, constructing a spatiotemporally correlated dynamic construction digital twin includes:

[0106] By improving the Kalman filter algorithm, spatiotemporal calibration of multidimensional sensing data is performed to eliminate clock skew between devices and differences in spatial coordinate systems.

[0107] A construction element association model is constructed based on knowledge graph technology, and a multi-dimensional association rule base for equipment, personnel, and environment is established.

[0108] Utilize a digital twin engine to generate physically accurate 3D visualized construction scenes;

[0109] The improved Kalman filter algorithm introduces an adaptive noise covariance matrix, as shown in Equation 1:

[0110] (1);

[0111] In the formula, Let covariance matrix be the variance matrix. This is an environmental dynamic coefficient, adjusted in real time based on the noise level at the construction site, with an adjustment range of 0.5-1.2. It is a diagonal matrix. Let Variance be the environmental noise. Let Variance be the variance of the equipment noise. The variance of human-generated noise.

[0112] Specifically, the variance of environmental noise Calculations are made using data from temperature, humidity, and noise sensors. = For noise weights, For noise variance, Temperature weighting, For temperature variance, For humidity weight, This represents the humidity variance.

[0113] Specifically, the variance of equipment noise Calculated from vibration sensor data (frequency, amplitude) and GNSS positioning error. Based on the amplitude sequence acquired by the vibration sensor. and frequency sequence Calculate the dynamic weighted vibration variance:

[0114] ;

[0115] In the formula, For dynamically weighted vibration variance, The amplitude sampling value, The average amplitude within the sliding time window. The vibration frequency at the corresponding moment. The reference frequency (e.g., the equipment's rated vibration frequency). This represents the total number of amplitude data points collected by the vibration sensor within the sliding time window.

[0116] Based on the positioning coordinate sequence of the GNSS module Calculate the variance of the 3D positioning error:

[0117] ;

[0118] In the formula, , , This refers to the positioning error in each direction; , , , These are the actual coordinates, which can be calibrated using a BIM model.

[0119] The variance of combined vibration and GNSS positioning error is determined by dynamic weighting coefficients. Weighted fusion:

[0120] ;

[0121] In the formula, It is dynamically adjusted according to the type of equipment.

[0122] Specifically, the variance of man-made noise Calculated from data such as personnel positioning offset and operation frequency, including: obtaining the variance of positioning offset, fluctuation of operation frequency, and degree of abnormality of safety status.

[0123] Among them, the deviation sequence based on personnel positioning coordinates and planned operation path Calculate the variance of the positioning offset:

[0124] ;

[0125] In the formula, The variance of the positioning offset. This refers to the total number of personnel location data points collected by the wearable positioning terminal within the sliding time window. , Locate the coordinates of the personnel. The coordinates are for the planned operation path.

[0126] Based on equipment operation interval time series Calculate the variance of operation frequency fluctuations:

[0127] ;

[0128] In the formula, The variance of the frequency of operations. This represents the average operation interval time. This refers to the number of time intervals between device operation events (such as start, stop, adjustment, etc.) within a sliding time window.

[0129] Based on the proportion of time safety equipment was not worn and the deviation of the fatigue index from the baseline value, the variance of the safety status anomaly was calculated:

[0130] ;

[0131] In the formula, For the variance of the anomaly degree of the safe state, The percentage of time safety equipment was not worn. The fatigue index deviates from the baseline value. , Weighting coefficients (e.g.) , ).

[0132] The variance of positioning offset, the variance of operation frequency fluctuation, and the variance of safety status anomaly are weighted and fused using dynamic weighting coefficients to generate the variance of artificial noise. :

[0133] ;

[0134] In the formula, The variance of human-caused noise. For dynamic weighting coefficients, For indexing, The variance of the corresponding positioning offset ( The weighting coefficients of ) The variance of the corresponding operation frequency ( The weighting coefficients of ) For the corresponding safety state anomaly variance ( The weighting coefficients of ) The corresponding positioning offset variance. Variance of corresponding operation frequency fluctuation The variance of the anomaly degree of the corresponding safety state.

[0135] Constraints: .

[0136] Dynamic weighting coefficients Optimize based on real-time construction scenarios, for example:

[0137] When frequent deviations from the path are detected ( (Raise), elevate ;

[0138] When the frequency of equipment operation is abnormal ( (Raise), elevate .

[0139] Specifically, a comparison of data fusion accuracy under different environmental dynamic coefficients:

[0140]

[0141] Test environment: 200 sets of sample data were collected at a construction site with heavy machinery interference, using adaptive testing. Compared to the fixed-parameter scheme, range adjustment improves spatial positioning accuracy by 34% and temporal alignment efficiency by 33%. Among these improvements... When the range is 0.5-1.2, the positioning error and timing alignment error are minimized, and the abnormal data recognition rate is highest.

[0142] Furthermore, when establishing a multi-dimensional association rule base for equipment, personnel, and environment, the following is included:

[0143] Based on dynamic ontology modeling technology, the entity types and associated attributes of the construction scenario are defined. The entity types include vibration-sensitive equipment, high-risk workers and weather-sensitive areas, and the associated attributes include spatial dependence, operation coupling coefficient and risk transmission factor.

[0144] An improved Apriori algorithm is used to mine frequent association patterns in multi-source heterogeneous data. The confidence of association rules is dynamically adjusted by a time decay factor. The calculation of the dynamic confidence threshold is shown in Formula 2.

[0145] (2);

[0146] In the formula, For dynamic confidence threshold, As the initial confidence level, The time decay coefficient, For data time intervals, This is the environmental sensitivity coefficient. and These are the actual environmental parameter values ​​and the predicted environmental parameter values, respectively. For the standard deviation of environmental parameters, It is a natural constant;

[0147] A multidimensional association tensor of equipment, personnel, and environment is constructed, implicit association rules are extracted through tensor decomposition algorithm, and the rule weights are dynamically adjusted based on fuzzy logic controller;

[0148] A real-time rule conflict detection mechanism is deployed, and the third-order gradient descent method is used to eliminate logical contradictions between equipment scheduling instructions and environmental constraints, generating a multi-dimensional association rule instance library with spatiotemporal labels.

[0149] Specifically, when the actual environmental parameters (such as temperature and noise) differ significantly from the predicted values, the numerator term... An increase in the absolute value of directly affects the adjustment of the confidence threshold:

[0150] Positive differences ( If the actual environmental value is higher than the prediction, the confidence threshold increases, and rules with higher confidence levels are required to be accepted.

[0151] negative difference ( If the actual environmental value is lower than the prediction, the confidence threshold is lowered, and the rule selection conditions are relaxed.

[0152] Environmental sensitivity coefficient The higher the value, the more significant the impact of environmental differences on the confidence level. For example, in meteorologically sensitive areas (such as rainstorm warnings and gale warnings). It automatically increases in size and responds quickly to sudden environmental changes.

[0153] Standard deviation normalization: Differences are normalized using the standard deviation of environmental parameters. Standardization is necessary to avoid deviations caused by different parameter units.

[0154] Over time, the confidence threshold of historical data gradually decreases. (Item), to ensure that the model relies more on recent data and adapts to dynamic construction scenarios.

[0155] This invention employs an improved Apriori algorithm to mine association rules from multi-source heterogeneous data, with a dynamic confidence threshold. Used for filtering frequent itemsets:

[0156] Threshold adaptation: When environmental changes (such as sudden strong noise) cause... When the confidence level is increased, only association rules with high confidence (such as "abnormal equipment vibration → safety risk") are retained;

[0157] Noise filtering: When environmental differences are small, This reduces the risk of fatigue by allowing more potential rules to participate in the analysis and uncover implicit correlations (such as "duration of stay + temperature and humidity → fatigue risk").

[0158] Example:

[0159] Assuming the actual temperature value at a certain moment Predicted value Standard deviation ,parameter ,but:

[0160] ;

[0161] Assumption , Minutes, substituting into formula 2:

[0162] ;

[0163] At this point, the confidence threshold is significantly increased, and the system only accepts rules with high confidence, thus avoiding misjudgments under high-temperature conditions.

[0164] Furthermore, when generating a 3D visualized construction scene, the following are included:

[0165] A hybrid physics engine based on the coupling of the discrete element method (DEM) and the finite element method (FEM) is constructed to perform multi-physics joint simulation of the motion trajectory of construction machinery, material stress distribution, and geological deformation. The particle diameter in the discrete element method is [not specified]. As shown in Formula 3:

[0166] (3);

[0167] In the formula, The maximum speed of the equipment. For simulating step size, The number of single-step particle contact pairs;

[0168] Specifically, simulation step size According to the construction progress (such as the duration of task delays) Dynamic adjustment. If progress lags behind, shorten the timeframe. To increase the simulation frequency (e.g., from 1 second to 0.5 seconds), the update speed of the digital twin can be accelerated.

[0169] Geological deformation data (such as soil moisture and geological hardness) affect the number of single-step particle contact pairs. For example, in high-humidity soils, particle viscosity increases. As the value increases, the geological hardness increases. A higher value indicates more particle contact.

[0170] A multi-level rendering pipeline is constructed using ray tracing and point cloud fusion technologies. GNSS positioning data and BIM models are spatially registered at the sub-meter level, and dynamic deformation visualization is achieved by injecting equipment vibration spectrum parameters through vertex shaders.

[0171] Deploy dynamic LOD (Level of Detail) technology, based on the relative motion speed obtained from mechanical body parameters and human behavior data. The model's detail levels are dynamically adjusted, as shown in Formula 4:

[0172] (4);

[0173] In the formula, For the level of detail, For screen pixel density, The human eye's visual persistence frequency threshold. The floor symbol;

[0174] Integrating a real-time collision detection algorithm, when the spatial overlap between the device's working radius and environmental obstacles exceeds a safety threshold... At that time, a warning coloring based on stress criteria is triggered, and a risk evolution animation with time-series markers is generated.

[0175] Specifically, relative speed is calculated by comparing the speed of a person's movement with the trajectory of the machine. For example, when a person approaches high-speed machinery... Increase, triggering model detail enhancement ( (Increase).

[0176] See Figure 3 In step S3 above, when establishing a dynamic optimization model based on the reinforcement learning algorithm, the following is included:

[0177] Construct a hierarchical reinforcement learning architecture that includes an upper decision-making layer, a lower execution layer, and a risk prediction layer;

[0178] Among them, the upper decision-making layer uses a deep deterministic strategy gradient algorithm to generate a global optimization strategy;

[0179] The lower layer implements the optimal allocation of local resources based on the Q-learning algorithm;

[0180] The risk prediction layer integrates a Bayesian network to assess construction safety risk values ​​in real time.

[0181] Among them, the deep deterministic policy gradient algorithm adopts a double-delay deep deterministic policy gradient architecture, sets the policy network update frequency to 1 / 3 of the value network update frequency, and sets the target network smoothness coefficient τ=0.005.

[0182] Specifically, comparative experiments were conducted to verify the impact of the selection of the τ value on the construction optimization effect:

[0183]

[0184] Experimental conditions: The system was connected and operated for 72 hours in a base station construction scenario. The optimization effect of the construction machinery scheduling scheme under different τ values ​​was compared. The results showed that when τ=0.005, the best balance was achieved in terms of construction efficiency improvement, loss reduction and strategy stability.

[0185] Furthermore, in achieving optimal allocation of local resources, the following steps are included:

[0186] A dynamic multidimensional state space is constructed, and the vibration spectrum of construction machinery, the topology of material transportation path and fault early warning parameters are mapped into an orthogonal basis of state vectors. The high-dimensional state space is then reduced in dimensionality by a variational autoencoder.

[0187] Design a multi-objective reward function that incorporates a resource utilization improvement coefficient. Task delay penalty factor and safety risk coefficient The reward value is calculated as shown in Formula 5:

[0188] (5);

[0189] In the formula, For a moment The reward value, For actual resource utilization rate, Based on the utilization rate, To delay the task time, For the first Class of safety risk indicators, For the first Class of safety risk indicators for time The partial derivatives, The number of categories of safety risk indicators;

[0190] A dual experience playback mechanism is deployed to store the equipment vibration spectrum characteristics and environmental noise data in the short-term memory pool and the long-term memory pool respectively, and a priority sampling strategy is used to extract training samples.

[0191] Specifically, vibration spectrum, material transport path topology, and fault early warning parameters serve as orthogonal bases for state vectors, affecting actual resource utilization. :

[0192] ;

[0193] In the formula, The state vector of the th Each orthogonal basis component (such as the component corresponding to the vibration spectrum, the component corresponding to the material transport path topology, and the component corresponding to the fault warning parameter). For the first Each weight coefficient is dynamically optimized through reinforcement learning; The number of orthogonal basis components. For bias terms, The Sigmoid activation function normalizes the output to [0,1].

[0194] Fault warning parameters such as abnormal equipment temperature and hydraulic system pressure fluctuations are assessed using a safety risk coefficient. Participation in reward calculation:

[0195] The actual pressure of the hydraulic system With safety baseline The deviation is quantified as a pressure fluctuation safety risk indicator: ;

[0196] In the formula, This serves as a safety risk indicator for pressure fluctuations. The maximum allowable pressure deviation;

[0197] Pressure fluctuation safety risk indicators When applied to reward value calculation, it is as follows:

[0198] ;

[0199] A dynamic ε-greedy strategy is introduced, and it is based on the construction progress. The exploration probability is adaptively adjusted, as shown in Formula 6:

[0200] (6);

[0201] In the formula, To explore probability values, This is the initial exploration probability value. The minimum exploration probability threshold, To explore the decay index, This represents the maximum construction progress.

[0202] By updating the Q-value matrix in real time through edge computing nodes, when resource allocation conflicts among multiple devices are detected, a game equilibrium based on Shapley values ​​is triggered, generating a resource reallocation instruction set with spatiotemporal constraints.

[0203] Specifically, construction progress This includes: time-dimensional parameters (task completion timestamp, process delay duration) and spatial progress parameters (completed volume of structure, geological excavation depth).

[0204] Progress-driven decay: construction progress The higher, The smaller the value, the higher the exploration probability. As the system decreases, it becomes more inclined to utilize known optimal strategies.

[0205] Dynamic response: If the construction progress is halted due to equipment failure or sudden environmental changes ( (Not growing as expected), explore probability value Maintain a high level and encourage the exploration of new strategies to break through bottlenecks.

[0206] Furthermore, when assessing construction safety risk values ​​in real time, this includes:

[0207] A dynamic topological Bayesian network is constructed to map equipment vibration spectrum, personnel positioning offset and meteorological parameters into multidimensional evidence variables of risk nodes. A three-dimensional risk transmission chain of equipment failure, personnel behavior error and environmental change is established through a causal inference engine.

[0208] The conditional probability table is updated in real time based on the sensor data stream within the sliding time window. The variational Bayesian inference algorithm is used to dynamically optimize the dependency strength weights between nodes. The posterior probability of the nodes is calculated as shown in Equation 7.

[0209] (7);

[0210] In the formula, In the evidence Under the assumption The posterior probability, Assumption Next The prior probabilities of nodes corresponding to each evidence variable. For the first The rate of change of the gradient of each evidence variable, The time decay factor, To exclude the assumption Other possible assumptions besides To correspond to other possible assumptions Next The prior probabilities of nodes corresponding to each evidence variable. The number of evidence variables;

[0211] Risk entropy fusion is deployed, and KL divergence is used to quantify the deviation between the actual construction status and the safety baseline. When the joint probability of multiple risk nodes exceeds a threshold, the risk is assessed. When this occurs, it triggers emergency response plan matching based on convolutional neural networks;

[0212] A risk heat map with spatiotemporal correlation is generated, and the risk gradient descent direction is fed back to the construction machinery control system through edge computing nodes to correct the action selection probability distribution of the reinforcement learning strategy network in real time.

[0213] Specifically, meteorological parameters are used as evidence variables. Its gradient rate of change ( Factors such as temperature, humidity, noise, wind speed, and rainfall over time affect the posterior probability calculation of Bayesian networks, which are used for risk transmission chain modeling.

[0214] Using equipment vibration spectrum as an evidentiary variable Its gradient rate of change ( (e.g., changes in frequency and amplitude over time) affect the calculation of the posterior probability of a Bayesian network, which is used for risk transmission chain modeling.

[0215] Personnel positioning offset as an evidentiary variable Its gradient rate of change ( (For example, the change in positioning offset over time) affects the calculation of the posterior probability of the Bayesian network, and the risk transmission chain is modeled through the Bayesian network.

[0216] In step S4 above, the generation of the multidimensional parameter optimization instruction set includes:

[0217] Based on the spatiotemporal correlation of risk heatmaps and the direction of risk gradient descent, a three-dimensional collaborative optimization model of geology, equipment, and personnel is constructed. The safety risk entropy value is embedded into the reward function correction term of the Q-learning algorithm, and the correction function is shown in Equation 8.

[0218] (8);

[0219] In the formula, For a moment The revised reward value, This is the risk constraint strength coefficient. The KL divergence between real-time risk distribution and safety baseline. The time-varying gradient of risk indicators;

[0220] Spatial safety distance constraints are imposed on equipment scheduling parameters based on the risk transmission chain prediction results of dynamic topology Bayesian networks.

[0221] Specifically, geological deformation data alters real-time risk distribution through risk transmission chains. With safety baseline The difference is quantified by KL divergence:

[0222] ;

[0223] In the formula, Risk assumptions (such as excessively high soil moisture or excessively low geological hardness); This represents the current risk probability. The preset probability for the safety baseline;

[0224] KL divergence measures the deviation of the actual risk distribution from the safety baseline, and corrects the reward value of Q-learning.

[0225] The equipment vibration spectrum is used to predict the risk transmission chain through a Bayesian network, generate safety distance constraints (such as the minimum interval between equipment), and embed Q-learning reward correction.

[0226] The fatigue index deviating from the baseline value and the proportion of time without wearing safety equipment are expressed through the safety risk entropy value. The following factors affect the calculation of KL divergence and dynamically adjust the reward function:

[0227] The fatigue index deviates from the baseline value. Quantified as fatigue risk entropy value:

[0228] ;

[0229] In the formula, The fatigue risk entropy value. This represents the maximum allowable deviation of the fatigue index.

[0230] The percentage of time when safety equipment was not worn Directly mapped to the wearer's risk entropy value:

[0231] ;

[0232] In the formula, The risk entropy value for wearing;

[0233] fatigue risk entropy value and wearing risk entropy value Through dynamic weights Combined into a comprehensive risk entropy value:

[0234] ;

[0235] Weighting:

[0236] ;

[0237] The weights are dynamically adjusted according to the rate of change of the risk entropy value, enabling rapid response to sudden risks.

[0238] By embedding the comprehensive risk entropy value into the KL divergence calculation, the deviation between the real-time risk distribution and the safety baseline is corrected.

[0239] ;

[0240] Correction mechanism:

[0241] when (High-risk personnel status) KL divergence value amplified, strengthening risk penalties;

[0242] when (For low-risk individuals), the KL divergence remains at its original calculated value.

[0243] In step S5 above, the process of real-time performance evaluation and dynamic adjustment of the dynamic optimization model includes:

[0244] A dynamic performance simulation sandbox is built based on a digital twin engine. A multi-dimensional parameter optimization instruction set is loaded and real-time perception data stream is injected to generate a three-dimensional performance evaluation index that includes resource consumption rate, construction progress deviation and risk entropy value.

[0245] The multi-dimensional parameter optimization instruction set is segmented for performance verification based on a sliding time window mechanism. Within each time window, the deviation between the actual construction trajectory and the digital twin predicted trajectory is calculated using grey relational analysis, as shown in Formula 9.

[0246] (9);

[0247] In the formula, For deviation, For the first Actual value of dimension parameter For the first Dimensional digital twin prediction value, This is the environmental noise tolerance threshold. To prevent division by zero of extremely small constants, The number of dimensions, For the first Dimensional weights;

[0248] When the deviation exceeds the preset safety boundary, parameter sensitivity analysis based on adversarial generative networks is triggered to identify the key parameter dimensions that lead to performance degradation.

[0249] The time-series impact of weight parameter adjustments is predicted by a lightweight LSTM model deployed at the edge, generating a feedback signal that includes the gradient correction direction and adjustment step size, driving the dynamic optimization model to perform online parameter iteration.

[0250] Specifically, the deviation calculation parameters include: resource consumption rate, construction schedule deviation, and risk entropy value. Among them, the resource consumption rate parameter includes: material utilization rate (obtained from material usage) and equipment cumulative operating rate (obtained from equipment cumulative operating time); the construction schedule deviation parameter includes: task completion time deviation (obtained from task completion timestamp); and the risk entropy value parameters include: fatigue risk entropy value and wearing risk entropy value.

[0251] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-substantial changes and substitutions made by those skilled in the art based on the present invention shall fall within the scope of protection claimed by the present invention.

Claims

1. A dynamic optimization method for communication engineering construction based on multi-dimensional sensing, characterized in that, Includes the following steps: Multi-dimensional sensing data of the construction site are collected in real time through a multi-source heterogeneous sensor network, including environmental parameters, equipment operating status, construction progress and personnel behavior data; Spatiotemporal alignment processing is performed on multidimensional sensing data to construct a dynamic construction digital twin with spatiotemporal correlation; A dynamic optimization model is established based on reinforcement learning algorithm, with the objective optimization function being to maximize construction efficiency, minimize resource consumption and safety risks, and risk constraints are embedded. Input the dynamic construction digital twin into the dynamic optimization model and output a multi-dimensional parameter optimization instruction set that includes equipment scheduling parameters, construction path parameters, and resource configuration parameters; Real-time performance evaluation of the multi-dimensional parameter optimization instruction set is performed through edge computing nodes, and the weight parameters of the dynamic optimization model are dynamically adjusted to form a closed-loop feedback control. Constructing a dynamic construction digital twin with spatiotemporal correlation, including: The multidimensional sensing data is spatiotemporally calibrated by improving the Kalman filter algorithm to eliminate clock deviations and spatial coordinate system differences between devices. Establish a multi-dimensional association rule base for equipment, personnel, and environment based on knowledge graphs; Generate physically accurate 3D visualizations of construction scenes using a digital twin engine; The improved Kalman filter algorithm incorporates an adaptive noise covariance matrix that includes the variances of environmental noise, equipment noise, and human noise. When establishing a dynamic optimization model, the following should be included: Construct a hierarchical reinforcement learning architecture that includes an upper decision-making layer, a lower execution layer, and a risk prediction layer; Among them, the upper decision-making layer uses a deep deterministic strategy gradient algorithm to generate a global optimization strategy; The lower layer implements the optimal allocation of local resources based on the Q-learning algorithm; The risk prediction layer integrates a Bayesian network to assess construction safety risk values ​​in real time. The deep deterministic policy gradient algorithm adopts a dual-delay deep deterministic policy gradient architecture, with a target network smoothing coefficient τ=0.

005.

2. The method for dynamic optimization of communication engineering construction based on multi-dimensional perception according to claim 1, characterized in that, The multi-source heterogeneous sensor network includes: Vibration sensors and GNSS positioning modules deployed on construction machinery; Temperature and humidity sensors, noise sensors, and image acquisition devices are installed at the construction site. Wearable positioning terminal and physiological parameter monitoring equipment for construction workers; RFID tracking system for material transport vehicles.

3. The method for dynamic optimization of communication engineering construction based on multi-dimensional perception according to claim 1, characterized in that, When establishing a multi-dimensional association rule base for equipment, personnel, and environment, the following should be included: Define the entity types and associated attributes of the construction scene based on dynamic ontology modeling technology; An improved Apriori algorithm is used to mine frequent association patterns in multi-source heterogeneous data, and the confidence of association rules is dynamically adjusted by a time decay factor. A multidimensional association tensor of equipment, personnel, and environment is constructed, implicit association rules are extracted through tensor decomposition algorithm, and the rule weights are dynamically adjusted based on fuzzy logic controller; A real-time rule conflict detection mechanism is deployed, and the third-order gradient descent method is used to eliminate logical contradictions between equipment scheduling instructions and environmental constraints, generating a multi-dimensional association rule instance library with spatiotemporal labels.

4. The method for dynamic optimization of communication engineering construction based on multi-dimensional perception according to claim 1, characterized in that, When generating a 3D visualized construction scene, the following are included: A hybrid physics engine based on the coupling of discrete element method and finite element method is constructed to perform multi-field joint simulation of construction machinery motion trajectory, material stress distribution and geological deformation; The diameter of the discrete element particle is determined by the maximum speed of the device, the simulation step size, and the number of particle contact pairs per step.

5. The method for dynamic optimization of communication engineering construction based on multi-dimensional perception according to claim 1, characterized in that, Achieving optimal allocation of local resources includes: A dynamic multidimensional state space is constructed based on a variational autoencoder, and the vibration spectrum of construction machinery, the topology of material transportation path and fault early warning parameters are mapped as an orthogonal basis for state vectors. Design a multi-objective reward function that integrates resource utilization, task delay, and security risks; Deploy a dual-memory pool experience playback mechanism: short-term storage of device vibration spectrum characteristics, long-term storage of environmental noise data, and use priority sampling for training; A dynamic ε-greedy strategy is introduced to adaptively adjust the exploration probability based on the construction progress. By updating the Q-value matrix in real time through edge computing nodes, a spatiotemporal constraint instruction set is generated by triggering Shapley value game equilibrium when there are resource allocation conflicts among multiple devices.

6. The method for dynamic optimization of communication engineering construction based on multi-dimensional perception according to claim 5, characterized in that, When assessing construction safety risk values ​​in real time, the following should be included: A dynamic topological Bayesian network is constructed to map equipment vibration spectrum, personnel positioning offset and meteorological parameters into multidimensional evidence variables, and a risk transmission chain of equipment-personnel-environment is established through a causal inference engine; The conditional probability table is updated in real time based on the sliding time window data stream, and the variational Bayesian inference algorithm is used to dynamically optimize the dependency strength weights between nodes. The deviation between the actual construction status and the safety baseline is quantified by KL divergence. When the joint probability of multiple risk nodes exceeds the threshold, emergency plan matching based on convolutional neural network is triggered. Generate a spatiotemporal risk heat map, with edge computing nodes providing feedback on the risk gradient descent direction, and adjust the action selection probability distribution of the reinforcement learning policy network in real time.

7. The method for dynamic optimization of communication engineering construction based on multi-dimensional perception according to claim 6, characterized in that, When generating a multidimensional parameter optimization instruction set, the following are included: Based on the spatiotemporal correlation of risk heatmaps and the direction of risk gradient descent, a three-dimensional collaborative optimization model of geology, equipment, and personnel is constructed, and the safety risk entropy value is embedded into the reward function correction term of the Q-learning algorithm. Spatial safety distance constraints are imposed on equipment scheduling parameters based on the risk transmission chain prediction results of dynamic topology Bayesian networks.

8. The method for dynamic optimization of communication engineering construction based on multi-dimensional perception according to claim 1, characterized in that, When conducting real-time performance evaluations, the following are included: A dynamic performance simulation sandbox is built based on a digital twin engine, a multi-dimensional parameter optimization instruction set is loaded and real-time perception data stream is injected to generate three-dimensional performance evaluation indicators. The performance of the multi-dimensional parameter optimization instruction set is verified by segmenting based on sliding time windows. Within each time window, the deviation between the actual construction trajectory and the predicted trajectory of the digital twin is calculated using grey relational analysis.

Images