A wellhead monitoring data transmission delay optimization system based on wireless communication

Abstract

The application relates to the technical field of artificial intelligence communication optimization, and discloses a wellhead monitoring data transmission delay optimization system based on wireless communication, which is suitable for complex environments such as oil and gas field drilling sites, coal mine wellheads and geothermal wellheads. The system comprises a sensing and collecting unit, a field RTU, an explosion-proof edge computing gateway and a backhaul side edge access server. By introducing a diffusion bridge-TCN-GRU delay prediction model and a double-target Actor-Critic scheduling mechanism in the edge computing gateway, multi-link state perception, delay prediction and intelligent scheduling are realized. The system can dynamically distribute and redundantly return between cellular, NB-IoT, LoRa and satellite links, and combine an EDF-LSTF hybrid scheduling algorithm to optimize message shaping and priority management. The application effectively improves the real-time performance, stability and energy efficiency of wellhead monitoring data transmission, and enhances the communication reliability and safety under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence communication optimization technology, and in particular to a wellhead monitoring data transmission delay optimization system based on wireless communication. Background Technology

[0002] Existing monitoring systems at oil and gas field drilling sites, coal mine wellheads, and geothermal wellheads generally use RTUs and single wireless communication links for data upload. However, in complex geological and electromagnetic environments, these systems are susceptible to signal attenuation, link jitter, and network congestion, resulting in high data transmission delays, high packet loss rates, and poor real-time performance. Traditional systems often employ fixed thresholds or static link switching strategies, failing to dynamically model communication link states or predict latency, making multi-link collaboration and resource optimization difficult. Furthermore, field terminals lack edge intelligent processing capabilities, relying on a central server for unified analysis, leading to sluggish response and increased energy consumption. Therefore, it is necessary to design a wellhead monitoring data transmission delay optimization system that combines artificial intelligence prediction and edge computing scheduling mechanisms. This system should enable link state awareness, latency prediction, and intelligent multi-path scheduling in complex environments, improving the stability and real-time reliability of wellhead data transmission. Summary of the Invention

[0003] This invention proposes a wellhead monitoring data transmission delay optimization system based on wireless communication. Addressing the issue of insufficient real-time data transmission in complex environments such as oil and gas field drilling sites, coal mine wellheads, and geothermal wellheads, it constructs a multi-link adaptive transmission architecture integrating edge computing and artificial intelligence. The system achieves intelligent shaping and multi-path backhaul of monitoring data by embedding time synchronization, link awareness, latency prediction, and scheduling decision modules within an explosion-proof edge computing gateway. Its innovation lies in introducing a diffusion bridge-TCN-GRU model to perform rolling prediction of future latency for wireless backhaul links, combining this with a dual-objective Actor-Critic algorithm to simultaneously optimize transmission latency and energy consumption, and utilizing an EDF-LSTF hybrid scheduling mechanism to achieve data priority control and rate shaping for different service types. This system can dynamically switch and offload traffic among multiple wireless links (5G / LTE, NB-IoT, LoRa, and satellite communication), significantly reducing end-to-end latency fluctuations, improving the stability, reliability, and energy efficiency of monitoring data transmission, and providing high real-time communication support for wellhead safety monitoring and remote control.

[0004] This invention provides a wellhead monitoring data transmission delay optimization system based on wireless communication, applicable to oil and gas field drilling sites, coal mine wellheads, and geothermal wellheads. The system includes:

[0005] The sensing and acquisition unit acquires multi-source monitoring signals;

[0006] The field RTU is electrically connected with the sensing and collecting unit, used for A / D conversion and preprocessing of the multi-source monitoring signals to obtain processed multi-source monitoring data; the RTU unit is provided with an RS-485 communication interface and DI / DO, AI / AO input and output channels, and is connected with the explosion-proof edge computing gateway through an industrial Ethernet;

[0007] The explosion-proof edge computing gateway comprises a time synchronization and timestamp module, a link sensing module, a delay prediction module, a scheduling decision module, a queue and shaping module and a multi-link backhaul module; the processed multi-source monitoring data are encapsulated into communication messages to generate scheduling and shaping message data; the explosion-proof edge computing gateway is wirelessly connected with the edge access server through a wireless backhaul link to realize multi-path uplink transmission and redundant backhaul of the scheduling and shaping message data; the wireless backhaul link comprises a cellular communication link, a low-power wide-area link and a satellite communication link;

[0008] The backhaul-side edge access server is deployed in the well site control room, receives the scheduling and shaping message data, and performs timestamp alignment, fusion and recombination and cache shaping operations thereon to complete consistency merging of the multi-link data; and implements error recovery, congestion control and flow management mechanisms at the link layer to ensure the integrity and continuity of data transmission; the scheduling and shaping message data are uploaded to the upstream SCADA system, data lake and historical database through OPC UA, MQTT and HTTPS standard interfaces to realize real-time monitoring of the field device running state, archiving management and intelligent analysis of the running data.

[0009] Further, the explosion-proof edge computing gateway realizes multi-path uplink transmission and redundant backhaul of the scheduling and shaping message data, and specifically comprises the following contents:

[0010] The time synchronization and timestamp module: establishes a unified time reference to ensure the time consistency and synchronization accuracy of the processed multi-source monitoring data within the edge computing gateway;

[0011] The link sensing module: takes the unified time reference as a synchronization reference, performs real-time communication state monitoring and feature extraction on the wireless backhaul link between the explosion-proof edge computing gateway and the backhaul-side edge access server, generates link feature data, performs time synchronization and time sequence normalization processing on the link feature data, and forms a link state sequence;

[0012] The delay prediction module takes the link state sequence as input, adopts the diffusion bridge-TCN-GRU model to perform rolling prediction on the end-to-end delay in the future 1-30 s, and generates a delay prediction result; the diffusion bridge-TCN-GRU model is constructed in the following manner: a TCN-GRU network is established, the TCN-GRU network is used to extract short-term and long-term features of the link state sequence and output a delay point prediction result, an adaptive diffusion bridge model based on fractional Brownian motion is constructed with reference to the delay point prediction result, the delay evolution is modeled and sampled at the distribution level, and after joint training by a joint loss function, the diffusion bridge-TCN-GRU model is constructed;

[0013] The scheduling decision module takes the link state sequence and the delay prediction result as input based on a dual-objective Actor-Critic model, and outputs an optimal transmission strategy with delay and energy consumption as joint optimization targets;

[0014] The queue and shaping module establishes four-level priority queues for alarm, control, periodic telemetry and batch file, takes the optimal transmission strategy output by the scheduling decision module as a parameter input, determines the queuing order of various types of service message data encapsulated from the processed multi-source monitoring data by using a mixed scheduling algorithm of EDF (Earliest-Deadline-First) and LSTF (Least-Slack-Time-First), and performs rate limiting and burst traffic shaping through a token bucket mechanism to generate scheduling and shaping message data that meets the link scheduling requirements;

[0015] The multi-link backhaul module configures the activation state, bandwidth allocation and shunting ratio of the cellular communication link, the low-power wide-area link and the satellite communication link according to the optimal transmission strategy, performs shunting and link aggregation of the MPTCP / QUIC protocol, realizes multi-path uplink transmission and redundant backhaul of the encapsulated message, ensures the reliability and optimal delay of data transmission, and sends the scheduling and shaping message data uplink to the backhaul side edge access server.

[0016] Further, the process of taking the link state sequence as input, adopting the diffusion bridge-TCN-GRU model to perform rolling prediction on the end-to-end delay in the future 1-30 s, and generating a delay prediction result, specifically includes the following steps:

[0017] Step S1: input the link state sequence into the TCN-GRU network, sequentially extract short-term and long-term time series features through the time series convolution layer and the gated recurrent layer, and output the end-to-end delay point prediction value in the future 1-30 s;

[0018] Step S2: the Hurst index is introduced to quantitatively model the long memory characteristics of the link state sequence, and a fractional Brownian motion model is formed; by estimating the Hurst index, the long-range dependence characteristics of the link delay can be determined; by constructing an Ornstein-Uhlenbeck (OU) process linear combination with a total layer number K, the fractional Brownian motion model is approximated by Markov, the modeling process converts the non-Markov type delay sequence into an analyzable random dynamics system, forms an augmented state space containing a principal variable and a covariant, and thus establishes a Markovized fractional Brownian motion model;

[0019] Step S3: the partial pinning fractional diffusion bridge mechanism is introduced to dynamically constrain and correct the random drift of the Markovized fractional Brownian motion model, realize controllable random propagation of the delay evolution, and form a diffusion bridge model; specifically, the principal variable is taken as a reference process, terminal constraints are only applied to the principal variable under the premise of maintaining the randomness of the covariant, the partial pinning fractional diffusion bridge equation is constructed, the analyzable drift term and diffusion term are introduced in the augmented state space, the Markovized fractional Brownian motion model has target-oriented and numerical solvability while maintaining the long memory characteristics, and thus the diffusion bridge model is formed;

[0020] Step S4: the time neural network is introduced as a learnable drift approximation function to model and dynamically correct the analyzable drift term, and an adaptive diffusion bridge model is formed;

[0021] Step S5: in the training stage, the point prediction output by the TCN-GRU network and the distribution characteristics output by the diffusion bridge model are taken as joint optimization objectives, and a joint loss function composed of a mean constraint term and a distribution constraint term is constructed; by minimizing the joint loss function, the TCN-GRU network parameters and the adaptive diffusion bridge model parameters are updated synchronously, and the collaborative learning of short-term trend prediction and long-range distribution modeling is realized;

[0022] Step S6: after the training is completed, the delay point prediction value output by the TCN-GRU network is input into the adaptive diffusion bridge model as a reference trajectory to drive it to perform multi-path random sampling, and a delay distribution sequence of 1-30 s in the future is generated; the quantile curve and the confidence interval width are calculated according to the delay distribution sequence, and are fused with the center estimation result of the delay point prediction value, and finally the delay prediction result is output.

[0023] By adopting the above scheme, the application has the following beneficial effects:

[0024] The application realizes the foresight modeling and dynamic prediction of the wireless backhaul link delay by integrating the diffusion bridge-TCN-GRU delay prediction model and the link state perception mechanism in the explosion-proof edge computing gateway. The mechanism can perceive the signal characteristic changes of cellular, low-power wide-area and satellite links in real time under the complex environment of wellhead (high temperature, high humidity, strong electromagnetic interference), and identify the potential delay fluctuation trend in advance, thereby reserving an optimization window before the scheduling decision. Compared with the traditional fixed threshold switching mode, the application effectively improves the real-time transmission accuracy and prediction a priori of the wellhead monitoring data, solves the problems of delay surge and data congestion caused by the sudden change of link state in the past, and enhances the dynamic robustness and adaptive ability of the system in the non-stationary channel environment.

[0025] The application realizes the joint optimization of transmission delay and energy consumption by constructing a scheduling decision mechanism based on a double-target Actor-Critic reinforcement learning algorithm. The system can dynamically output the optimal transmission strategy according to the diffusion bridge model prediction result, intelligently distribute and select the main and standby links for the multi-path channel, so that different types of business messages (alarm, control, telemetry, batch) can achieve a balance between delay sensitivity and energy consumption. The mechanism significantly improves the bandwidth utilization and energy efficiency ratio of the wellhead data link, solves the problems of high energy consumption and low distribution efficiency in existing multi-link communication, and ensures the stable uplink and low-power operation of monitoring data in complex terrain and multi-work area scenarios.

[0026] In addition, the application introduces an EDF-LSTF hybrid scheduling and token bucket traffic shaping mechanism to realize real-time management and control of different priority services and smooth output of burst traffic. In the scene of wellhead burst alarm or high-frequency acquisition, the mechanism can prioritize the immediate transmission of critical safety data, while avoiding communication congestion and message accumulation through adaptive speed limiting and buffer shaping. Through this mechanism, the application significantly improves the real-time response capability and data continuity of the wellhead monitoring system, solves the problems of uncertain data delay and declining reliability of traditional systems under high load, and enhances the reliable communication guarantee capability and field intelligence level of wellhead safety monitoring and remote control. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 A system structure block diagram of a wellhead monitoring data transmission delay optimization system based on wireless communication is proposed for the application;

[0028] Figure 2 A TCN-GRU network structure diagram is proposed in Example Three.

[0029] Figure 1 In the above, RTU represents: Remote Terminal Unit, remote terminal unit; Figure 2In the specific implementation, TCN represents a temporal convolutional network, and GRU represents a gated recurrent unit. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0031] In an embodiment, the wellhead monitoring data transmission delay optimization system based on wireless communication is applied to an oil and gas field drilling site, a coal mine wellhead and a geothermal wellhead. Figure 1 The system comprises the following components.

[0032] The sensing and collecting unit is arranged at the wellhead equipment and auxiliary devices, including an oil (gas) tree, a throttle valve, a casing and tubing pressure port, a wellhead blowout preventer, a geothermal recharge valve group and a coal mine ventilation wellhead monitoring point, and is configured with a 4-20 mA intrinsic safety type pressure transmitter, a HART communication temperature instrument and a pulse counting type flow sensor, for acquiring multiple source monitoring signals, including wellhead pressure, temperature, casing pressure, oil pressure, flow, vibration amplitude, valve position state and gas concentration (H2S, CH4, CO2, CO).

[0033] The field RTU is electrically connected with the sensing and collecting unit, for A / D conversion and preprocessing of the multiple source monitoring signals, to acquire processed multiple source monitoring data. The RTU unit is provided with an RS-485 communication interface and DI / DO, AI / AO input and output channels, is connected with an explosion-proof edge computing gateway through an industrial Ethernet, and realizes real-time transmission and safety control of wellhead data.

[0034] The explosion-proof edge computing gateway adopts an aluminum alloy explosion-proof shell, has a protection level of IP66, an explosion-proof level of Ex dIIB T4 Gb, is suitable for a wellhead environment with high temperature, high humidity and explosive gas, has a working temperature range of -40°C to +75°C, is provided with a GNSS (GPS / Beidou) time service module and a hardware watchdog, supports DC 12-36 V wide power supply and reserves a solar MPPT charging interface. The gateway integrates multiple functional modules, including a time synchronization and timestamp module, a link sensing module, a delay prediction module, a scheduling decision module, a queue and shaping module and a multi-link backhaul module. The gateway encapsulates the processed multi-source monitoring data into a communication message, generates scheduling and shaping message data, and realizes multi-path uplink transmission and redundant backhaul of the scheduling and shaping message data through a wireless backhaul link and a wireless communication connection with an edge access server. The wireless backhaul link includes a cellular communication link (5G / LTE), a low-power wide-area link (NB-IoT, LoRa) and a satellite communication link (LEO / Ku band).

[0035] The backhaul-side edge access server is deployed in a well site control room, receives the scheduling and shaping message data, and performs timestamp alignment, fusion reorganization and cache shaping operations on the scheduling and shaping message data, completes consistent merging of multi-link data, and implements error recovery, congestion control and flow management mechanisms at the link layer to ensure the integrity and continuity of data transmission. The scheduling and shaping message data is uploaded to an upstream SCADA system, a data lake and a historical database through OPC UA, MQTT and HTTPS standard interfaces, to realize real-time monitoring of the running state of field devices, archival management of operation data and intelligent analysis.

[0036] In embodiment two, based on embodiment one, the explosion-proof edge computing gateway realizes multi-path uplink transmission and redundant backhaul of scheduling and shaping message data, and specifically includes the following contents:

[0037] The time synchronization and timestamp module establishes a unified time reference to ensure the time consistency and synchronization accuracy of the processed multi-source monitoring data within the edge computing gateway.

[0038] The link sensing module takes the unified time reference as a synchronization reference, monitors the real-time communication state and extracts features of the wireless backhaul link between the explosion-proof edge computing gateway and the backhaul-side edge access server, generates link feature data, and performs time synchronization and time sequence normalization processing on the link feature data to form a link state sequence, which provides input for the delay prediction module. The link feature data includes RSRP, RSRQ, SINR, uplink and downlink PRB occupation rate, round-trip time (RTT), jitter, packet loss rate, RLC retransmission count, base station load indication, TCP congestion window and queue length.

[0039] Delay prediction module: used for forward-looking prediction of future performance of the wireless backhaul link; taking the link state sequence as input, using the diffusion bridge-TCN-GRU model to perform rolling prediction of end-to-end delay within 1-30 s in the future, and generating delay prediction results; the diffusion bridge-TCN-GRU model is constructed in the following way: a TCN-GRU network is established, the TCN-GRU network is used to extract short-term and long-term features of the link state sequence and output delay point prediction results, an adaptive diffusion bridge model based on fractional Brownian motion is constructed with the delay point prediction results as reference, and the delay evolution is modeled and sampled at the distribution level, after joint training by a joint loss function, the diffusion bridge-TCN-GRU model is constructed;

[0040] Scheduling decision module: based on a dual-objective Actor-Critic model, taking delay and energy consumption as joint optimization objectives, inputting the link state sequence and delay prediction results, and outputting an optimal transmission strategy, the optimal transmission strategy includes primary and backup link selection, shunting ratio, fragmentation granularity, sending window, ARQ threshold, FEC redundancy rate and retransmission strategy, realizing intelligent scheduling of multi-link transmission;

[0041] Queue and shaping module: four-level priority queues of alarm type, control type, periodic telemetry type and batch file type are established, the optimal transmission strategy output by the scheduling decision module is taken as parameter input, for various types of service message data encapsulated from processed multi-source monitoring data, an EDF (Earliest-Deadline-First) and LSTF (Least-Slack-Time-First) hybrid scheduling algorithm is used to determine the order of dequeuing, and a token bucket mechanism is used to perform rate limiting and burst traffic shaping, so as to balance end-to-end delay, bandwidth utilization and transmission reliability while meeting the real-time requirements of different services, and generate scheduling and shaping message data meeting the link scheduling requirements;

[0042] Multi-link backhaul module: according to the optimal transmission strategy, the activation state, bandwidth allocation and shunting ratio of the cellular communication link, the low-power wide-area link and the satellite communication link are configured, the shunting and link aggregation of the MPTCP / QUIC protocol are performed, the multi-path uplink transmission and redundant backhaul of the encapsulated message are realized, the reliability and optimal delay of data transmission are ensured, and the scheduling and shaping message data (encapsulated alarm message, control message, periodic telemetry message and batch file message) are sent uplink to the backhaul side edge access server.

[0043] Embodiment three, according to Figure 2 , the embodiment is based on embodiment two, in the embodiment, the link state sequence is taken as input, the diffusion bridge-TCN-GRU model is used to perform rolling prediction of end-to-end delay within 1-30 s in the future, and the process of generating delay prediction results, specifically including the following steps:

[0044] Step S1: Input the link state sequence into the TCN-GRU network, and extract short-term and long-term temporal features by passing through the temporal convolutional layer and the gated recurrent layer in sequence, and output the predicted end-to-end delay point value for the next 1–30 s.

[0045] Step S2: Introduce the Hurst exponent to quantify and model the long-memory characteristics of the link state sequence, forming a fractional Brownian motion model. Here, a Hurst exponent of 0.5 represents memoryless standard Brownian motion; a Hurst exponent > 0.5 represents a positively correlated super-diffusion process; and a Hurst exponent < 0.5 represents a negatively correlated sub-diffusion process. By estimating the Hurst exponent, the long-range dependence of the link delay can be determined. A Markov approximation is applied to the fractional Brownian motion model by constructing a linear combination of Ornstein–Uhlenbeck (OU) processes with a total of K layers. This modeling process transforms the non-Markovian delay sequence into an analytical stochastic dynamic system, forming an augmented state space containing main variables and covariates, thus establishing a Markovized fractional Brownian motion model.

[0046] The Markov approximation formula for the fractional Brownian motion model is as follows:

[0047] ;

[0048] in, Represents the Hurst exponent. Indicates the current time. Represents the fractional Brownian motion (fBM) of the Markov approximation in time. The value; Represents the total number of layers, the total number of Ornstein–Uhlenbeck (OU) subprocesses used to approximate fractional Brownian motion (fBM); Indicates 1 to index variables; Indicates the first The weighting coefficients of each component; This represents the integration time variable, from the product of zero to the current time. ; Indicates the first The decay rate of each component; Indicates exponential core / memory core, Represents Ito's random integral;

[0049] Step S3: Introduce a partially pinned fractional diffusion bridge mechanism to dynamically constrain and randomly drift the Markovized fractional Brownian motion model, achieving controllable random propagation of time-delayed evolution and forming a diffusion bridge model; specifically: using the main variable as the reference process, while maintaining the randomness of the covariates, only terminal constraints are applied to the main variable. By constructing a partially pinned fractional diffusion bridge equation, analytical drift terms and diffusion terms are introduced into the augmented state space, enabling the Markovized fractional Brownian motion model to maintain long memory characteristics while possessing goal orientation and numerical solvability, thus forming a diffusion bridge model;

[0050] Partially pinned fractional diffusion bridge equation:

[0051] ;

[0052] in, This represents the state vector in the augmented state space, which includes the main variables and covariates; This represents the system drift matrix (linear deterministic part). Represents the noise injection vector / diffusion load matrix. Indicates the transpose symbol. Represents standard Brownian motion. Indicates a resolvable drift term;

[0053] ;

[0054] in, Indicates the value of the current augmented state (putting...) (specific values ​​after substitution) The weights of the main variables reflect their direct driving force under terminal constraints. In the diffusion bridge model, the first... Linear weights of each OU component, Represents the time-sensitive kernel, measuring the current moment. Between to terminal 1, the first The response strength of each OU component to the terminal constraint; This represents the linear weight of the first OU component in the diffusion bridge model. Represents the time-sensitive kernel, measuring the current moment. Between terminal 1 and terminal 2, the response strength of the first OU component to the terminal constraint; This indicates the terminal anchor point (the target location for "pinning") of the main variable. Represents the conditional mean function; Represents the conditional variance function; This represents the variance-weighted terminal residual term;

[0055] Step S4: Introduce a temporal neural network as a learnable drift approximation function to perform neural modeling and dynamic correction of the analytic drift term, forming an adaptive diffusion bridge model. Specifically, the conditional expectation term of the analytic drift term is expressed in a trainable form, and the mapping from the link data space to the diffusion augmentation space is decoupled through a temporal neural network, enabling the diffusion bridge model to perform distribution-level compensation and dynamic correction for long-range dependent features. This structure achieves nonlinear fitting of dynamic drift and cross-scale feature fusion without increasing the overall computational complexity, thereby significantly enhancing the robustness and generalization ability of the model in different link environments, forming an adaptive diffusion bridge model.

[0056] Step S5: During the training phase, the point predictions output by the TCN-GRU network and the distribution characteristics output by the diffusion bridge model are used as the joint optimization objective to construct a joint loss function consisting of mean constraint terms and distribution constraint terms. By minimizing the joint loss function, the parameters of the TCN-GRU network and the parameters of the adaptive diffusion bridge model are updated synchronously to achieve collaborative learning of short-term trend prediction and long-term distribution modeling.

[0057] Joint loss function formula:

[0058] ;

[0059] in, Denotes the joint loss function. This represents the conventional supervision loss (mean constraint term) of the TCN-GRU network. This represents the distributed level loss (distributed constraint term) of the adaptive diffusion bridge model. This represents the loss balance coefficient, used to adjust the weight ratio between the mean constraint term and the distribution constraint term;

[0060] ;

[0061] in, This indicates that, under the reference process, the state vector in the augmented state space at the current moment is... The conditional mean result obtained after substituting into the conditional mean function; This indicates that, under the reference process, the state vector in the augmented state space at the current moment is... The conditional standard deviation result obtained after substituting into the conditional standard deviation function. Represents the standardized squared residuals of the terminal main variables; This represents the set of trainable parameters for the drift approximation function in the adaptive diffusion bridge model. It is used to adjust the dynamic response behavior of the diffusion bridge in the time domain and uncertainty space, so that the model can adaptively approximate the real time delay distribution evolution process. denotes a learnable drift approximation function parameterized by neural networks, which takes the time variable with conditional standard deviation for input, through a set of trainable parameters Parameterized modeling is performed for nonlinear approximation and dynamic correction of the analyzable drift term; denotes the integral of the time variable over the normalized time interval [0, 1]; denotes the mathematical expectation;

[0062] Step S6: After training, the time delay point prediction value output by the TCN-GRU network is input into the adaptive diffusion bridge model as a reference trajectory to drive it to perform multi-path random sampling, generating a time delay distribution sequence for the future 1-30 s; the quantile curve and confidence interval width are calculated according to the time delay distribution sequence, and are fused with the center estimation result of the time delay point prediction value, and finally the comprehensive time delay prediction result containing the expected value, confidence interval and uncertainty measure is output, realizing the distribution level prediction and credibility evaluation of the link time delay.

[0063] Embodiment four, based on embodiment two, in this embodiment, the link state sequence is taken as input, and the diffusion bridge-TCN-GRU model is used to perform rolling prediction of the end-to-end time delay for the future 1-30 s, and the process of generating the time delay prediction result includes the following steps:

[0064] Step S1: input the link state sequence into the TCN-GRU network, and sequentially extract short-term and long-term time sequence features through the time sequence convolution layer and the gated recurrent layer, and output the end-to-end time delay point prediction value for the future 1-30 s;

[0065] Step S2: in the model training phase, the mean square error (MSE) between the predicted time delay and the real time delay is taken as the supervision loss function, and the network parameters are optimized through back propagation, realizing the fitting and trend learning of the link time delay time sequence features by the model;

[0066] Step S3: in the inference phase, input the real-time link state sequence into the trained TCN-GRU network, and perform rolling prediction to obtain the time delay prediction sequence for the future multiple time steps;

[0067] Step S4: smooth the time delay prediction sequence and estimate the confidence interval, and output the time delay prediction result containing the expected value and the predicted fluctuation range, which is used to guide the link scheduling and transmission optimization.

[0068] The above describes the present application and its embodiments, which are not limited, and the drawings only show one of the embodiments of the present application, and the actual structure is not limited thereto; in general, if a person skilled in the art is inspired thereby, without departing from the purpose of the present application, without creative design, similar structure and embodiments to the technical solution, which should belong to the protection scope of the present application.

Claims

1. A wellhead monitoring data transmission delay optimization system based on wireless communication, characterized in that: The system includes: The sensing and acquisition unit acquires multi-source monitoring signals; The field RTU is electrically connected to the sensor acquisition unit and is used to perform A / D conversion and preprocessing of multi-source monitoring signals to obtain processed multi-source monitoring data. An explosion-proof edge computing gateway includes a time synchronization and timestamp module, a link awareness module, a delay prediction module, a scheduling decision module, a queueing and shaping module, and a multi-link backhaul module. It encapsulates processed multi-source monitoring data into communication messages, generating scheduling shaped message data. The explosion-proof edge computing gateway wirelessly connects to the backhaul-side edge access server via a wireless backhaul link, performing multi-path uplink transmission and redundant backhaul of the scheduling shaped message data. The backhaul edge access server receives the scheduled shaped message data and performs timestamp alignment, fusion and reassembly, and cache shaping operations on it to complete the consistency merging of multi-link data; and implements error recovery, congestion control, and traffic management mechanisms to ensure the integrity and continuity of data transmission. Link awareness module: performs real-time communication status monitoring and feature extraction on the wireless backhaul link, forming a link status sequence; Delay prediction module: Taking the link state sequence as input, it uses the diffusion bridge-TCN-GRU model to perform rolling prediction of end-to-end delay within the next 1–30 seconds and generates delay prediction results; Scheduling decision module: Based on the dual-objective Actor-Critic model, with latency and energy consumption as joint optimization objectives, it takes the link state sequence and latency prediction results as inputs and outputs the optimal transmission strategy. Queuing and Shaping Module: A four-level priority queue is established, with the optimal transmission strategy as the parameter input. For the business message data formed by encapsulating processed multi-source monitoring data, the EDF and LSTF hybrid scheduling algorithm is used to determine the dequeue order. The token bucket mechanism is used to perform rate limiting and burst traffic shaping to obtain the scheduled and shaped message data. The diffusion bridge-TCN-GRU model is constructed as follows: a TCN-GRU network is established, the short-term and long-term features of the link state sequence are extracted using the TCN-GRU network and the delay point prediction results are output. An adaptive diffusion bridge model is constructed with the delay point prediction results as a reference. The delay evolution is modeled at the distribution level and randomly sampled. After joint training with the joint loss function, the diffusion bridge-TCN-GRU model is constructed. The process of generating time delay prediction results using the diffusion bridge-TCN-GRU model includes the following steps: Step S1: Input the link state sequence into the TCN-GRU network and output the predicted delay point value; Step S2: Introduce the Hurst exponent to quantify and model the long memory features of the link state sequence; establish a Markovized fractional Brownian motion model by constructing a linear combination of the Ornstein–Uhlenbeck process for Markov approximation. Step S3: Introduce a partially pinned fractional diffusion bridge mechanism to dynamically constrain and randomly drift the Markovized fractional Brownian motion model, forming a diffusion bridge model; Step S4: Introduce a time neural network as a learnable drift approximation function, perform neural modeling and dynamic correction, and form an adaptive diffusion bridge model; Step S5: During the training phase, construct the joint loss function; by minimizing the joint loss function, synchronously update the TCN-GRU network parameters and the adaptive diffusion bridge model parameters; Step S6: After training is completed, the predicted time delay points output by the TCN-GRU network are used as reference trajectories and input into the adaptive diffusion bridge model to output the time delay prediction results.

2. The wellhead monitoring data transmission delay optimization system based on wireless communication according to claim 1, characterized in that: Wireless backhaul links include cellular communication links, low-power wide-area links, and satellite communication links.

3. The wellhead monitoring data transmission delay optimization system based on wireless communication according to claim 1, characterized in that, include: Time synchronization and timestamp module: Establishes a unified time base; The link awareness module specifically performs real-time communication status monitoring and feature extraction on the wireless backhaul link using a unified time base as a synchronization reference, generates link feature data, and performs time synchronization and time sequence normalization processing on the link feature data to form a link status sequence. The multi-link backhaul module specifically configures the activation status, bandwidth allocation, and traffic splitting ratio of cellular communication links, low-power wide-area links, and satellite communication links according to the optimal transmission strategy, performs traffic splitting and link aggregation based on MPTCP and QUIC protocols, realizes multi-path uplink transmission and redundant backhaul of encapsulated messages, and sends the scheduling shaped message data uplink to the backhaul side edge access server.

4. The wellhead monitoring data transmission delay optimization system based on wireless communication according to claim 3, characterized in that: Step S1 specifically involves: inputting the link state sequence into the TCN-GRU network, sequentially passing it through a temporal convolutional layer and a gated recurrent layer to extract short-term and long-term temporal features, and outputting the predicted delay point value; Step S2 specifically involves: introducing the Hurst exponent to quantify and model the long memory features of the link state sequence, forming a fractional Brownian motion model; and using a linear combination of Ornstein–Uhlenbeck processes with a total number of K layers to perform Markov approximation on the fractional Brownian motion model, forming an augmented state space containing main variables and covariates, and establishing a Markovized fractional Brownian motion model. Step S3 specifically involves: introducing a partially pinned fractional diffusion bridge mechanism to dynamically constrain and randomly drift-correct the Markovized fractional Brownian motion model, thus forming a diffusion bridge model; specifically: using the main variable as the reference process, while maintaining the randomness of the covariates, only terminal constraints are applied to the main variable, and by constructing a partially pinned fractional diffusion bridge equation, analytical drift terms and diffusion terms are introduced into the augmented state space to form a diffusion bridge model; Step S4 specifically involves: introducing a time neural network as a learnable drift approximation function, performing neural modeling and dynamic correction on the analytic drift term, and forming an adaptive diffusion bridge model; Step S5 specifically involves: during the training phase, using the point predictions output by the TCN-GRU network and the distribution characteristics output by the diffusion bridge model as the joint optimization objective, constructing a joint loss function composed of mean constraint terms and distribution constraint terms; by minimizing the joint loss function, synchronously updating the parameters of the TCN-GRU network and the parameters of the adaptive diffusion bridge model, thereby achieving collaborative learning of short-term trend prediction and long-term distribution modeling. Step S6 is as follows: After training is completed, the predicted time delay points output by the TCN-GRU network are used as reference trajectories and input into the adaptive diffusion bridge model to drive it to perform multi-path random sampling and generate a time delay distribution sequence; the quantile curve and confidence interval width are calculated based on the time delay distribution sequence and fused with the center estimation results of the predicted time delay points to finally output the time delay prediction result.

Images