Multi-dimensional sensor network deployment and collaborative optimization method based on N00 construction method

Abstract

The invention discloses a multi-dimensional sensor network deployment and collaborative optimization method based on an N00 construction method. The method comprises the steps that S1, underground multi-source data fusion spatial data is collected and preprocessed; s2, based on the preprocessed data, establishing a multi-level spatial topological structure model based on an N00 construction method; s3, establishing a deployment dynamic weight function, and obtaining a deployment optimization model based on the deployment dynamic weight function and the multi-level spatial topological structure model; s4, obtaining a deployment optimization model based on the deployment dynamic weighting function and the multi-level spatial topological structure model; and S5, optimizing and deploying the optimization model based on a deep reinforcement learning algorithm of Transform. According to the embodiment of the invention, the spatial model is established based on multi-source data, the established model is more accurate, the dynamic weighting function is adopted to optimize the deployment model, the adaptive deployment scheme is formed, the deployment optimization model is further optimized, collaborative optimization is completed, and the optimal deployment scheme is given.

Description

Technical Field

[0001] The present invention relates to the field of multidimensional sensor deployment, and in particular to a multidimensional sensor network deployment and collaborative optimization method based on the N00 method. Background Art

[0002] In the industrial production process, multi-dimensional sensor deployment is required, and a spatial model needs to be established during the deployment process. Traditional modeling methods lack multi-source data, resulting in inaccurate modeling. Existing technologies have some deployment methods with simple algorithms, but when the environment in the industrial production process changes rapidly or multiple constraints (such as coverage, connectivity, and cost) are superimposed, how to dynamically adjust the sensor layout has always been a difficulty in the industry. Summary of the Invention

[0003] The purpose of the present invention is to provide a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method, aiming to solve the spatial modeling of complex environments and the adaptive deployment of sensor networks.

[0004] The present invention provides a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method, including:

[0005] S1, collect spatial data of underground multi-source data fusion and perform preprocessing;

[0006] S2. Based on the pre-processed data, a multi-level spatial topological structure model based on the N00 method is established;

[0007] S3. Establish a deployment dynamic weight function, and obtain a deployment optimization model based on the deployment dynamic weight function and the multi-level spatial topology structure model;

[0008] S4, obtaining a deployment optimization model based on the deployment dynamic weight function and the multi-level spatial topology structure model;

[0009] S5. Optimize and deploy the optimization model based on Transformer-based deep reinforcement learning algorithm.

[0010] The embodiment of the present invention uses multi-source data to establish a spatial model, which makes the established model more accurate. It uses a dynamic weight function to optimize the deployment model, and uses the Transformer's deep reinforcement learning algorithm to optimize the deployment model to form an adaptive deployment solution.

[0011] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it is implemented in accordance with the contents of the specification, and in order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0013] Figure 1 This is a flow chart of a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention;

[0014] Figure 2 This is a specific schematic diagram of spatial modeling of a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention;

[0015] Figure 3 1. This is a schematic diagram of point cloud data triangulation for spatial modeling of a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention;

[0016] Figure 4 1. This is a schematic diagram of boundary smoothing processing for spatial modeling of a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention;

[0017] Figure 5 Schematic diagram of node relationships in spatial modeling of a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention;

[0018] Figure 6 1 is a specific schematic diagram of a NURBS curve for spatial modeling of a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention;

[0019] Figure 7 Schematic diagram of the design of dynamic weight function and sensor deployment strategy of a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention;

[0020] Figure 8 Schematic diagram of sensor network deployment according to a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention;

[0021] Figure 9 This is a specific deployment diagram of a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention;

[0022] Figure 101 is a schematic diagram of a multidimensional sensor network collaborative optimization process of a multidimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention;

[0023] Figure 11 1 is a schematic diagram of a collaborative optimization algorithm of a multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] Method Example

[0026] According to an embodiment of the present invention, a spatial modeling method for sensor network deployment is provided. Figure 1 As shown, specifically including:

[0027] The N00 method is a pillarless mining technology that achieves efficient and safe coal mining by removing pressure from the top and forming a self-forming roadway. However, the spatial structure of the goaf and roadway-working face under this method is extremely complex. Traditional modeling methods are difficult to meet the requirements of refined sensor network deployment. Therefore, this paper proposes the following solution:

[0028] like Figure 2 As shown;

[0029] S1, collect spatial data fused from multiple sources and perform preprocessing;

[0030] Utilizing 3D laser scanning (LiDAR) equipment, high-density point cloud data is collected from goafs and tunnel-working faces to obtain precise 3D spatial morphology. Mobile devices (drones or robots) are equipped with inertial navigation systems (INS) to obtain spatial position and attitude information, compensating for LiDAR's limitations in dynamic environments. Total stations are also used to precisely measure key nodes and obtain high-precision geographic coordinates, which serve as a benchmark for data calibration and verification.

[0031] Said S1 specifically includes: collecting spatial data fused from multiple sources, performing time stamp calibration on the spatial data, unifying the spatial data into a coordinate system to obtain point cloud data, and performing denoising and fusion processing on the point cloud data to obtain first point cloud data.

[0032] Data collected by different devices is timestamped to ensure data integration based on the same time base. A unified mine coordinate system is established to convert all source data to this unified coordinate system, eliminating coordinate discrepancies. An improved adaptive filtering algorithm is used to denoise point cloud data, eliminating random noise and redundant points generated during the acquisition process and improving data quality.

[0033] The denoising and fusing of the point cloud data to obtain the first point cloud data specifically includes: using an improved adaptive filtering algorithm to denoise the point cloud data and perform weighted average fusion to obtain the first point cloud data. The formula of the improved adaptive filtering algorithm is as follows:

[0034]

[0035] Among them, σ i is the filtering threshold of the i-th point cloud, σ0 is the initial filtering threshold, ρ i is the local point cloud density of the i-th point cloud, ρ0 is the global average point cloud density, and β is the adjustment coefficient.

[0036] The algorithm dynamically adjusts the filtering threshold according to the point cloud density to avoid detail loss caused by over-filtering.

[0037] Furthermore, a weighted average fusion method is used to fuse LiDAR, INS and total station data to improve the accuracy and integrity of spatial data.

[0038] The calculation formula of weighted average fusion weight coefficient is as follows:

[0039]

[0040] Among them, w i is the weight of the i-th data source, σ i is the accuracy of the i-th data source, σ j is the accuracy of the jth data source, and n is the number of data sources.

[0041] The weighted coefficient w is calculated by the above weighted average fusion weight coefficient formula i , giving greater weight to the higher quality point clouds in each sensor data source, thus forming high-precision 'fused point cloud data'.

[0042] The fused high-precision point cloud can effectively reduce the risk of node misalignment caused by measurement errors, and provide more reliable raw data support for the following node matching and curve / surface fitting.

[0043] S2. Based on the preprocessed data, a multi-level spatial topological structure model is established;

[0044] S2 specifically includes: defining a first node and a first edge based on the first point cloud data, constructing a multi-dimensional topological graph model based on the first node and the first edge, establishing association relationships between different layers, and establishing a multi-level spatial topological structure model based on the association relationships.

[0045] Integrating the application scenario, we first defined new nodes and edges based on the process characteristics of the N00 method and actual site conditions. Nodes (V) define key points such as goaf boundaries, roadway intersections, working face endpoints, and support structure locations. Edges (E) are established between nodes based on actual connection relationships, including straight line segments, curve segments, and curved surface segments, reflecting the continuity and complexity of the spatial structure.

[0046] Next, we construct spatial topological relationships: We construct a three-dimensional topological graph model G = (V, E), where V is the node set and E is the edge set. The model includes a goaf layer, a roadway-working face layer, and an equipment layer. We establish relationships between these layers, such as the spatial relationship between the goaf and the roadway-working face, and the positional relationship between equipment and roadways, to facilitate the development of sensor deployment strategies.

[0047] Finally, the spatial topological relationship is optimized, and an automatic matching algorithm based on three-dimensional spatial density clustering is proposed to address the node mismatch problem caused by measurement errors.

[0048] The method of defining a first node and a first edge based on the first point cloud data, constructing a multi-dimensional topological graph model based on the first node and the first edge, establishing association relationships between different layers, and establishing a multi-level spatial topological structure model based on the association relationships specifically includes:

[0049] Based on the first point cloud data, a first node and a first edge are defined, and a multi-dimensional topological graph model is constructed based on the first node and the first edge. In view of the complex morphology and irregular boundary characteristics of a certain layer, point cloud data of the goaf is obtained from the first point cloud data, and a boundary modeling method based on an improved α-shape algorithm is used to establish the topological graph model;

[0050] like Figure 3 As shown;

[0051] The method of establishing a topology model by using a boundary modeling method based on an improved α-shape algorithm specifically includes: triangulating the point cloud data of the goaf to generate an initial triangular mesh, introducing an adaptive α parameter, dynamically adjusting the α value according to the point cloud density and spatial distribution, and establishing a topology map;

[0052] The adaptive α value calculation formula is as follows:

[0053]

[0054] Where: α i : α value of the i-th region, α0: initial α value, ρi : point cloud density of the i-th region, ρ max : Maximum point cloud density in the region, γ: Adjustment coefficient, which controls the sensitivity of α value to density;

[0055] like Figure 4 As shown;

[0056] The topology graph is smoothed using the Laplace smoothing algorithm to obtain a topology graph model. The Laplace smoothing algorithm is as follows:

[0057]

[0058] Among them, P i new : Updated node position, P i old : original node position, λ: smoothing factor, ranging from 0 to 1, k is the number of neighbor nodes of node i, P j is the coordinate of the jth neighbor node.

[0059] like Figure 5 As shown;

[0060] Calculate the local density ρ of each node i , the formula is as follows, ρ i is the number of nodes within a certain radius r;

[0061]

[0062] Among them, δ() is the Dirac function, N is the total number of nodes, P i : coordinates of the i-th node, P j : jth node coordinate, P i Used to fix a node, P j Used to compare all nodes;

[0063] Set the density threshold ρ th , filter out nodes in high-density areas; cluster high-density nodes, and use the cluster center as the new node position. The cluster node condition formula is as follows:

[0064] ||P i -P j ||≤δ;

[0065] Among them, δ is the tolerance threshold;

[0066] For curve and surface structures, a curve and surface fitting method based on non-uniform rational B-splines is used to establish a multi-level spatial topological structure model;

[0067] like Figure 6 As shown;

[0068] The NURBS curve formula is as follows:

[0069]

[0070] where N i,p (u) is the i-th basis function, w i is the weight of the i-th node of the curve and surface structure, P i is the i-th node coordinate of the curve and surface structure, u is a parameter variable defined within the node vector range, n is the number of nodes of the curve and surface structure, p is the order of the B-spline curve, [u min ,u max ] is the range of the parameter variable u, and the positional relationship between different layers is established. Based on this positional relationship, a multi-level spatial topological structure model is established. Usually, the connection between the layers is determined by calculating the coordinate distance, overlap, and proximity.

[0071] The present invention establishes a multi-level three-dimensional topological model of goaf, tunnel-working face and equipment, builds cross-level association relationships, and supports the precise deployment of multi-dimensional sensor networks.

[0072] First, a hierarchical division is performed: the goaf layer describes the spatial form, boundary characteristics, and internal structure of the goaf. The roadway-working surface layer describes the spatial location, shape, orientation, and topological connection between the roadway and working surface. The equipment layer describes the location, type, function, and attributes of underground equipment.

[0073] Next, cross-layer correlation is performed: through spatial relationships, the relative positions of goafs, roadways, and working faces are determined, identifying high-risk areas and key monitoring areas. Based on the layout of equipment in the roadway, the deployment requirements of equipment monitoring sensors are determined.

[0074] Finally, the data structure of the topological model is designed: an object-oriented data structure is used to encapsulate spatial entities and topological relationships at different levels to facilitate the automatic generation of sensor deployment strategies.

[0075] This method can accurately fit complex spatial curves and surfaces, reflecting the actual morphology of goaf and roadway-working surface under the N00 construction method.

[0076] Based on the construction of a complex spatial topological model of goaf and tunnel-working face under the N00 construction method, how to efficiently and accurately deploy a multidimensional sensor network becomes the key to ensuring safe production in mines. Due to the real-time changes in underground operating parameters (such as roof pressure, gas concentration, equipment density, etc.), traditional static deployment strategies are difficult to adapt to dynamic environments. To this end, this paper proposes a sensor deployment strategy based on a dynamic weight function, which incorporates real-time operating parameters into deployment decisions to achieve adaptive optimization of the sensor network.

[0077] like Figure 7 As shown;

[0078] Establish a deployment dynamic weight function, and obtain a deployment optimization model based on the deployment dynamic weight function and the multi-level spatial topology structure model;

[0079] Combine real-time operating parameters with sensor deployment requirements and dynamically adjust sensor deployment priorities based on the risk level and monitoring needs of different areas.

[0080] First, based on the special working conditions under the N00 method, key parameters that have a greater impact on mine safety are selected, including:

[0081] Top plate pressure (P top ): reflects the stability of goaf and roadway roof.

[0082] Gas concentration (C gas ): Affects the risk of gas explosion and poisoning.

[0083] Equipment density (D equip ): Reflects the number and importance of equipment in the area and affects the equipment monitoring needs.

[0084] Temperature change rate (ΔT): Early warning of spontaneous combustion and fire risks.

[0085] The establishment and deployment of a dynamic weight function specifically includes:

[0086] Obtain real-time operating parameters and standardize them. The standardization formula is as follows:

[0087]

[0088] Among them, F k is the kth real-time operating condition parameter, and are the minimum and maximum values of the parameter respectively;

[0089] A dynamic weight function is established based on real-time operating parameters. The formula is as follows:

[0090]

[0091] w s (i) is the dynamic weight of the s-th sensor at spatial position i; w 0s is the basic weight of the s-th sensor, reflecting the initial setting of its importance; α sk is the sensitivity coefficient of the s-th sensor to the k-th operating parameter; φ(f k(i)) is the influence function of the operating parameters, which reflects the influence of the parameters on the weight; β is the adjustment index, which controls the nonlinear degree of weight growth.

[0092] The deployment optimization model is obtained based on the deployment dynamic weight function and the multi-level spatial topology structure model. Specifically, the deployment optimization model is obtained based on the deployment dynamic weight function and the multi-level spatial topology structure model. The formula of the deployment optimization model is as follows:

[0093]

[0094] Z is the optimization objective function, which comprehensively considers the monitoring effect and deployment cost, N is the total number of spatial locations to be deployed, S is the total number of sensor types, and w s (i) is the dynamic weight of the s-th sensor at spatial position i, x is is a binary variable. If the s-th sensor is deployed at spatial position i, then x is =1; otherwise x is =0,c s is the unit cost of the s-th type of sensor, λ is the trade-off coefficient, balancing the monitoring effect and deployment cost;

[0095] Set constraints to ensure that key areas are effectively covered by the corresponding types of sensors;

[0096] 1. The coverage constraint formula is as follows:

[0097] Where: S i is the set of sensor types required for spatial position i; N(i) is the set of neighborhood positions of spatial position i, is the coverage range of the sensor, x js is a binary variable. If the s-th sensor is deployed at location j, then x js =1; otherwise x js =0,;

[0098] If the sth type sensor effectively covers the spatial position i at the spatial position j, then Otherwise, it is 0; θ i : the minimum coverage required for spatial position i;

[0099] 2. Connectivity constraints:

[0100] According to the spatial topology model and the communication range of the sensor, the connectivity matrix between nodes is constructed, and the connectivity constraint condition is: network connectivity ≥ δ;

[0101] The network connectivity is calculated by the algebraic connectivity of the minimum cut set or Laplace matrix, and δ is the preset connectivity threshold.

[0102] 3. Deployment location constraints:

[0103]

[0104] Among them, N valid is the set of deployable spatial locations i.

[0105] Dynamic weight function w s In (i), f k (i) is the working condition parameter of spatial position i (such as roof pressure, gas concentration, equipment density, etc.). These working condition parameters are closely related to the spatial topology: the roof pressure value is usually measured according to the roadway-working face node or goaf boundary node (that is, the stress applied to the topological node). Equipment density D equip It is necessary to calculate the gas concentration based on the topology of the equipment layer and the location of the equipment. gas Measurements or estimates are also required at the corresponding spatial nodes.

[0106] In summary, “behind position i are topological nodes / grid coordinates, and working condition f k (i) is obtained by topological model + real-time monitoring data. The two are combined in the dynamic weight function so that w s (i) It combines spatial location dependence with real-time working condition dependence. The combination of “spatial topology” and “dynamic weight” is mainly reflected in the following three points:

[0107] ① Topology is the basis for positioning: let each position i determine its own coordinates and adjacency information in the goaf or tunnel; ② The correspondence between the working condition parameters and the position: these working condition values f k (i) Need to be allocated based on the mapping of nodes / grids in the topology; ③ Dynamic weight w s (i) The deployment priority of sensor type s can be calculated based on the real-time operating parameters of topological location i.

[0108] Dynamic weight function part w s (i) is responsible for determining the attractiveness of the "working condition priority" to sensor s at topological location i (i.e., deployment value). The spatial topology structure guarantees "where deployment is allowed and how to maintain network coverage and communication" through coverage constraints N(i) and connectivity constraints.

[0109] Spatial topology and dynamic weighting are combined in the objective function (using weighting to calculate deployment benefits) and constraints (using topological adjacency and network connectivity thresholds to maintain coverage and communication stability). Spatial topology determines the "deployable range + coverage / connectivity requirements," while dynamic weighting determines the "value weight of each location and sensor type." These two elements, combined within the objective function and constraint system, enable the overall deployment to be "temporally and spatially adaptive."

[0110] like Figure 8 and 9 As shown;

[0111] In view of the special risks in goaf, tunnel-working face and equipment-intensive areas under the N00 construction method, a sensor deployment strategy for key areas is proposed to improve the accuracy and real-time performance of safety monitoring.

[0112] In the goaf, high-precision stress sensors and microseismic monitoring equipment are deployed in the high stress area of the goaf roof. top , dynamically adjust the sensor weight w stress (i) High stress areas are given higher weights. Gas sensors are deployed in upper corners and blind alleys where gas is likely to accumulate. The sensitivity coefficient α of the gas sensor to gas concentration is gas Set it higher to ensure priority deployment in high-gas areas.

[0113] In terms of tunnel-working face, in equipment-intensive areas, equipment status monitoring sensors such as temperature sensors, current sensors, etc. are deployed. equip , adjust the sensor weight w equip (i) Deploy temperature, humidity, and hazardous gas sensors in areas with high human activity. Adjust the weights of the relevant sensors based on the temperature change rate ΔT and humidity parameters.

[0114] To ensure the integrity of the communication network, relay nodes are deployed in areas with severe signal attenuation to ensure network connectivity. The deployment weights of relay nodes are adjusted based on the signal attenuation parameters in the spatial topology model. Backup nodes are deployed near key nodes to improve network robustness and resilience. The locations of backup nodes are planned by analyzing key network nodes (highly concentrated data aggregation points).

[0115] The present invention also provides a multi-dimensional sensor network deployment optimization method, which is based on a Transformer optimization deployment optimization model.

[0116] A deep reinforcement learning algorithm (DQN-Trans) based on the Transformer architecture and an improved multi-objective optimization algorithm fully utilize dynamic weight functions and spatial topology models to achieve real-time adaptive adjustment of sensor network deployment and solve the global optimal solution.

[0117] The multi-dimensional sensor network deployment problem under the N00 method is modeled as a sequential decision problem in deep reinforcement learning (DRL). The learning ability of intelligent agents in dynamic environments is utilized to achieve real-time optimization of sensor deployment strategies.

[0118] like Figure 10 As shown:

[0119] First, the state space:

[0120] The state vector st formula is defined as follows: At time step t, the state st contains the current sensor deployment status, real-time operating parameters and spatial topology information.

[0121] s t =[X t ,F t ,G t ];

[0122] in:

[0123] Sensor deployment matrix, The sth type of sensor is deployed for location i.

[0124] Working condition parameter matrix, including the real-time working condition parameters of each position i

[0125] G t : Spatial topology information, including the connection relationship between nodes and edges.

[0126] Next, define the action space (ActionSpace):

[0127] The action set A is that the agent can choose to deploy or remove a sensor of a certain type s at a certain location i at each time step, or keep it unchanged.

[0128] a t =(i,s,a op );

[0129] Among them, a op ∈{deploy, remove, keep}.

[0130] At the same time, a reward function is designed and a dynamic weight function is integrated into the reward function to balance monitoring effect, deployment cost and network connectivity.

[0131] R t =ΔZ t -η·Penalty t ;

[0132] Where: ΔZ t =Z t -Z t -1: The increment of the optimization objective function, Z t is the objective function value at time step t. t The definition of is consistent with the above optimization objective function:

[0133] Penaltyt: Penalty for violating constraints, including coverage constraints and connectivity constraints.

[0134] η: Penalty coefficient, which controls the influence of the penalty term.

[0135] Finally determine the goal of maximizing cumulative rewards:

[0136]

[0137] Where π is the policy, γ∈[0,1) is the discount factor, and T is the total number of time steps.

[0138] Specifically including: Transformer-based deep reinforcement learning algorithm optimization deployment optimization model.

[0139] Real-time operating parameters are obtained, and the state vectors of the real-time operating parameters, deployment optimization model, and multi-level spatial topology structure model are converted into embedding vectors. Position encoding is added to the embedding vector to obtain an embedded position vector. The embedded position vector is input into the multi-head self-attention layer to extract global features. The global features are nonlinearly transformed based on the feedforward neural network, and an evaluation value is output. The model with the highest evaluation value is selected as the optimal deployment model.

[0140] Under the N00 method, a deep Q network (DQN-Trans) based on the Transformer architecture is introduced, utilizing the powerful global feature extraction capability of the Transformer to address the limitations of traditional DQN in processing large-scale, high-dimensional state spaces.

[0141] The building blocks of the DQN-Trans model are as follows:

[0142] Input layer: The state vector s t Convert it into an embedding vector, including the embedding of sensor deployment status, working condition parameters and spatial topology information. Add position encoding to maintain the position information of each element in the sequence.

[0143] Transformer Encoder: Uses a multi-head self-attention layer to extract global features of the input sequence. A feed-forward neural network is used to perform nonlinear transformations on the output of the attention layer.

[0144] The action value function Q(s) of the output layer t , a t ,θ) outputs corresponding to each possible action a t Q value.

[0145] like Figure 11 As shown;

[0146] The execution flow of the algorithm is as follows:

[0147] Step 1: Initialize the parameters θ of the DQN-Trans model.

[0148] Initialize the experience replay pool D.

[0149] Step 2: Interaction and data collection;

[0150] For each time step t: get the current state s from the environment t .

[0151] Select action a according to ∈ greedy strategy t :

[0152] ∈: Exploration rate; used to decide whether to take random actions (exploration) or use the optimal action given by the network in the current state.

[0153] Execute action a t , get reward R t and the next state s t+1 . R t : Instant reward; composed of "monitoring effect - cost - penalty (violation of coverage / connectivity, etc.)".

[0154] (s t ,a t ,R t ,s t+1 ) is stored in the experience replay pool D.

[0155] Step 3: Model training;

[0156] For each training step:

[0157] Randomly sample mini-batches of data from D {(s j ,a j ,R j ,s j+1 )}.

[0158] Calculate the target value y j :

[0159] a' is the next action;

[0160] Among them, θ- is the parameter of the target network, which is updated regularly.

[0161] Minimize the loss function L(θ):

[0162] Mini-batch size: denoted as |B|;

[0163] Update the model parameters θ.

[0164] Step 4: Parameter update;

[0165] The deployment strategy of the sensors is as follows:

[0166] In view of the special risks in goaf, tunnel-working face and equipment-intensive areas under the N00 construction method, a sensor deployment strategy for key areas is proposed to improve the accuracy and real-time performance of safety monitoring.

[0167] In the goaf, high-precision stress sensors and microseismic monitoring equipment are deployed in the high stress area of the goaf roof. top , dynamically adjust the sensor weight w stress (i) High stress areas are given higher weights. Gas sensors are deployed in upper corners and blind alleys where gas is likely to accumulate. The sensitivity coefficient α of the gas sensor to gas concentration is gas Set it higher to ensure priority deployment in high-gas areas.

[0168] In terms of tunnel-working face, in equipment-intensive areas, equipment status monitoring sensors such as temperature sensors, current sensors, etc. are deployed. equip , adjust the sensor weight w equip (i) Deploy temperature, humidity, and hazardous gas sensors in areas with high human activity. Adjust the weights of the relevant sensors based on the temperature change rate ΔT and humidity parameters.

[0169] To ensure the integrity of the communication network, relay nodes are deployed in areas with severe signal attenuation to ensure network connectivity. The deployment weights of relay nodes are adjusted based on the signal attenuation parameters in the spatial topology model. Backup nodes are deployed near key nodes to improve network robustness and resilience. The locations of backup nodes are planned by analyzing key network nodes (highly concentrated data aggregation points).

[0170] The implementation process is designed as follows:

[0171] Step 1: Obtain the latest operating parameter data through the existing monitoring system.

[0172] Step 2: Use the dynamic weight function to calculate the weight of each location and type of sensor.

[0173] Step 3: Find the optimal sensor deployment solution.

[0174] Step 4: Based on the solution results, actually deploy sensors or adjust the configuration of existing sensors.

[0175] Step 5: Regularly update operating parameters, recalculate weights, and adjust deployment plans to achieve dynamic optimization.

[0176] The beneficial effects of the present invention are as follows:

[0177] The present invention proposes a three-dimensional spatial topological modeling method based on multi-source data fusion, which solves the modeling problem of the complex spatial structure of goaf and tunnel under the N00 construction method.

[0178] Real-time operating condition parameters are integrated into the sensor deployment strategy, a dynamic weight function is proposed, and adaptive optimization of sensor deployment is achieved.

[0179] An optimization model that comprehensively considers monitoring effect and deployment cost is established to optimize sensor deployment.

[0180] For the first time, the Transformer-based deep reinforcement learning algorithm was applied to sensor network optimization, and the DQN-Trans algorithm was proposed, which realized real-time adjustment and global optimization of sensor deployment strategies.

[0181] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. These modifications or replacements of the technical solutions of the embodiments of the present invention do not cause the essence of the corresponding technical solutions to deviate from the scope of this solution.

Claims

1. A multi-dimensional sensor network deployment and collaborative optimization method based on the N00 method, characterized in that: include: S1, collect spatial data of underground multi-source data fusion and perform preprocessing; S2. Based on the pre-processed data, a multi-level spatial topological structure model based on the N00 method is established; S3. Establish a deployment dynamic weight function, and obtain a deployment optimization model based on the deployment dynamic weight function and the multi-level spatial topology structure model; S4, obtaining a deployment optimization model based on the deployment dynamic weight function and the multi-level spatial topology structure model; S5. Optimize and deploy the optimization model based on Transformer-based deep reinforcement learning algorithm.

2. The method according to claim 1, characterized in that Said S1 specifically includes: collecting spatial data fused from multiple sources of underground data, performing time stamp calibration on the spatial data, unifying the spatial data into a coordinate system to obtain point cloud data, and performing denoising and fusion processing on the point cloud data to obtain first point cloud data.

3. The method according to claim 2, characterized in that The denoising and fusing of the point cloud data to obtain the first point cloud data specifically includes: using an improved adaptive filtering algorithm to denoise the point cloud data and perform weighted average fusion to obtain the first point cloud data. The formula of the improved adaptive filtering algorithm is as follows: Among them, σ i is the filtering threshold of the i-th point cloud, σ0 is the initial filtering threshold, ρ i is the local point cloud density of the i-th point cloud, ρ0 is the global average point cloud density, and β is the adjustment coefficient; The calculation formula of the weighted average fusion weight parameter is as follows: Among them, w i is the weight of the i-th data source, σ i is the accuracy of the i-th data source, σ j is the j-th data source accuracy, n is the number of data sources, 1≤j≤n.

4. The method according to claim 2, characterized in that The S2 specifically includes: defining a first node and a first edge based on the first point cloud data, constructing a multidimensional topological graph model based on the first node and the first edge, establishing an association relationship between different layers, and establishing a multi-level spatial topological structure model based on the N00 method according to the association relationship.

5. The method according to claim 4, characterized in that The method further comprises: defining a first node and a first edge based on the first point cloud data, constructing a multi-dimensional topological graph model based on the first node and the first edge, establishing association relationships between different layers, and establishing a multi-level spatial topological structure model based on the association relationships. Based on the first point cloud data, first nodes and first edges are defined, where the first nodes include: goaf boundary points, roadway intersection points, working face endpoints, and support structure locations; the first edges are connections between the first nodes; and a multidimensional topological graph model is constructed based on the first nodes and the first edges; Based on the first edge, the goaf layer, the roadway-working face layer, and the equipment layer are established. The goaf layer describes the spatial form, boundary characteristics, and internal structure information of the goaf. The roadway-working face layer describes the spatial position, shape, direction, and topological connection relationship between the roadway and the working face. The equipment layer describes the location, type, function, and attributes of the underground equipment. Establish the positional relationship between the goaf layer, the roadway-working face layer, and the equipment layer, that is, calculate the coordinate distance, overlap, and proximity between different layers; According to the characteristics of complex shapes and irregular boundaries in each position relationship, the local density m of each node is calculated i , the formula is as follows, Where δ(·) is the Dirac function, N is the number of nodes within the radius r, and P i is the coordinate of the i-th node, P j is the coordinate of the jth node; Set the density threshold ρ th , filter out nodes in high-density areas; cluster high-density nodes, and use the cluster center as the new node position. The cluster node condition formula is as follows: ||P i -P j ||≤δ, Where δ is the tolerance threshold; In view of the complex shape and irregular boundary of the goaf, the point cloud data of the goaf in the first point cloud data is obtained, and a boundary modeling method based on the improved α-shape algorithm is used to establish a topological graph model; For the curve and surface structures in each position relationship, a curve and surface fitting method based on non-uniform rational B-splines is used to establish a multi-dimensional topological structure model. The NURBS curve formula is as follows: where N i,p (u) is the i-th basis function, w i is the weight of the i-th control point of the curve and surface structure. The control point is a term used in NURBS fitting. i is the coordinate of the ith control point of the curve and surface structure, u is a parameter variable defined within the range of the control point vector, n is the number of control points of the curve and surface structure, p is the order of the B-spline curve, [u min ,u max ] is the value range of parameter variable u; A multi-level spatial topological structure model is established based on positional relationships.

6. The method according to claim 5, characterized in that The topology model is established by adopting a boundary modeling method based on an improved α-shape algorithm, specifically comprising: Triangulate the point cloud data of the goaf to generate an initial triangular mesh, introduce an adaptive α parameter, dynamically adjust the α value according to the point cloud density and spatial distribution, and establish a topological map; The adaptive α value calculation formula is as follows: Where: α i is the α value of the i-th region, α0 is the initial α value, n i is the point cloud density of the i-th region, ρ max is the maximum point cloud density in the region, γ is the adjustment coefficient, which controls the sensitivity of α value to density; The topology graph is smoothed using the Laplace smoothing algorithm to obtain a topology graph model. The Laplace smoothing algorithm is as follows: Among them, P i new is the updated node position, P i old is the original node position, λ is the smoothing factor, ranging from 0 to 1, k is the number of neighbor nodes of node i, K j is the coordinate of the jth neighbor node.

7. The method according to claim 1, characterized in that The establishment and deployment of a dynamic weight function specifically includes: Obtain the downhole real-time working condition parameters and standardize the downhole working condition parameters. The standardization formula is as follows: Among them, F k is the kth real-time operating condition parameter, and are the minimum and maximum values of the parameter respectively; A dynamic weight function is established based on real-time operating parameters. The formula is as follows: w s (i) is the dynamic weight of the s-th sensor at spatial position i; w 0s is the basic weight of the s-th sensor; α sk is the sensitivity coefficient of the s-th sensor to the k-th operating parameter; φ(f k (i)) is the influence function of the operating parameters, which reflects the influence of the parameters on the weight; β is the adjustment index, which controls the nonlinear degree of weight growth.

8. The method according to claim 7, characterized in that The S4 specifically includes: obtaining a deployment optimization model based on the deployment dynamic weight function and the multi-level spatial topology structure model. The formula of the deployment optimization model is as follows: Z is the optimization objective function, which comprehensively considers monitoring effect and deployment cost; N is the total number of spatial locations to be deployed; S is the total number of sensor types; w s (i) is the dynamic weight of the s-th sensor at spatial position i; x is is a binary variable. If the s-th sensor is deployed at spatial position i, then x is =1, otherwise x is =0;c s is the unit cost of the s-th type of sensor; λ is the trade-off coefficient, balancing the monitoring effect and deployment cost; Set constraints to ensure that key areas are effectively covered by the corresponding types of sensors; 1. The coverage constraint formula is as follows: Where: S i is the set of sensor types required for spatial position i; N(i) is the set of neighborhood positions of spatial position i, and is the coverage range of the sensor; x js is a binary variable. If the s-th sensor is deployed at spatial position j, then x js =1, otherwise x js = 0; if the sth type sensor effectively covers position i at spatial position j, then otherwise θ i is the minimum coverage required for spatial position i; 2. Connectivity constraints: According to the spatial topology model and the communication range of the sensor, the connectivity matrix between nodes is constructed, and the connectivity constraint condition is: network connectivity ≥ δ; The network connectivity is calculated by the algebraic connectivity of the minimum cut set or Laplace matrix, and δ is the preset connectivity threshold.

3. Deployment location constraints: Among them, N valid is the set of deployable spatial locations i.

9. The method according to claim 1, characterized in that The Transformer-based deep reinforcement learning algorithm optimization deployment optimization model specifically includes: Obtain real-time downhole operating parameters, convert the state vectors of the downhole real-time operating parameters, deployment optimization model, and multi-level spatial topology structure model into embedding vectors, add position encoding to the embedding vector to obtain an embedded position vector, input the embedded position vector into the multi-head self-attention layer to extract global features, perform nonlinear transformation on the global features based on the feedforward neural network, output evaluation values, and select the model with the highest evaluation value as the optimal deployment model.

Images