Resource dynamic scheduling method and system for underground coal mine self-powered wireless sensor network

Abstract

A resource dynamic scheduling method and system for a coal mine underground self-powered wireless sensor network, the method comprising: wind energy collection and real-time power supply calculation; wind speed distribution parameter acquisition and distribution type judgment; wind speed distribution type-based energy supply interruption probability constraint conversion; wind speed-energy mapping and node working mode adjustment; sensor node data sensing and energy consumption calculation; two-hop wireless communication transmission and energy consumption calculation; cluster-level information age quantification and dynamic update; optimization target and constraint condition construction. The system comprises: a plurality of micro wind turbines respectively corresponding to a plurality of ventilation roadways installed underground; energy storage units respectively connected with the plurality of micro wind turbines; a plurality of sensor clusters respectively corresponding to the plurality of ventilation roadways; and a local aggregator internally integrated with a wind power supply control module and a wireless resource scheduling module, respectively connected with the energy storage units and the plurality of sensor clusters. The present application can ensure the stability and reliability of self-powered wireless data transmission.

Description

Technical Field

[0001] This invention belongs to the field of wireless sensor monitoring technology in underground coal mines, specifically relating to a method and system for dynamic resource scheduling of self-powered wireless sensor networks in underground coal mines. Background Technology

[0002] Wireless sensor networks (WSNs) in coal mines are a key technology for environmental sensing, disaster early warning, and safety monitoring. However, the inherent strong multipath fading, high electromagnetic interference, and especially the difficulty in power supply in the underground environment make it difficult to achieve long-term and stable deployment of traditional WSNs that rely on batteries or wired power. Therefore, self-powered WSNs based on environmental energy harvesting, especially technologies that utilize wind energy harvesting from the mine ventilation system for self-powering, have become the most promising solution.

[0003] While wind energy harvesting makes long-term operation of underground wind-powered network (WSN) possible, its core challenge lies in the randomness and instability of energy supply. Although the airflow generated by the ventilation system is continuous, it fluctuates dramatically due to factors such as tunnel structure and equipment operation, resulting in intermittent energy supply. This unpredictable energy input creates a natural contradiction with the stringent real-time data requirements of underground monitoring tasks. Traditional static resource allocation strategies are inadequate in this dynamic environment, failing to effectively balance energy efficiency and information age, severely restricting the practical application of wind-powered WSNs. More critically, traditional network performance evaluation metrics have significant limitations in unstable energy supply scenarios. Whether it's node-to-remote monitoring center (RMC) latency or network throughput, these metrics primarily focus on the efficiency or rate characteristics of data transmission, failing to effectively characterize the dynamic impact of energy fluctuations on data timeliness, and especially struggling to quantify the data freshness degradation caused by unstable energy supply. Against this backdrop, Information Age (AoI), as a novel performance metric with greater scenario adaptability, precisely quantifies the complete time span from data generation to successful acquisition by the receiver. It comprehensively and dynamically characterizes data freshness and has gradually become a core standard for evaluating the real-time performance of wireless sensing systems with unstable power supplies. Further analysis reveals that when nodes enter periodic dormancy due to insufficient energy supply, or when data updates are significantly delayed due to momentary energy shortages, AoI exhibits a step-like upward trend. This sharp deterioration in data freshness directly leads to the backend decision-making system's inability to make timely and accurate responses based on the latest sensed data, thus constraining the operational reliability and decision-making effectiveness of the entire monitoring system. Therefore, there is an urgent need to provide a resource dynamic scheduling method and system for self-powered wireless sensor networks in coal mines. Summary of the Invention

[0004] To address the problems of existing technologies, this invention provides a method and system for dynamic resource scheduling of self-powered wireless sensor networks (WSNs) in coal mines. This method effectively ensures a continuous and stable energy supply and real-time data transmission for underground WSNs, and effectively solves the problems of difficult deployment, short lifespan, and high energy consumption associated with traditional wired / battery-powered WSNs. The system is highly intelligent and ensures the stability and reliability of self-powered wireless data transmission.

[0005] To achieve the above objectives, the present invention provides a resource dynamic scheduling method for self-powered wireless sensor networks in coal mines, comprising the following steps; Step 1: Wind energy harvesting and real-time power supply calculation; using micro wind turbines deployed in ventilation roadways in coal mines to capture wind energy and convert it into electrical energy, and using energy storage units to cache the electrical energy to provide continuous wireless power supply to WSN nodes; Step 2: Obtaining wind speed distribution parameters and determining distribution type; wind speed data in the mine is collected by wind speed sensors deployed in the ventilation tunnel, and Kolmogorov-Smirnov detection is used to determine whether the data conforms to a truncated Weibull distribution in order to determine the wind speed distribution type in the target area in the mine. Step 3: Power supply interruption probability constraint conversion based on wind speed distribution type; Based on the wind speed distribution type, the probabilistic power supply interruption constraint is converted into a deterministic constraint to ensure the stability of the node cluster's power supply. Step 4: Wind speed-energy mapping and node working mode adjustment; Based on wind speed distribution parameters and energy supply constraints, establish a wind speed-energy mapping model, and combine real-time wind speed monitoring values ​​to predict the energy supply intensity in future periods; During periods of high energy supply, nodes prioritize data acquisition and transmission tasks to maximize data throughput; During periods of low energy supply, nodes switch to a low-power sleep state, retaining only core sensing functions to avoid energy waste. Step 5: Sensor node data sensing and energy consumption calculation; Sensor nodes collect downhole environmental data according to optimized parameters to ensure that sensing energy consumption does not exceed the real-time wind power supply capacity; Step Six: Two-hop wireless communication transmission and energy consumption calculation; A two-hop transmission architecture of sensor node → local aggregator → remote monitoring center is adopted. Interference is reduced by orthogonal sub-channel division, and the transmission rate and energy consumption of each hop are calculated. Step 7: Cluster-level information age quantification and dynamic update; evaluate data freshness in two-hop transmission scenarios through cluster-level AoI, and dynamically update AoI values ​​to reflect data timeliness; Step 8: Optimization Objectives and Constraints Construction; With minimizing the average cluster-level AoI and maximizing the cluster-level energy efficiency as the core objectives, construct a multi-constraint optimization model; Step 9: Dynamic resource allocation based on risk-sensitive reinforcement learning; construct a reinforcement learning framework to intelligently optimize node perception duration, communication bandwidth allocation and transmission power, and achieve a coordinated balance between energy consumption and communication performance and closed-loop scheduling.

[0006] As a preferred option, in step one, the first step is calculated according to formula (1). Clusters in time slices Wind energy harvesting power within ; (1); In the formula, The preset time slice length; air density, ; The swept area of ​​the turbine blades. , The radius of the turbine blade; Energy conversion efficiency; For mechanical efficiency; This represents the real-time wind speed at the turbine.

[0007] As a preferred option, in step two, the process of using the Kolmogorov-Smirnov test to determine whether the data conforms to a truncated Weibull distribution, in order to determine the wind speed distribution type in the downhole target area, is as follows: S21: Construction of historical wind speed dataset; In the ventilation tunnel corresponding to each sensor cluster, deploy wind speed sensors with a measurement accuracy of ±0.1 m / s and a sampling frequency of 1 Hz, continuously collect wind speed data for no less than 72 hours, and construct a historical wind speed dataset. ,in, For the first Second sampling time The wind speed value, For sampling sequence number, This represents the total number of samples. S22: Classification type determination; The Kolmogorov-Smirnov test is used to test the goodness of fit of the historical wind speed dataset. The historical wind speed dataset is compared with a truncated Weibull distribution defined as [0.5 m / s, 8 m / s]. If the probability value corresponding to the test statistic is ≥0.05, the wind speed is determined to follow a truncated Weibull distribution. The shape parameter is calculated by the maximum likelihood estimation method. Scale parameters and actual domain ,in, This is the minimum wind speed value from historical data. The maximum wind speed value in the historical data is given; if the probability value corresponding to the test statistic is <0.05, the wind speed is determined to be of arbitrary distribution, and the mean wind speed is calculated according to formula (2) and formula (3) respectively. and variance ; (2); (3).

[0008] As a preferred option, in step three, the process of transforming the probabilistic power supply interruption constraint into a deterministic constraint is as follows: S31: Definition of Net Energy Surplus; Wind Energy Harvesting Power Factor According to formula (4), the first... Clusters in time slices Net energy surplus ; (4); In the formula, The system loss coefficient, air density; For the first Total energy consumption of each cluster, including energy consumption for sensing, integration, and transmission; and These represent the minimum and maximum values ​​of the net energy surplus, respectively. , ; S32: Constraint transformation for truncating the Weibull distribution; S32-1: Calculate the mean of the net energy surplus according to formulas (5) and (6) respectively. and variance Among them, the wind speed is calculated according to formula (7). of Step Moment ; (5); (6); (7); In the formula, Wind speed The third moment; Wind speed The 6th moment; An index for a stage or time step; For the first The weighting coefficients for each stage; This represents the value of the gamma function under specific parameters; , These are the upper and lower bound parameters of the gamma function, respectively. , ; To truncate the Weibull distribution function, , , These represent the upper and lower bound parameters of the distribution function, respectively; S32-2: Based on Bernstein's inequality, constrain the probability of power outage. This is transformed into a deterministic constraint, as shown in formula (8), where, For the first A cluster of sensors in time slices The probability of power outage within the area; (8); In the formula, For the first Individuals in time slices The mean of the net internal energy surplus; The threshold value for the probability of power outage is 0.1 to 0.2. S33: Constraint transformation under arbitrary distribution; S33-1: First, use the average wind speed. As the starting point, for Performing a second Taylor expansion yields formula (9); then let , , According to formula (10), the quadratic approximate function of net energy surplus is obtained. ; (9); (10); S33-2: Based on the conditional value at risk theory, the interruption probability constraint is transformed into a deterministic constraint in the worst case, as shown in formula (11); (11); In the formula, This is the conditional risk value.

[0009] S33-3: Solve the constraints using semidefinite programming to satisfy... ,in, It is a positive semi-definite matrix of order two. ,satisfy and ; This is the wind speed statistical feature matrix. ; The CVaR constraint is used to relax the variables.

[0010] As a preferred option, in step five, the process of ensuring that the sensed energy consumption does not exceed the real-time wind power supply capacity is as follows: S51: Data acquisition volume calculation; nodes are calculated at a fixed frequency. Collect downhole environmental data and calculate the first step according to formula (12). tudi Each node in the time slice The amount of data collected within ; (12); In the formula, For node sensing duration; S52: Sensing energy consumption calculation; calculate node sensing energy consumption according to formula (13). and ensure ; (13); In the formula, To maintain constant power consumption of the sensing circuit, .

[0011] As a preferred option, the process of calculating the transmission rate and energy consumption for each hop in step six is ​​as follows: S61: Calculation of first-hop transmission rate and energy consumption; S61-1: The 2.4 GHz band is divided into 15 to 20 sub-channels to avoid interference between nodes within the cluster; S61-2: Calculate the path loss of the first hop according to formula (14). , (14); In the formula, For reference path loss, ; This represents the loss index for short-distance underground paths. This represents the distance between the node and the aggregator. ; S61-3: Calculate the channel gain of the first hop according to formula (15). ; (15); In the formula, The small-scale fading coefficient ranges from 0.5 to 1.0 and follows a Nakagami distribution. As the decay factor, For fading power; S61-4: Calculate the transmission rate of the first hop according to formula (16). ; (16); In the formula, The communication bandwidth allocated to the node; This refers to the node's transmit power. Channel noise power; S61-5: Calculate the transmission duration of the first hop according to formulas (17) and (18) respectively. and transmission energy consumption And ensure that the constraints are met: ,and ,in, This represents the upper limit of the sensor node's transmit power. This represents the lower limit of the signal-to-noise ratio for sensor nodes; (17); (18); S62: Calculation of second-hop transmission rate and energy consumption; S62-1: Calculate the path loss of the second hop according to formula (19) ; (19); In the formula, The distance from the aggregator to the remote monitoring center; S62-2: Calculate the transmission rate of the second hop according to formula (20). ; (20); In the formula, This refers to the communication bandwidth of the local aggregator. For the aggregator's transmit power, For aggregator link channel gain; S62-3: Calculate the total amount of data within the cluster according to formula (21). ; (twenty one); In the formula, For the first The set of nodes in a cluster; S62-4: Calculate the transmission duration of the second hop according to formulas (22) and (23) respectively. and transmission energy consumption And ensure that the constraints are met: ,and ,in, This is the upper limit of the transmit power of the local aggregator. This is the lower limit of the signal-to-noise ratio for the local aggregator. (twenty two); (twenty three).

[0012] As a preferred option, in step seven, the process of dynamically updating the AoI value to reflect the timeliness of the data is as follows: S71: Cluster-level AoI calculation; calculate the first AoI according to formula (24). Average information freshness of the cluster ; (twenty four); In the formula, Indicates the AoI of the first hop; Indicates the AoI of the second hop; This represents the number of nodes within the cluster. S72: Dynamically update AoI; First jump update: If satisfied This indicates that the first hop transmission was successful. ,otherwise ; Second jump update: If satisfied This indicates that the second hop transmission was successful. ,otherwise ,in, , These are the delay correction terms for the first hop and the second hop, respectively.

[0013] As a preferred option, the process of constructing the multi-constraint optimization model in step eight is as follows: S81: Optimize the objective function; construct the optimization objective function according to formula (25); (25); In the formula, The total number of time slices for optimization; This represents the total number of node clusters. These are the weighting coefficients. ; For clusters Total energy consumption ; For the first Cluster-level energy efficiency; S82: Construct constraints; construct power interruption probability constraints according to formula (26), construct transmit power constraints according to formula (27), construct signal-to-noise ratio constraints according to formula (28), and construct bandwidth constraints according to formula (29); (26); (27); (28); (29); In the formula, Total available bandwidth, .

[0014] As a preferred option, the process of achieving a coordinated balance between energy consumption and communication performance and closed-loop scheduling in step nine is as follows: S91: Definition of state and action space; state space Includes estimated available power of wind turbines. and instantaneous channel gain and Action space Includes bandwidth Perceiving time and power distribution ,in, These correspond to the first hop and the second hop wireless communication transmission, respectively. S92: State input, action generation, and constraint mapping; S92-1: State Input and Initial Action Generation; This involves generating preprocessed state information (estimated available power of the wind turbine). and instantaneous channel gain and Input the actor network, which, based on the current power-channel joint state, from the action space The initial resource allocation action is generated in the middle; S92-2: Action space constraint mapping; Bandwidth normalization is performed according to formula (30) to ensure that the initial bandwidth meets the total bandwidth constraint; Sensing duration scaling is performed according to formula (31) to ensure that the initial sensing duration action is within the effective range; Transmission power parameterization is performed according to formula (32) to ensure that the initial transmission power action meets the hardware constraints. (30); In the formula, For the first The actual allocated bandwidth for wireless communication; This represents the maximum total bandwidth for the corresponding link. (31); In the formula, For time slices Actual sensing duration of internal sensor nodes; This is the minimum sensing time for the node; This represents the total duration of the current time slice; For element-wise multiplication; (32); In the formula, For time slices Inner Actual transmit power of wireless communication ; , ; , ; , These are the upper and lower limits of the link's transmit power, respectively. S93: Reward function design and immediate reward calculation; Based on the optimization objective, construct the reward function according to formula (33). To quantify the trade-off between timeliness, energy consumption, and violation of constraints; (33); S94: Risk-sensitive reinforcement learning training; S94-1: Risk Adjustment and Objective Function Definition; Incorporating State-Adaptive Risk Coefficient Risk-adjusted rewards are obtained by applying risk adjustment to the rewards, and the exploration benefits and security constraints are dynamically balanced. Based on this, a risk-sensitive objective function is defined according to formula (34). ; (34); In the formula, As a discount factor, ; , , These are two different weighting coefficients. Real-time fluctuation values ​​constrained by the probability of power outage; Representation strategy The variance term below is used to penalize high-risk actions; S94-2: Risk-Sensitive Advantage Estimation; The commentator network receives state information and risk-adjusted returns, and first calculates the time slices according to formulas (35) and (36) respectively. Risk-adjusted cumulative return within the return range and constrained risk-adjusted cumulative return Then, calculate the risk-sensitive advantage function according to formulas (37) and (38) respectively. and constrained advantage function ; (35); (36); (37); (38); In the formula , The current state estimated by the commentator network, respectively. Reward-type and constraint-type state-value functions, , The next state estimated by the commentator network, respectively. Reward-type and constraint-type state-value functions; , These are the weighting coefficients for rewards and constraints, respectively. This is an average of historical returns; The historical average constraint cost; To constrain violations; To constrain relevant cost items; S94-3: Objective function construction; defining the importance sampling ratio based on the dominance estimation results. Combining the PPO pruning mechanism, constrained cost items, and strategy entropy incentive items, the PPO pruning objective function is constructed according to formulas (39) and (40), respectively. Cost term function related to constraints ; (39); (40); In the formula, This is the clipping threshold; This represents the probability of the current policy. The probability is the old strategy probability; S94-4: Actor Network Parameter Update and Constraint Verification; Optimization via Gradient Descent Algorithm and An enhanced objective function is constructed by combining the policy entropy incentive term, and the actor network parameters are adjusted to optimize the action generation strategy. At the same time, according to formula (41), the constraint verification is achieved through Lagrange multiplier update to ensure that the resource allocation action meets the energy supply, power, and signal-to-noise ratio constraints, and an enhanced objective function is constructed according to formula (42). ; (41); In the formula, It is a non-negative Lagrange multiplier; Step size; The threshold for allowing violations; To constrain violations; This indicates the non-negation operation; (42); In the formula, This is the entropy regularization coefficient; This is the policy entropy, used to encourage exploration; S94-5: Critics Network Parameter Update; Optimize State Value Function Estimation Accuracy through Mean Squared Error Regression, and Construct Risk-Sensitive Value Loss Functions for the Critics Network Based on Formulas (43) and (44) respectively. and constrained cost value loss function ; (43); (44); S94-6: Policy convergence judgment and iterative optimization; repeat S94-1 to S94-5 until policy convergence; in multiple consecutive iterations, the enhanced objective function... The fluctuation value is less than the preset threshold, and the constraint violation indicator Continuously below the allowed violation threshold This ensures strategy stability and constraint satisfaction. S94-7: Optimal policy output; after the policy converges, output a stable optimal resource allocation policy.

[0015] This invention relates to a low-power wireless data transmission system and implementation method for a wind-powered underground wireless sensor (WSN) network in coal mines. It is applicable to environmental monitoring and equipment status monitoring scenarios in deep mine environments. The core objective is to address the problems of difficult deployment, short lifespan, and high energy consumption associated with traditional wired / battery-powered WSNs. Furthermore, in the complex scenario of uncertain energy and channel conditions in underground coal mines, it aims to achieve intelligent resource scheduling with the joint objectives of minimizing Information Age (AoI) and maximizing energy efficiency. This core task faces three key challenges: the randomness and intermittency of underground wind energy supply conflict with the real-time requirements of communication tasks, necessitating dynamic strategies to ensure data freshness; the complex and variable system state caused by multiple uncertainties such as wind speed fluctuations and channel time-varying conditions makes traditional optimization methods difficult to adapt, requiring an intelligent adaptive decision-making framework; and resource scheduling must also meet hard communication constraints such as energy interruption probability and minimum signal-to-noise ratio (SNR), maximizing system performance while ensuring long-term network reliability.

[0016] To address the aforementioned intertwined challenges, this invention constructs a complete technical system of "self-powered guarantee - intelligent scheduling - efficiency optimization," with each module working together to form a complete self-powered wireless data transmission system. First, a micro wind energy acquisition module is deployed based on the mine ventilation system to capture wind energy in the roadway in real time and convert it into electrical energy. This is combined with an energy storage unit to dynamically cache electrical energy, providing continuous and stable wireless power supply to WSN nodes, thus avoiding the deployment and maintenance difficulties of traditional power supply methods. Addressing the randomness of wind energy supply, a wind speed-energy mapping model is designed. Combining historical wind speed statistics with real-time monitoring values, the future energy supply intensity is predicted, and the node's operating mode is dynamically adjusted. Data acquisition and transmission are prioritized during high-energy periods, while switching to a low-power sleep state during low-energy periods. Simultaneously, a probability constraint conversion strategy for energy supply interruption under different wind speed distributions is proposed, accurately adapting to truncated Weibull distributions and arbitrary distributions of underground wind speeds, effectively controlling the risk of energy supply interruption. Secondly, a risk-sensitive reinforcement learning proximity policy optimization (PPO) framework with interruption probability constraints is constructed. This framework transforms the random energy interruption constraints and communication hardware constraints into differentiable Lagrange penalty terms, addressing the difficulty of directly solving constraints under uncertainty. Leveraging the dynamic interaction between reinforcement learning and the environment, it intelligently optimizes node sensing cycles, wireless transmission power, and communication bandwidth allocation. Combined with a multi-hop wireless communication architecture and a node-to-RMC cluster-level energy efficiency assessment model, it automatically increases transmission power to ensure successful transmission when channel interference is strong, and reduces power to save energy when channel quality is good, achieving a synergistic balance between low power consumption and data AoI timeliness. Finally, a distributed data fusion strategy is introduced. Neighboring WSN nodes first preprocess and fuse the collected environmental monitoring data locally, removing redundant information before transmitting it to higher-level nodes, further reducing the overall network energy consumption.

[0017] Compared with existing technologies, this method has significant advantages: First, it constructs a reinforcement learning optimization framework under the joint constraints of probabilistic constraints and communication hardware, specifically addressing the inherent conflict between the randomness of wind energy supply and the demand for real-time communication. This improves the robustness of wireless communication in complex and uncertain underground environments, with adaptability far exceeding traditional optimization methods. Second, through a wind speed-energy mapping model and a dedicated interruption probability constraint transformation strategy, it precisely controls the risk of power supply interruption, ensuring long-term stable network operation and outperforming methods without targeted constraint transformation. Third, based on a risk-sensitive reinforcement learning-based dynamic resource allocation method, it achieves adaptive collaborative optimization of sensing duration, transmission power, and bandwidth. Combined with a distributed data fusion strategy, it achieves an efficient balance between low power consumption, high energy utilization, and high real-time communication performance under intermittent wind power supply conditions. Fourth, it utilizes the mine ventilation system to achieve continuous self-powering of the WSN, avoiding the deployment and maintenance difficulties of traditional power supply, significantly extending the continuous working cycle of WSN nodes, and providing long-term and reliable communication support for environmental perception and safety production monitoring in coal mines.

[0018] This method can effectively ensure the continuous and stable supply of energy and the real-time data of underground WSNs. It can effectively solve the problems of difficult deployment, short lifespan and high energy consumption of traditional wired / battery-powered WSNs. At the same time, it can not only adapt to the unstable characteristics of wind power supply in coal mines, but also ensure the real-time and reliable transmission of monitoring data. It is suitable for environmental monitoring and equipment status monitoring scenarios in deep mine environments.

[0019] The present invention also provides a resource dynamic scheduling system for self-powered wireless sensor networks in coal mines, including a micro wind turbine, an energy storage unit, a sensor cluster, a local aggregator, and a remote monitoring center; Multiple micro wind turbines are installed in multiple ventilation tunnels underground to capture airflow energy and convert it into electrical energy; The energy storage unit is connected to multiple micro wind turbines to buffer the electrical energy converted by the wind turbines. Multiple sensor clusters are installed in multiple ventilation tunnels. Each sensor cluster includes multiple sensor nodes, and each sensor node has multiple built-in monitoring sensors. The sensor nodes are used to collect environmental parameters in the corresponding ventilation tunnel and adjust the sensing parameters according to the adjustment signals sent by the local aggregator. The local aggregator is deployed in the mine and integrates a wind power control module and a wireless resource scheduling module; the local aggregator is connected to the energy storage unit and multiple sensor clusters respectively; The wind power supply control module is used to calculate the power supply of the wind turbine based on wind speed data, and to convert the probabilistic constraint of power supply interruption into a deterministic constraint according to the wind speed distribution type. It outputs a power supply status signal containing available power and constraint boundaries. At the same time, it is used to dynamically adjust the state constraint conversion parameters according to power supply fluctuations to ensure the stability of power supply. The wireless resource scheduling module is used to collect the power supply status of the wind power supply control module, the channel status and environmental monitoring data uploaded by the sensor cluster, and the quality feedback of its own transmission link. Based on the risk-sensitive PPO framework and combined with preset constraints, it generates the optimal strategy for sensing duration, transmission power, and bandwidth allocation. At the same time, it is used to send adjustment signals to the sensor cluster, aggregate and process the environmental monitoring data uploaded by the sensor cluster, and then forward it to the remote monitoring center. It is also used to update the PPO model parameters based on the feedback data and iteratively optimize the resource allocation strategy. The remote monitoring center is deployed in a ground monitoring room to receive aggregated data uploaded by the local aggregator, enabling data storage, analysis, and anomaly early warning.

[0020] In this invention, micro wind turbines and sensor clusters are deployed in ventilation tunnels to precisely match energy supply and consumption scenarios. Combined with energy storage units to cache electrical energy, this architectural approach addresses the pain points of traditional power supply deployment difficulties and reliance on batteries, achieving a closed-loop passive power supply. The wind power supply control module calculates the power supply based on wind speed data, transforming probabilistic power interruption constraints into deterministic constraints and dynamically adjusting conversion parameters to effectively cope with wind speed fluctuations and ensure power supply stability. Sensor nodes combine environmental parameter acquisition with adjustable sensing parameters, responding to local aggregator adjustment commands to achieve dynamic "acquisition-adaptation." The wireless resource scheduling module integrates power supply status, channel status, monitoring data, and link feedback, generating optimal strategies based on a risk-sensitive PPO framework. Simultaneously, iteratively updating the model through feedback data ensures that resource allocation (sensing duration, transmission power, bandwidth) accurately adapts to the dynamic downhole environment, balancing power supply constraints and communication needs. The local aggregator, acting as the core hub, handles data acquisition from the sensor cluster, power supply status feedback, and data forwarding from the remote monitoring center. It optimizes data transmission efficiency through aggregation processing. The remote monitoring center handles data storage, analysis, and anomaly warnings, forming a complete data chain of "acquisition-processing-forwarding-monitoring," ensuring the practical value of the monitoring data. The two main modules integrated into the local aggregator focus on power supply stability and resource scheduling respectively. Their precise functional division and smooth interaction ensure the system's efficient and orderly operation.

[0021] The system has a simple structure and a high degree of intelligence, which can ensure the stability and reliability of self-powered wireless data transmission. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method portion of this invention; Figure 2 This is a decision-making flowchart of the risk-sensitive PPO framework in the method section of this invention. Figure 3 This is a schematic diagram of the system components in the invention being assembled in a mine; Figure 4 This is a block diagram of the system component in the invention. Detailed Implementation

[0023] The invention will now be further described with reference to the accompanying drawings.

[0024] like Figure 1 and Figure 2 As shown, the present invention provides a method for dynamic resource scheduling of self-powered wireless sensor networks in coal mines, comprising the following steps; Step 1: Wind energy harvesting and real-time power supply calculation; using micro wind turbines deployed in ventilation roadways in coal mines to capture wind energy and convert it into electrical energy, and using energy storage units to cache electrical energy to provide continuous wireless power supply to WSN nodes, thus solving the problems of short battery life and difficult deployment of wired power supply from the source; Step 2: Obtaining wind speed distribution parameters and determining distribution type; wind speed data in the mine is collected by wind speed sensors deployed in the ventilation tunnel, and Kolmogorov-Smirnov detection is used to determine whether the data conforms to a truncated Weibull distribution in order to determine the wind speed distribution type in the target area in the mine. Step 3: Energy interruption probability constraint conversion based on wind speed distribution type; Based on the wind speed distribution type, the probabilistic energy interruption constraint is converted into a deterministic constraint to ensure the stability of the node cluster's energy supply and provide energy constraint boundaries for subsequent resource scheduling. Step 4: Wind speed-energy mapping and node working mode adjustment; Based on wind speed distribution parameters and energy supply constraints, establish a wind speed-energy mapping model, and combine real-time wind speed monitoring values ​​to predict the energy supply intensity in future periods; During periods of high energy supply, nodes prioritize data acquisition and transmission tasks to maximize data throughput; During periods of low energy supply, nodes switch to a low-power sleep state, retaining only core sensing functions to avoid energy waste. Step 5: Sensor node data sensing and energy consumption calculation; Sensor nodes collect downhole environmental data according to optimized parameters to ensure that sensing energy consumption does not exceed the real-time wind power supply capacity; Step Six: Two-hop wireless communication transmission and energy consumption calculation; A two-hop transmission architecture of sensor node → local aggregator → remote monitoring center is adopted. Interference is reduced by orthogonal sub-channel division, and the transmission rate and energy consumption of each hop are calculated. Step 7: Cluster-level Information Age (AoI) Quantification and Dynamic Update; The cluster-level AoI is used to evaluate the data freshness in two-hop transmission scenarios, and the AoI value is dynamically updated to reflect the data timeliness; Step 8: Optimization Objectives and Constraints Construction; With minimizing the average cluster-level AoI and maximizing the cluster-level energy efficiency as the core objectives, construct a multi-constraint optimization model; Step 9: Dynamic resource allocation based on risk-sensitive reinforcement learning; construct a reinforcement learning framework to intelligently optimize node perception duration, communication bandwidth allocation and transmission power, and achieve a coordinated balance between energy consumption and communication performance and closed-loop scheduling.

[0025] As a preferred option, in step one, the first step is calculated according to formula (1). Clusters in time slices Wind energy harvesting power within ; (1); In the formula, The preset time slice length (values ​​range from 1 to 5 seconds, preferably 2 seconds); air density, ; The swept area of ​​the turbine blades. , The radius of the turbine blade; Energy conversion efficiency; For mechanical efficiency; This represents the real-time wind speed at the turbine.

[0026] As a preferred option, in step two, the process of using the Kolmogorov-Smirnov test to determine whether the data conforms to a truncated Weibull distribution, in order to determine the wind speed distribution type in the downhole target area, is as follows: S21: Construction of historical wind speed dataset; In the ventilation tunnel corresponding to each sensor cluster, deploy wind speed sensors with a measurement accuracy of ±0.1 m / s and a sampling frequency of 1 Hz, continuously collect wind speed data for no less than 72 hours, and construct a historical wind speed dataset. ,in, For the first Second sampling time The wind speed value, For sampling sequence number, This represents the total number of samples. S22: Classification type determination; The Kolmogorov-Smirnov test is used to test the goodness of fit of the historical wind speed dataset. The historical wind speed dataset is compared with the preset definition [0.5 m / s, 8 m / s] (shape parameter). Initial range 1~4, scale parameters Comparison of truncated Weibull distributions (initial range 1~5 m / s): If the probability value corresponding to the test statistic is ≥0.05, the wind speed is determined to follow a truncated Weibull distribution, and the shape parameter is calculated using the maximum likelihood estimation method. Scale parameters and actual domain ,in, This is the minimum wind speed value from historical data. The maximum wind speed value in the historical data is given; if the probability value corresponding to the test statistic is <0.05, the wind speed is determined to be of arbitrary distribution, and the mean wind speed is calculated according to formula (2) and formula (3) respectively. and variance ; (2); (3).

[0027] As a preferred option, in step three, the process of transforming the probabilistic power supply interruption constraint into a deterministic constraint is as follows: S31: Definition of Net Energy Surplus; Wind Energy Harvesting Power Factor According to formula (4), the first... Clusters in time slices Net energy surplus ; (4); In the formula, The system loss coefficient, air density; For the first Total energy consumption of each cluster, including energy consumption for sensing, integration, and transmission; and These represent the minimum and maximum values ​​of the net energy surplus, respectively. , ; S32: Constraint transformation for truncating the Weibull distribution; S32-1: Calculate the mean of the net energy surplus according to formulas (5) and (6) respectively. and variance Among them, the wind speed is calculated according to formula (7). of Step Moment ; (5); (6); (7); In the formula, Wind speed The third moment; Wind speed The 6th moment; An index for a stage or time step; For the first The weighting coefficients for each stage; This represents the value of the gamma function under specific parameters; , These are the upper and lower bound parameters of the gamma function, respectively. , ; To truncate the Weibull distribution function, , , These represent the upper and lower bound parameters of the distribution function, respectively; S32-2: Based on Bernstein's inequality, constrain the probability of power outage. This is transformed into a deterministic constraint, as shown in formula (8), where, For the first A cluster of sensors in time slices The probability of power outage within the area; (8); In the formula, For the first Individuals in time slices The mean of the net internal energy surplus; The threshold value for the probability of power outage is set between 0.1 and 0.2, preferably 0.15. This constraint ensures that the probability of power outage under the truncated Weibull distribution wind speed does not exceed [the threshold value]. ; S33: Constraint transformation under arbitrary distribution; S33-1: First, use the average wind speed. As the starting point, for Performing a second Taylor expansion yields formula (9); then let , , According to formula (10), the quadratic approximate function of net energy surplus is obtained. ; (9); (10); S33-2: Based on the Conditional Value at Risk (CVaR) theory, the interruption probability constraint is transformed into a deterministic constraint in the worst case, as shown in formula (11); (11); In the formula, This is the conditional risk value.

[0028] S33-3: Solve the constraints using semidefinite programming to satisfy... ,in, It is a positive semi-definite matrix of order two. ,satisfy and ; This is the wind speed statistical feature matrix. ; The CVaR constraint is used to relax the variables.

[0029] As a preferred option, in step five, the process of ensuring that the sensed energy consumption does not exceed the real-time wind power supply capacity is as follows: S51: Data acquisition volume calculation; nodes are calculated at a fixed frequency. (Preferred 100Hz) Collect downhole environmental data, and calculate the first according to formula (12). tudi Each node in the time slice The amount of data collected within ; (12); In the formula, The node sensing duration (a variable to be optimized, with a value of 0.1~0.5s); S52: Sensing energy consumption calculation; calculate node sensing energy consumption according to formula (13). and ensure Ensure that the perceived energy consumption does not exceed the real-time power supply capacity of the wind turbine in step one; (13); In the formula, To maintain constant power consumption of the sensing circuit, .

[0030] As a preferred option, the process of calculating the transmission rate and energy consumption for each hop in step six is ​​as follows: S61: Calculation of transmission rate and energy consumption for the first hop (sensor node → local aggregator); S61-1: The 2.4 GHz band is divided into 15 to 20 sub-channels to avoid interference between nodes within the cluster; S61-2: Calculate the path loss of the first hop according to formula (14). , (14); In the formula, For reference path loss, ; This represents the loss index for short-distance underground paths. This represents the distance between the node and the aggregator. ; S61-3: Calculate the channel gain of the first hop according to formula (15). ; (15); In the formula, The small-scale fading coefficient ranges from 0.5 to 1.0 and follows a Nakagami distribution. As the decay factor, For fading power; S61-4: Calculate the transmission rate of the first hop according to formula (16). ; (16); In the formula, The communication bandwidth allocated to the node; This refers to the node's transmit power. Channel noise power; S61-5: Calculate the transmission duration of the first hop according to formulas (17) and (18) respectively. and transmission energy consumption And ensure that the constraints are met: ,and ,in, This represents the upper limit of the sensor node's transmit power. This represents the lower limit of the signal-to-noise ratio for sensor nodes; (17); (18); S62: Calculation of transmission rate and energy consumption for the second hop (local aggregator → remote monitoring center RMC); S62-1: Calculate the path loss of the second hop according to formula (19) ; (19); In the formula, The distance from the aggregator to the remote monitoring center; S62-2: Calculate the transmission rate of the second hop according to formula (20). ; (20); In the formula, This refers to the communication bandwidth of the local aggregator. For the aggregator's transmit power, For aggregator link channel gain; S62-3: Calculate the total amount of data within the cluster according to formula (21). ; (twenty one); In the formula, For the first The set of nodes in a cluster; S62-4: Calculate the transmission duration of the second hop according to formulas (22) and (23) respectively. and transmission energy consumption And ensure that the constraints are met: ,and ,in, This is the upper limit of the transmit power of the local aggregator. This is the lower limit of the signal-to-noise ratio for the local aggregator. (twenty two); (twenty three).

[0031] As a preferred option, in step seven, the process of dynamically updating the AoI value to reflect the timeliness of the data is as follows: S71: Cluster-level AoI calculation; calculate the first AoI according to formula (24). Average information freshness of the cluster ; (twenty four); In the formula, Indicates the AoI of the first hop; Indicates the AoI of the second hop; This represents the number of nodes within the cluster. S72: Dynamically update AoI; First jump update: If satisfied This indicates that the first hop transmission was successful. ,otherwise ; Second jump update: If satisfied This indicates that the second hop transmission was successful. ,otherwise ,in, , These are the delay correction terms for the first hop and the second hop, respectively.

[0032] As a preferred option, the process of constructing the multi-constraint optimization model in step eight is as follows: S81: Optimize the objective function; construct the optimization objective function according to formula (25); (25); In the formula, The total number of time slices for optimization; This represents the total number of node clusters. These are the weighting coefficients. It can be adjusted according to actual needs; For clusters Total energy consumption ; For the first Cluster-level energy efficiency; S82: Construct constraints; construct power interruption probability constraints according to formula (26), construct transmit power constraints according to formula (27), construct signal-to-noise ratio constraints according to formula (28), and construct bandwidth constraints according to formula (29); (26); (27); (28); (29); In the formula, Total available bandwidth, It is used to allocate short-range communication links and LoRa long-range communication links to each sensor cluster, so as to meet the bandwidth requirements for data transmission within the cluster nodes and data uploading between cluster aggregates.

[0033] As a preferred option, the process of achieving a coordinated balance between energy consumption and communication performance and closed-loop scheduling in step nine is as follows: S91: Definition of state and action space; state space Includes estimated available power of wind turbines. and instantaneous channel gain and Action space Includes bandwidth Perceiving time and power distribution ,in, These correspond to the first hop and the second hop wireless communication transmission, respectively. S92: State input, action generation, and constraint mapping; S92-1: State Input and Initial Action Generation; This involves generating preprocessed state information (estimated available power of the wind turbine). and instantaneous channel gain and Input the actor network, which, based on the current power-channel joint state, from the action space Initial resource allocation actions are generated in (bandwidth allocation, sensing duration, and transmit power); S92-2: Action space constraint mapping; Bandwidth normalization is performed according to formula (30) to ensure that the initial bandwidth meets the total bandwidth constraint; Sensing duration scaling is performed according to formula (31) to ensure that the initial sensing duration action is within the effective range; Transmission power parameterization is performed according to formula (32) to ensure that the initial transmission power action meets the hardware constraints. (30); In the formula, For the first The actual allocated bandwidth for wireless communication; This represents the maximum total bandwidth for the corresponding link. (31); In the formula, For time slices Actual sensing duration of internal sensor nodes; This is the minimum sensing time for the node; This represents the total duration of the current time slice; For element-wise multiplication; (32); In the formula, For time slices Inner Actual transmit power of wireless communication ; , ; , ; , These are the upper and lower limits of the link's transmit power, respectively. S93: Reward function design and immediate reward calculation; Based on the optimization objective, construct the reward function according to formula (33). To quantify the trade-off between timeliness, energy consumption, and violation of constraints; (33); S94: Risk-sensitive reinforcement learning training; S94-1: Risk Adjustment and Objective Function Definition; Incorporating State-Adaptive Risk Coefficient Risk-adjusted rewards are obtained by applying risk adjustment to the rewards, and the exploration benefits and security constraints are dynamically balanced. Based on this, a risk-sensitive objective function is defined according to formula (34). ; (34); In the formula, As a discount factor, ; , , These are two different weighting coefficients. Real-time fluctuation values ​​constrained by the probability of power outage; Representation strategy The variance term below is used to penalize high-risk actions; S94-2: Risk-Sensitive Advantage Estimation; The commentator network receives state information and risk-adjusted returns, and first calculates the time slices according to formulas (35) and (36) respectively. Risk-adjusted cumulative return within the return range and constrained risk-adjusted cumulative return Then, calculate the risk-sensitive advantage function according to formulas (37) and (38) respectively. and constrained advantage function ; (35); (36); (37); (38); In the formula , The current state estimated by the commentator network, respectively. Reward-type and constraint-type state-value functions, , The next state estimated by the commentator network, respectively. Reward-type and constraint-type state-value functions; , These are the weighting coefficients for rewards and constraints, respectively. This is an average of historical returns; The historical average constraint cost; To constrain violations; To constrain relevant cost items; S94-3: Objective function construction; defining the importance sampling ratio based on the dominance estimation results. Combining the PPO pruning mechanism, constrained cost items, and strategy entropy incentive items, the PPO pruning objective function is constructed according to formulas (39) and (40), respectively. Cost term function related to constraints ; (39); (40); In the formula, This is the clipping threshold; This represents the probability of the current policy. The probability is the old strategy probability; S94-4: Actor Network Parameter Update and Constraint Verification; Optimization via Gradient Descent Algorithm and An enhanced objective function is constructed by combining the policy entropy incentive term, and the actor network parameters are adjusted to optimize the action generation strategy. At the same time, according to formula (41), the constraint verification is achieved through Lagrange multiplier update to ensure that the resource allocation action meets the constraints of energy supply, power, signal-to-noise ratio, etc., and an enhanced objective function is constructed according to formula (42). ; (41); In the formula, It is a non-negative Lagrange multiplier; Step size; The threshold for allowing violations; To constrain violations; This indicates the non-negation operation; (42); In the formula, This is the entropy regularization coefficient; This is the policy entropy, used to encourage exploration; S94-5: Critics Network Parameter Update; Optimize State Value Function Estimation Accuracy through Mean Squared Error Regression, and Construct Risk-Sensitive Value Loss Functions for the Critics Network Based on Formulas (43) and (44) respectively. and constrained cost value loss function ; (43); (44); S94-6: Policy convergence judgment and iterative optimization; repeat S94-1 to S94-5 until policy convergence; in multiple consecutive iterations, the enhanced objective function... The fluctuation value is less than the preset threshold, and the constraint violation indicator Continuously below the allowed violation threshold This ensures strategy stability and constraint satisfaction. S94-7: Optimal policy output; after policy convergence, output a stable optimal resource allocation policy and specify the sensing duration of sensor nodes. Subchannel allocation and transmit power for the first hop transmission Bandwidth allocation for the second hop transmission of the local aggregator With transmission power Core scheduling parameters; S95: Decision Output and Execution Feedback; S95-1: Strategy Distribution and Execution; The optimal resource allocation strategy is distributed to each sensor node and the local aggregator to guide the actual scheduling execution; the sensor nodes, according to... Adjust the data sensing duration to ensure that sensing energy consumption matches wind power supply; the sub-channels allocated to sensor nodes according to the strategy and Execute the first hop data transfer; the local aggregator allocates data according to the policy. and Perform the second-hop data transmission; S95-2: Execution Data Feedback; Real-time feedback of actual operating data after execution to the status acquisition stage. Feedback data includes: actual energy consumption (wind turbine output power, actual node energy consumption); channel transmission quality (actual signal-to-noise ratio of two-hop links, transmission success rate); AoI changes; constraint satisfaction status (actual number of power interruptions, power / bandwidth constraint violations). S95-3: Closed-loop optimization; after preprocessing, the feedback data is used as the state input for the next cycle to achieve continuous iteration of decision-execution-feedback-optimization, ensuring that the scheduling strategy can adapt to the dynamic changes in underground coal mine power supply and channel environment, and maintain the long-term stable and efficient operation of WSN.

[0034] This invention relates to a low-power wireless data transmission system and implementation method for a wind-powered underground wireless sensor (WSN) network in coal mines. It is applicable to environmental monitoring and equipment status monitoring scenarios in deep mine environments. The core objective is to address the problems of difficult deployment, short lifespan, and high energy consumption associated with traditional wired / battery-powered WSNs. Furthermore, in the complex scenario of uncertain energy and channel conditions in underground coal mines, it aims to achieve intelligent resource scheduling with the joint objectives of minimizing Information Age (AoI) and maximizing energy efficiency. This core task faces three key challenges: the randomness and intermittency of underground wind energy supply conflict with the real-time requirements of communication tasks, necessitating dynamic strategies to ensure data freshness; the complex and variable system state caused by multiple uncertainties such as wind speed fluctuations and channel time-varying conditions makes traditional optimization methods difficult to adapt, requiring an intelligent adaptive decision-making framework; and resource scheduling must also meet hard communication constraints such as energy interruption probability and minimum signal-to-noise ratio (SNR), maximizing system performance while ensuring long-term network reliability.

[0035] To address the aforementioned intertwined challenges, this invention constructs a complete technical system of "self-powered guarantee - intelligent scheduling - efficiency optimization," with each module working together to form a complete self-powered wireless data transmission system. First, a micro wind energy acquisition module is deployed based on the mine ventilation system to capture wind energy in the roadway in real time and convert it into electrical energy. This is combined with an energy storage unit to dynamically cache electrical energy, providing continuous and stable wireless power supply to WSN nodes, thus avoiding the deployment and maintenance difficulties of traditional power supply methods. Addressing the randomness of wind energy supply, a wind speed-energy mapping model is designed. Combining historical wind speed statistics with real-time monitoring values, the future energy supply intensity is predicted, and the node's operating mode is dynamically adjusted. Data acquisition and transmission are prioritized during high-energy periods, while switching to a low-power sleep state during low-energy periods. Simultaneously, a probability constraint conversion strategy for energy supply interruption under different wind speed distributions is proposed, accurately adapting to truncated Weibull distributions and arbitrary distributions of underground wind speeds, effectively controlling the risk of energy supply interruption. Secondly, a risk-sensitive reinforcement learning proximity policy optimization (PPO) framework with interruption probability constraints is constructed. This framework transforms the random energy interruption constraints and communication hardware constraints into differentiable Lagrange penalty terms, addressing the difficulty of directly solving constraints under uncertainty. Leveraging the dynamic interaction between reinforcement learning and the environment, it intelligently optimizes node sensing cycles, wireless transmission power, and communication bandwidth allocation. Combined with a multi-hop wireless communication architecture and a node-to-RMC cluster-level energy efficiency assessment model, it automatically increases transmission power to ensure successful transmission when channel interference is strong, and reduces power to save energy when channel quality is good, achieving a synergistic balance between low power consumption and data AoI timeliness. Finally, a distributed data fusion strategy is introduced. Neighboring WSN nodes first preprocess and fuse the collected environmental monitoring data locally, removing redundant information before transmitting it to higher-level nodes, further reducing the overall network energy consumption.

[0036] Compared with existing technologies, this method has significant advantages: First, it constructs a reinforcement learning optimization framework under the joint constraints of probabilistic constraints and communication hardware, specifically addressing the inherent conflict between the randomness of wind energy supply and the demand for real-time communication. This improves the robustness of wireless communication in complex and uncertain underground environments, with adaptability far exceeding traditional optimization methods. Second, through a wind speed-energy mapping model and a dedicated interruption probability constraint transformation strategy, it precisely controls the risk of power supply interruption, ensuring long-term stable network operation and outperforming methods without targeted constraint transformation. Third, based on a risk-sensitive reinforcement learning-based dynamic resource allocation method, it achieves adaptive collaborative optimization of sensing duration, transmission power, and bandwidth. Combined with a distributed data fusion strategy, it achieves an efficient balance between low power consumption, high energy utilization, and high real-time communication performance under intermittent wind power supply conditions. Fourth, it utilizes the mine ventilation system to achieve continuous self-powering of the WSN, avoiding the deployment and maintenance difficulties of traditional power supply, significantly extending the continuous working cycle of WSN nodes, and providing long-term and reliable communication support for environmental perception and safety production monitoring in coal mines.

[0037] This method can effectively ensure the continuous and stable supply of energy and the real-time data of underground WSNs. It can effectively solve the problems of difficult deployment, short lifespan and high energy consumption of traditional wired / battery-powered WSNs. At the same time, it can not only adapt to the unstable characteristics of wind power supply in coal mines, but also ensure the real-time and reliable transmission of monitoring data. It is suitable for environmental monitoring and equipment status monitoring scenarios in deep mine environments.

[0038] like Figure 3 and Figure 4 As shown, the present invention also provides a resource dynamic scheduling system for self-powered wireless sensor networks in coal mines, including a micro wind turbine, an energy storage unit, a sensor cluster, a local aggregator, and a remote monitoring center; Multiple micro wind turbines are installed in various ventilation roadways underground, preferably in areas with stable airflow within the mine ventilation roadways, to capture wind energy and convert it into electrical energy; as a preferred option, the blade sweep area of ​​the micro wind turbines is 0.2 m². 2 The energy conversion efficiency is not less than 0.4, and the mechanical efficiency is not less than 0.8. The energy storage unit is connected to multiple micro wind turbines to buffer the electrical energy converted by the wind turbines, avoid power outages at nodes caused by power supply fluctuations, and provide continuous and stable wireless power supply for the sensor cluster. Multiple sensor clusters are installed in multiple ventilation tunnels. Each sensor cluster includes multiple sensor nodes evenly distributed in the monitoring area within the cluster. The number of sensor nodes is N, and the value of N ranges from 1 to 10 (preferably 5). Each sensor node has multiple built-in monitoring sensors and a built-in microprocessor, preferably a CC2530. The sensor node's sensing circuit power consumption is ≤5mW, and the sampling frequency is 1~5Hz. It is used to collect environmental parameters in the corresponding ventilation tunnel, including underground gas concentration, temperature, etc., and transmits the data to the local aggregator via a 2.4GHz wireless link. At the same time, it is used to adjust the sensing parameters according to the adjustment signal issued by the local aggregator to adapt to the power supply status and transmission requirements. As a preferred option, the sensor node is only used to transmit the raw environmental data it collects to the local aggregator in real time, without undertaking data processing or forwarding functions, to ensure low power consumption operation of the node; The local aggregator is deployed in the mine. It is equipped with a high-performance microprocessor (such as an STM32H743) and dual communication modules. One module is a 2.4GHz short-range wireless communication module used to receive raw data from all sensor nodes within the cluster. The other is a LoRa long-range communication module (operating frequency 433 / 868MHz, maximum transmit power ≤20dBm) used to establish a data connection with the remote monitoring center (RMC). The local aggregator integrates a wind power supply control module and a wireless resource scheduling module. It connects to the energy storage unit and multiple sensor clusters. The wind power supply control module has a built-in wind speed model. The wind power supply control module is used to calculate the power supply of the wind turbine based on wind speed data, and to convert the probabilistic constraint of power supply interruption into deterministic constraint according to the wind speed distribution type. It outputs a power supply status signal containing available power and constraint boundaries. At the same time, it is used to dynamically adjust the state constraint transformation parameters according to power supply fluctuations to ensure the stability of power supply. The wireless resource scheduling module is used to collect the power supply status of the wind power supply control module and the channel status (channel gain, path loss, etc.) uploaded by the sensor cluster, as well as environmental monitoring data and quality feedback of its own transmission link. Based on the risk-sensitive PPO framework and combined with preset constraints (such as power supply interruption probability constraints, transmit power limits, signal-to-noise ratio lower limit, and bandwidth constraints), it generates the optimal strategy for sensing duration, transmit power, and bandwidth allocation, realizing the coordinated optimization of system low power consumption and communication performance. At the same time, it is used to send adjustment signals to the sensor cluster, and to perform aggregation processing such as data format standardization and redundant data removal on the environmental monitoring data uploaded by the sensor cluster. Then, it forwards the aggregated data to the remote monitoring center through the LoRa long-distance wireless link, realizing centralized management and long-distance transmission of data within the cluster. It is also used to update the PPO model parameters (actor network, critic network weights, Lagrange multipliers, etc.) based on feedback data (actual power consumption, power supply interruption occurrence, channel transmission quality, AOI dynamic changes), and iteratively optimize the resource allocation strategy. The remote monitoring center (RMC) is deployed in a ground monitoring room to receive aggregated data uploaded by a local aggregator, enabling data storage, analysis, and anomaly early warning, thus providing decision support for mine safety management. Preferably, the remote monitoring center (RMC) uses an industrial server.

[0039] In this invention, micro wind turbines and sensor clusters are deployed in ventilation tunnels to precisely match energy supply and consumption scenarios. Combined with energy storage units to cache electrical energy, this architectural approach addresses the pain points of traditional power supply deployment difficulties and reliance on batteries, achieving a closed-loop passive power supply. The wind power supply control module calculates the power supply based on wind speed data, transforming probabilistic power interruption constraints into deterministic constraints and dynamically adjusting conversion parameters to effectively cope with wind speed fluctuations and ensure power supply stability. Sensor nodes combine environmental parameter acquisition with adjustable sensing parameters, responding to local aggregator adjustment commands to achieve dynamic "acquisition-adaptation." The wireless resource scheduling module integrates power supply status, channel status, monitoring data, and link feedback, generating optimal strategies based on a risk-sensitive PPO framework. Simultaneously, iteratively updating the model through feedback data ensures that resource allocation (sensing duration, transmission power, bandwidth) accurately adapts to the dynamic downhole environment, balancing power supply constraints and communication needs. The local aggregator, acting as the core hub, handles data acquisition from the sensor cluster, power supply status feedback, and data forwarding from the remote monitoring center. It optimizes data transmission efficiency through aggregation processing. The remote monitoring center handles data storage, analysis, and anomaly warnings, forming a complete data chain of "acquisition-processing-forwarding-monitoring," ensuring the practical value of the monitoring data. The two main modules integrated into the local aggregator focus on power supply stability and resource scheduling respectively. Their precise functional division and smooth interaction ensure the system's efficient and orderly operation.

[0040] The system has a simple structure and a high degree of intelligence, which can ensure the stability and reliability of self-powered wireless data transmission.

Claims

1. A method for dynamic resource scheduling in self-powered wireless sensor networks in underground coal mines, characterized in that, Includes the following steps; Step 1: Wind energy harvesting and real-time power supply calculation; using micro wind turbines deployed in ventilation roadways in coal mines to capture wind energy and convert it into electrical energy, and using energy storage units to cache the electrical energy to provide continuous wireless power supply to WSN nodes; Step 2: Obtaining wind speed distribution parameters and determining distribution type; wind speed data in the mine is collected by wind speed sensors deployed in the ventilation tunnel, and Kolmogorov-Smirnov detection is used to determine whether the data conforms to a truncated Weibull distribution in order to determine the wind speed distribution type in the target area in the mine. Step 3: Power outage probability constraint conversion based on wind speed distribution type; Based on the wind speed distribution type, the probabilistic power supply interruption constraint is transformed into a deterministic constraint to ensure the stability of the power supply of the node cluster. Step 4: Wind speed-energy mapping and node operating mode adjustment; Based on wind speed distribution parameters and energy supply constraints, establish a wind speed-energy mapping model, and combine real-time wind speed monitoring values ​​to predict the energy supply intensity in future periods. During periods of high energy supply, nodes prioritize data acquisition and transmission tasks to maximize data throughput. During periods of low energy supply, nodes switch to a low-power sleep state, retaining only core sensing functions to avoid energy waste; Step 5: Sensor node data sensing and energy consumption calculation; Sensor nodes collect downhole environmental data according to optimized parameters to ensure that sensing energy consumption does not exceed the real-time wind power supply capacity; Step Six: Two-hop wireless communication transmission and energy consumption calculation; A two-hop transmission architecture of sensor node → local aggregator → remote monitoring center is adopted. Interference is reduced by orthogonal sub-channel division, and the transmission rate and energy consumption of each hop are calculated. Step 7: Cluster-level information age quantification and dynamic update; evaluate data freshness in two-hop transmission scenarios through cluster-level AoI, and dynamically update AoI values ​​to reflect data timeliness; Step 8: Optimization Objectives and Constraints Construction; With minimizing the average cluster-level AoI and maximizing the cluster-level energy efficiency as the core objectives, construct a multi-constraint optimization model; Step 9: Dynamic resource allocation based on risk-sensitive reinforcement learning; construct a reinforcement learning framework to intelligently optimize node perception duration, communication bandwidth allocation and transmission power, and achieve a coordinated balance between energy consumption and communication performance and closed-loop scheduling.

2. The resource dynamic scheduling method for self-powered wireless sensor networks in coal mines according to claim 1, characterized in that, In step one, the first step is calculated according to formula (1). Clusters in time slices Wind energy harvesting power within ; （1）； In the formula, The preset time slice length; air density, ; The swept area of ​​the turbine blades. , The radius of the turbine blade; Energy conversion efficiency; For mechanical efficiency; This represents the real-time wind speed at the turbine.

3. A resource dynamic scheduling method for self-powered wireless sensor networks in coal mines according to claim 2, characterized in that, In step two, the process of using the Kolmogorov-Smirnov test to determine whether the data conforms to a truncated Weibull distribution, in order to determine the wind speed distribution type in the downhole target area, is as follows: S21: Construction of historical wind speed dataset; In the ventilation tunnel corresponding to each sensor cluster, deploy wind speed sensors with a measurement accuracy of ±0.1 m / s and a sampling frequency of 1 Hz, continuously collect wind speed data for no less than 72 hours, and construct a historical wind speed dataset. ,in, For the first Second sampling time The wind speed value, For sampling sequence number, This represents the total number of samples. S22: Classification type determination; The Kolmogorov-Smirnov test is used to test the goodness of fit of the historical wind speed dataset. The historical wind speed dataset is compared with a truncated Weibull distribution defined as [0.5 m / s, 8 m / s]. If the probability value corresponding to the test statistic is ≥0.05, the wind speed is determined to follow a truncated Weibull distribution. The shape parameter is calculated by the maximum likelihood estimation method. Scale parameters and actual domain ,in, This is the minimum wind speed value from historical data. The maximum wind speed value in the historical data is given; if the probability value corresponding to the test statistic is <0.05, the wind speed is determined to be of arbitrary distribution, and the mean wind speed is calculated according to formula (2) and formula (3) respectively. and variance ; （2）； （3）。 4. A resource dynamic scheduling method for self-powered wireless sensor networks in coal mines according to claim 3, characterized in that, In step three, the process of transforming the probabilistic power supply interruption constraint into a deterministic constraint is as follows: S31: Definition of Net Energy Surplus; Wind Energy Harvesting Power Factor According to formula (4), the first... Clusters in time slices Net energy surplus ; （4）； In the formula, The system loss coefficient, air density; For the first Total energy consumption of each cluster, including energy consumption for sensing, integration, and transmission; and These represent the minimum and maximum values ​​of the net energy surplus, respectively. , ; S32: Constraint transformation for truncating the Weibull distribution; S32-1: Calculate the mean of the net energy surplus according to formulas (5) and (6) respectively. and variance Among them, the wind speed is calculated according to formula (7). of Step Moment ; （5）； （6）； （7）； In the formula, Wind speed The third moment; Wind speed The 6th moment; An index for a stage or time step; For the first The weighting coefficients for each stage; This represents the value of the gamma function under specific parameters; , These are the upper and lower bound parameters of the gamma function, respectively. , ; To truncate the Weibull distribution function, , , These represent the upper and lower bound parameters of the distribution function, respectively; S32-2: Based on Bernstein's inequality, constrain the probability of power outage. This is transformed into a deterministic constraint, as shown in formula (8), where, For the first A cluster of sensors in time slices The probability of power outage within the area; （8）； In the formula, For the first Individuals in time slices The mean of the net internal energy surplus; The threshold value for the probability of power outage is 0.1 to 0.

2. S33: Constraint transformation under arbitrary distribution; S33-1: First, use the average wind speed. As the starting point, for Performing a second Taylor expansion yields formula (9); then let , , According to formula (10), the quadratic approximate function of net energy surplus is obtained. ; （9）； （10）； S33-2: Based on the conditional value at risk theory, the interruption probability constraint is transformed into a deterministic constraint in the worst case, as shown in formula (11); （11）； In the formula, Conditional risk value; S33-3: Solve the constraints using semidefinite programming to satisfy... ,in, It is a positive semi-definite matrix of order two. ,satisfy and ; This is the wind speed statistical feature matrix. ; The CVaR constraint is used to relax the variables.

5. A resource dynamic scheduling method for self-powered wireless sensor networks in coal mines according to claim 4, characterized in that, In step five, the process of ensuring that the sensed energy consumption does not exceed the real-time wind power supply capacity is as follows: S51: Data acquisition volume calculation; nodes are calculated at a fixed frequency. Collect downhole environmental data and calculate the first step according to formula (12). tudi Each node in the time slice The amount of data collected within ; （12）； In the formula, For node sensing duration; S52: Sensing energy consumption calculation; calculate node sensing energy consumption according to formula (13). and ensure ; （13）； In the formula, To maintain constant power consumption of the sensing circuit, .

6. A resource dynamic scheduling method for self-powered wireless sensor networks in coal mines according to claim 5, characterized in that, In step six, the process of calculating the transmission rate and energy consumption for each hop is as follows: S61: Calculation of first-hop transmission rate and energy consumption; S61-1: The 2.4 GHz band is divided into 15 to 20 sub-channels to avoid interference between nodes within the cluster; S61-2: Calculate the path loss of the first hop according to formula (14). , （14）； In the formula, For reference path loss, ; This represents the loss index for short-distance underground paths. This represents the distance between the node and the aggregator. ; S61-3: Calculate the channel gain of the first hop according to formula (15). ; （15）； In the formula, The small-scale fading coefficient ranges from 0.5 to 1.0 and follows a Nakagami distribution. As the decay factor, This refers to the fading power. S61-4: Calculate the transmission rate of the first hop according to formula (16). ; （16）； In the formula, The communication bandwidth allocated to the node; This refers to the node's transmit power. Channel noise power; S61-5: Calculate the transmission duration of the first hop according to formulas (17) and (18) respectively. and transmission energy consumption And ensure that the constraints are met: ,and ,in, This represents the upper limit of the sensor node's transmit power. This represents the lower limit of the signal-to-noise ratio for sensor nodes; （17）； （18）； S62: Calculation of second-hop transmission rate and energy consumption; S62-1: Calculate the path loss of the second hop according to formula (19) ; （19）； In the formula, The distance from the aggregator to the remote monitoring center; S62-2: Calculate the transmission rate of the second hop according to formula (20). ; （20）； In the formula, This refers to the communication bandwidth of the local aggregator. For the aggregator's transmit power, For aggregator link channel gain; S62-3: Calculate the total amount of data within the cluster according to formula (21). ; （21）； In the formula, For the first The set of nodes in a cluster; S62-4: Calculate the transmission duration of the second hop according to formulas (22) and (23) respectively. and transmission energy consumption And ensure that the constraints are met: ,and ,in, This is the upper limit of the transmit power of the local aggregator. This is the lower limit of the signal-to-noise ratio for the local aggregator. （22）； （23）。 7. A resource dynamic scheduling method for self-powered wireless sensor networks in coal mines according to claim 6, characterized in that, In step seven, the process of dynamically updating the AoI value to reflect the timeliness of the data is as follows: S71: Cluster-level AoI calculation; calculate the first AoI according to formula (24). Average information freshness of the cluster ; （24）； In the formula, Indicates the AoI of the first hop; Indicates the AoI of the second hop; This represents the number of nodes within the cluster. S72: Dynamically update AoI; First jump update: If satisfied This indicates that the first hop transmission was successful. ,otherwise ; Second jump update: If satisfied This indicates that the second hop transmission was successful. ,otherwise ,in, , These are the delay correction terms for the first hop and the second hop, respectively.

8. A resource dynamic scheduling method for self-powered wireless sensor networks in coal mines according to claim 7, characterized in that, In step eight, the process of constructing the multi-constraint optimization model is as follows: S81: Optimize the objective function; construct the optimization objective function according to formula (25); （25）； In the formula, The total number of optimization time slices; This represents the total number of node clusters. These are the weighting coefficients. ; For clusters Total energy consumption ; For the first Cluster-level energy efficiency; S82: Construct constraints; construct power interruption probability constraints according to formula (26), construct transmit power constraints according to formula (27), construct signal-to-noise ratio constraints according to formula (28), and construct bandwidth constraints according to formula (29); （26）； （27）； （28）； （29）； In the formula, Total available bandwidth, .

9. A resource dynamic scheduling method for self-powered wireless sensor networks in coal mines according to claim 8, characterized in that, In step nine, the process of achieving a coordinated balance between energy consumption and communication performance, as well as closed-loop scheduling, is as follows: S91: Definition of state and action space; state space Includes estimated available power of wind turbines. and instantaneous channel gain and Action space Includes bandwidth Perceiving time and power distribution ,in, These correspond to the first hop and the second hop wireless communication transmission, respectively. S92: State input, action generation, and constraint mapping; S92-1: State Input and Initial Action Generation; This involves generating preprocessed state information (estimated available power of the wind turbine). and instantaneous channel gain and Input the actor network, which, based on the current power-channel joint state, from the action space The initial resource allocation action is generated in the middle; S92-2: Action space constraint mapping; Bandwidth normalization is performed according to formula (30) to ensure that the initial bandwidth meets the total bandwidth constraint; Sensing duration scaling is performed according to formula (31) to ensure that the initial sensing duration action is within the effective range; Transmission power parameterization is performed according to formula (32) to ensure that the initial transmission power action meets the hardware constraints. （30）； In the formula, For the first The actual allocated bandwidth for wireless communication; This represents the maximum total bandwidth for the corresponding link. （31）； In the formula, For time slices Actual sensing duration of internal sensor nodes; This is the minimum sensing time for the node; This represents the total duration of the current time slice; For element-wise multiplication; （32）； In the formula, For time slices Inner Actual transmit power of wireless communication ; , ; , ; , These are the upper and lower limits of the link's transmit power, respectively. S93: Reward function design and immediate reward calculation; Based on the optimization objective, construct the reward function according to formula (33). To quantify the trade-off between timeliness, energy consumption, and violation of constraints; （33）； S94: Risk-sensitive reinforcement learning training; S94-1: Risk Adjustment and Objective Function Definition; Incorporating State-Adaptive Risk Coefficient Risk-adjusted rewards are obtained by applying risk adjustment to the rewards, and the exploration benefits and security constraints are dynamically balanced. Based on this, a risk-sensitive objective function is defined according to formula (34). ; （34）； In the formula, As a discount factor, ; , , These are two different weighting coefficients. Real-time fluctuation values ​​constrained by the probability of power outage; Representation Strategy The variance term below is used to penalize high-risk actions; S94-2: Risk-Sensitive Advantage Estimation; The commentator network receives state information and risk-adjusted returns, and first calculates the time slices according to formulas (35) and (36) respectively. Risk-adjusted cumulative return within the return range and constrained risk-adjusted cumulative return Then, calculate the risk-sensitive advantage function according to formulas (37) and (38) respectively. and constrained advantage function ; （35）； （36）； （37）； （38）； In the formula , The current state estimated by the commentator network, respectively. Reward-type and constraint-type state-value functions, , The next state estimated by the commentator network, respectively. Reward-type and constraint-type state-value functions; , These are the weighting coefficients for rewards and constraints, respectively. This is an average of historical returns; The historical average constraint cost; To constrain violations; To constrain relevant cost items; S94-3: Objective function construction; defining the importance sampling ratio based on the dominance estimation results. Combining the PPO pruning mechanism, constrained cost items, and strategy entropy incentive items, the PPO pruning objective function is constructed according to formulas (39) and (40), respectively. Cost term function related to constraints ; （39）； （40）； In the formula, This is the clipping threshold; The probability of the current policy; The probability is the old strategy probability; S94-4: Actor Network Parameter Update and Constraint Verification; Optimization via Gradient Descent Algorithm and An enhanced objective function is constructed by combining the policy entropy incentive term, and the actor network parameters are adjusted to optimize the action generation strategy. At the same time, according to formula (41), the constraint verification is achieved through Lagrange multiplier update to ensure that the resource allocation action meets the energy supply, power, and signal-to-noise ratio constraints, and an enhanced objective function is constructed according to formula (42). ; （41）； In the formula, It is a non-negative Lagrange multiplier; Step size; The threshold for allowing violations; To constrain violations; This indicates the non-negation operation; （42）； In the formula, This is the entropy regularization coefficient; This is the policy entropy, used to encourage exploration; S94-5: Critics Network Parameter Update; Optimize State Value Function Estimation Accuracy through Mean Squared Error Regression, and Construct Risk-Sensitive Value Loss Function for the Critics Network Based on Formulas (43) and (44) respectively. and constrained cost-value loss function ; （43）； （44）； S94-6: Policy convergence judgment and iterative optimization; repeat S94-1 to S94-5 until policy convergence; in multiple consecutive iterations, the enhanced objective function... The fluctuation value is less than the preset threshold, and the constraint violation indicator Continuously below the allowed violation threshold This ensures strategy stability and constraint satisfaction. S94-7: Optimal policy output; after the policy converges, output a stable optimal resource allocation policy.

10. A resource dynamic scheduling system for self-powered wireless sensor networks in underground coal mines, characterized in that, This includes micro wind turbines, energy storage units, sensor clusters, local aggregators, and remote monitoring centers; Multiple micro wind turbines are installed in multiple ventilation tunnels underground to capture airflow energy and convert it into electrical energy; The energy storage unit is connected to multiple micro wind turbines to buffer the electrical energy converted by the wind turbines. Multiple sensor clusters are installed in multiple ventilation tunnels. Each sensor cluster includes multiple sensor nodes, and each sensor node has multiple built-in monitoring sensors. The sensor nodes are used to collect environmental parameters in the corresponding ventilation tunnel and adjust the sensing parameters according to the adjustment signals sent by the local aggregator. The local aggregator is deployed in the mine and integrates a wind power control module and a wireless resource scheduling module; the local aggregator is connected to the energy storage unit and multiple sensor clusters respectively; The wind power supply control module is used to calculate the power supply of the wind turbine based on wind speed data, and to convert the probabilistic constraint of power supply interruption into a deterministic constraint according to the wind speed distribution type. It outputs a power supply status signal containing available power and constraint boundaries. At the same time, it is used to dynamically adjust the state constraint conversion parameters according to power supply fluctuations to ensure the stability of power supply. The wireless resource scheduling module is used to collect the power supply status of the wind power supply control module, the channel status and environmental monitoring data uploaded by the sensor cluster, and the quality feedback of its own transmission link. Based on the risk-sensitive PPO framework and combined with preset constraints, it generates the optimal strategy for sensing duration, transmission power, and bandwidth allocation. At the same time, it is used to send adjustment signals to the sensor cluster, aggregate and process the environmental monitoring data uploaded by the sensor cluster, and then forward it to the remote monitoring center. It is also used to update the PPO model parameters based on the feedback data and iteratively optimize the resource allocation strategy. The remote monitoring center is deployed in a ground monitoring room to receive aggregated data uploaded by the local aggregator, enabling data storage, analysis, and anomaly early warning.

Images