A geothermal well monitoring and maintenance management system and method based on the Internet of Things

Abstract

The present invention relates to the field of geothermal resource development technology, specifically a geothermal well monitoring and maintenance management system and method based on the Internet of Things; the method steps are: through a denoising method combining adaptive fuzzy clustering and deep convolutional neural network, the real signal of sensor data is restored; by generating a check mark, the integrity of the collected geothermal well monitoring data is checked; combined with LSTM time series modeling and deep reinforcement learning for fault prediction and optimization, support vector machine is used for fault type identification, and maintenance strategies are formulated according to the prediction results, the appropriate maintenance type and timing are selected, and resource scheduling optimization is performed to generate a detailed maintenance plan; data and maintenance information are visualized and pushed to on-site staff in real time, and the staff performs operations according to the maintenance plan, and automatically records and updates the health status of the equipment. The present invention improves the monitoring and maintenance efficiency of geothermal wells through intelligent data processing and decision support.

Description

Technical Field

[0001] The present invention relates to the technical field of geothermal resource development, and in particular to a geothermal well monitoring and maintenance management system and method based on the Internet of Things. Background Art

[0002] With the continuous growth of energy demand, geothermal energy, as a clean and sustainable energy resource, has gradually attracted global attention. As the core facility for the development and utilization of geothermal resources, the stability and safety of geothermal wells directly affect the collection efficiency and sustainable utilization of geothermal energy. However, in the long-term operation of geothermal wells, due to the influence of geological environment, equipment aging, untimely maintenance and other factors, they are prone to temperature, pressure anomalies, vibration and other faults, resulting in system performance degradation, and even equipment damage or safety accidents. Therefore, how to achieve real-time monitoring and fault warning of geothermal wells to ensure their efficient and safe operation has become a key issue that needs to be urgently solved in the field of geothermal energy development. In recent years, with the development of Internet of Things technology, geothermal well monitoring has gradually developed towards intelligence and automation.

[0003] A Chinese invention patent application published as CN116843276B discloses a smart garden maintenance management system and method based on the Internet of Things. The system includes a wetland environment monitoring and analysis module, a wetland ecological extraction and analysis module, a wetland blockage monitoring module, a wetland blockage analysis module, a wetland terminal control module and a wetland database. By monitoring the environmental information, extracting historical information and monitoring the infiltration information of the target wetland garden area, the winter overwintering problem of the target wetland garden area is solved, and the biodiversity and regional ecological balance of the wetland garden in winter are maintained. At the same time, the winter biological community structure of the wetland garden is avoided, and the hydraulic performance of the wetland is guaranteed to a certain extent to avoid the problem of wetland degradation, thereby improving the treatment effect and service life of the wetland. At the same time, the climate regulation function of the wetland is improved, the water resource supply security is guaranteed, and the flood storage and waterlogging prevention capacity along the river is enhanced; Traditional methods often rely on manual inspections and regular checks, which make it difficult to achieve real-time, comprehensive monitoring and accurate fault prediction. Therefore, how to improve the stability and safety of geothermal well operation and reduce the occurrence rate of failures has become an urgent issue that needs to be addressed in the development and utilization of geothermal energy. Summary of the invention

[0004] The purpose of the present invention is to address the problems existing in the background technology and to propose a geothermal well monitoring and maintenance management system and method based on the Internet of Things.

[0005] The technical solution of the present invention is a geothermal well monitoring and maintenance management method based on the Internet of Things, which includes the following specific implementation steps: S1. Real-time monitoring of the working status of the geothermal well and continuous collection of data on the working status of the geothermal well; S2, aggregate sensor data, i.e., geothermal well working status data; S3, analyze the sensor data through adaptive fuzzy clustering, identify the noise data, and use convolutional neural network to repair the noise data, retain the real signal, and dynamically adjust the convolution kernel parameters to adapt to different noise patterns, optimize the data processing process, and then generate a verification mark; S4. Verify the integrity and compliance of sensor data and input it into the long short-term memory network (LSTM) for time series data processing. Combine deep reinforcement learning with traditional Q learning to build a fault prediction model, update the Q value through a deep neural network, evaluate actions under different states, and optimize the decision-making strategy based on real-time feedback, outputting prediction results and optimization strategies. S5. Use support vector machine to identify fault types, distinguish different fault categories by constructing hyperplanes, and select maintenance strategies based on prediction results, including maintenance decisions, maintenance types, timing, and resource scheduling, and use simulated annealing algorithm to optimize resource scheduling, generate detailed maintenance plans, and output classification results and maintenance plans. S6, visualize the sensor data, prediction results, classification results and maintenance plan, and the staff arranges specific maintenance work according to the maintenance plan and records the maintenance records; S7. Save sensor data, prediction results, classification results, maintenance plans, and repair records.

[0006] Preferably, the process of repairing noisy data is as follows: S21. Perform cluster analysis on sensor data based on adaptive fuzzy clustering and divide the data into several clusters. The process is as follows: S2101. The basic formula of fuzzy clustering is: ; In the formula, μ ij Represents data point x i For cluster c j The membership degree of x i represents the i-th data point; d(x i ,c j ) represents the data point x i To cluster center c j The Euclidean distance of the clusters; m represents the fuzzy factor, which controls the membership distribution of data points in multiple clusters; c represents the total number of clusters; c k represents the cluster center of the kth cluster; S2102, clustering the sensor data by a fuzzy clustering algorithm to identify the data points belonging to noise, and after clustering, analyzing the membership of each cluster. If the mean membership of a cluster is lower than a set threshold, the sample corresponding to the cluster is considered to be noise; S22. Use the convolutional neural network (CNN) to further repair the noise data. The convolutional layer learns the noise characteristics in the input signal, removes the noise, and retains the real signal. Specifically: S2201. Through the convolution operation, the filter analyzes and identifies the noise pattern in the signal, filters out the noise part, and repairs the noise data. The convolution operation formula is: ; In the formula, y[i] represents the convolution output; x[i+τ] represents the i+τth data point of the input signal; w[τ] represents the τth weight parameter of the convolution kernel; K represents the size of the convolution kernel; S2202. The training process of CNN uses the back propagation algorithm to adjust the parameters θ of the convolution kernel by minimizing the loss function. The formula of the loss function is: ; Where L(θ) represents the loss function, that is, the error between the predicted value and the true value; y i represents the true value of the i-th data point, that is, the signal after noise repair; represents the predicted value of the i-th data point, which is calculated by the input data and parameter θ through the network forward propagation; N represents the size of the data set, that is, the total number of input data; S2203. After the training is completed, the convolutional neural network repairs the input noise signal and restores the true content of the signal.

[0007] Preferably, the adjustment process of dynamically adjusting the convolution kernel parameters to adapt to different noise modes is as follows: An adaptive adjustment mechanism is introduced to dynamically adjust the parameters of the convolution kernel to adapt to new noise patterns. This is achieved through a dynamic learning rate: ; In the formula, θ t represents the parameters of the model at time step t; θ t-1 represents the parameters of the model at the previous time step t-1; η represents the learning rate, that is, the step size of the model update; Represents the gradient of the loss function, that is, the direction and magnitude of the model parameter adjustment.

[0008] Preferably, the verification mark generation process is as follows: S41, sensor data i Convert to binary string hdata i ; Among them, data i Indicates T i Sensor data acquired in real time at all times; S42, select a random number r∈Z q , taking the random number r as the first-order verification parameter; Where q is the order of the cyclic group G; G is a predefined cyclic group of order q; based on this, the bilinear map e is defined as: G×G→G T ; G T represents a cyclic group; Z p is the ring of integers modulo q; S43, calculate the second-order calibration parameter r': ; Where g is the generator of the cyclic group G; α and β represent predefined encoding parameters; α∈Z q ; β∈Z q ; Decoding parameters α' and β' are defined accordingly: α'=g α ; β' = g β ; and make the bilinear mapping e, decoding parameters α' and β' public; S44, output verification mark Lc i =(r, r').

[0009] Preferably, the verification process for verifying the integrity and compliance of sensor data is as follows: S51, the received sensor data data i Convert to binary string hdata i ' ; S52, calculate the first-order check code Ce1: ; Wherein, g is a generator of a predefined cyclic group G; e is a predefined bilinear map; S53, obtaining a predefined second-order check code Ce2: Ce2=e(g, g); S54, if Ce1=Ce2, it means that the received sensor data data i Verification passed, that is, the received sensor data data i Complete and compliant.

[0010] Preferably, the generation process of the prediction results and optimization strategy is as follows: S61. Input the preprocessed data into the LSTM network. At each time step, the LSTM updates its own state according to the current input and the state at the previous moment. The LSTM learns long-term dependencies and processes the time series changes in the geothermal well data by controlling the forget gate, input gate, candidate memory unit, and output gate. S62: Build an intelligent fault prediction and optimization model, combine traditional Q-learning methods and deep reinforcement learning, and continuously optimize the prediction strategy based on real-time feedback, as follows: S6201, the intelligent fault prediction and optimization model uses a deep neural network to approximate the Q function. Specifically, the Q value is updated by the following formula: ; In the formula, s t Indicates the current state, corresponding to the current state of sensor data in the geothermal well system; a t represents the current action taken; θ Q Represents the parameters of the current neural network, i.e., weights, which are optimized through training; r t represents the reward at the current moment, reflecting the effect of the current action; γ represents the discount factor, which determines the importance of future rewards; Represents the parameters of the target network, which is used to calculate the target Q value; a' represents the next state s t+1 One of all the actions taken in The goal of updating the Q value is to optimize the current decision by maximizing future rewards. The specific update process is as follows: ; In the formula, y t Indicates the target Q value calculated according to the target network, which is used to compare with the Q value output by the current network; S6302: Update the parameters of the Q network and use the mean square error as the loss function to measure the difference between the current Q value and the target Q value: ; Where D represents the experience pool, which contains the historical state-action-reward quadruple, each of which represents the complete experience of the agent interacting with the environment at a certain moment: ; represents the expected operation on the samples sampled from the experience pool D; L(θ Q ) represents the loss function; The purpose of the loss function is to minimize the difference between the current Q value and the target Q value; S6303, Fault prediction and optimization model performs fault prediction, uses reward function to evaluate prediction accuracy, gives positive rewards when faults are successfully predicted, and imposes penalties when warnings are not given in time; S6304. Output prediction results and optimization strategies.

[0011] Preferably, the fault prediction process is as follows: S71. Define the state space s t : The state space consists of sensor data, forming a vector s t ; S72. Define action space a t : Define the following actions: a1: No fault, continue monitoring; a2: A fault occurs and the early warning program is started; a3: A failure occurs and the equipment is shut down for maintenance; S73. Define reward function: Positive reward: If the model successfully predicts a failure and takes action, a positive reward is given: r t =+10, that is, if the fault is warned in advance and effectively responded to; Negative reward: If the fault is not discovered in time, resulting in equipment damage, a negative reward will be given: t =-50, that is, if a failure occurs and there is no early warning; Penalty Reward: Give negative rewards if unnecessary actions are taken prematurely: r t =-20, i.e. if premature fault warning leads to unnecessary downtime; S74: Q value update and model optimization, as follows: The state adopts: the processed sensor data received by the data analysis unit; Action selection: Select an action based on the current state, and use the ε-greedy strategy to select the action with the largest Q value; Experience replay: The current state, action, reward, and next state are stored in the experience pool. During each training, batches of data are randomly extracted from the experience pool for training to avoid the impact of the correlation of time series data on training. Q value update: according to the target Q value y t The difference between the Q value and the current Q value updates the parameters of the neural network: ; Target network update: Set the update time interval to copy the weights of the current Q network to the target network to maintain the stability of the target Q value: .

[0012] Preferably, the optimization objective function of the simulated annealing algorithm is: ; In the formula, C total represents the total maintenance cost; represents the personnel cost required for the ith maintenance task; represents the cost of the equipment required for the i-th task; Represents the cost of materials required for the i-th task.

[0013] The technical solution of the present invention is a geothermal well monitoring and maintenance management system based on the Internet of Things, which is used to execute the above-mentioned geothermal well monitoring and maintenance management method based on the Internet of Things, including: A sensor module, which integrates several sensor units and is used to collect geothermal well monitoring data; The data acquisition and transmission module is used to receive data from the sensor, convert it into digital signals, and upload the data to the data processing module using wireless communication technology; The data processing module is used to pre-process the received sensor data and transmit the pre-processed sensor data to the cloud platform and the big data analysis module; User interface and application module, used to provide an operation interface and a visual interface for field staff; Cloud platform and big data analysis module, used to store and analyze sensor data, including: A cloud data platform for storing, processing and analyzing data from geothermal wells, supporting large-scale data storage and efficient query; Data analysis unit, which is used to analyze the working status of equipment through machine learning and data mining technology, support anomaly detection and trend analysis, and help formulate maintenance plans; Intelligent decision-making unit, used to provide real-time decision support, generate maintenance strategies and predict future trends based on data analysis results.

[0014] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects: The present invention designs a geothermal well monitoring and maintenance management system and method based on the Internet of Things. By combining the Internet of Things technology, deep learning, reinforcement learning and intelligent decision-making technology, it has significant technical effects in improving the operation efficiency of geothermal wells, reducing the failure rate, optimizing maintenance decisions, etc.: (1) Improving monitoring accuracy and real-time performance: By installing a variety of sensors at key locations of geothermal wells, the operating status of geothermal wells can be fully and real-time monitored, which provides high-precision basic data support for subsequent fault warning and equipment maintenance; (2) Intelligent noise removal and data recovery: The denoising method that combines adaptive fuzzy clustering (FCM) with deep convolutional neural network (CNN) can effectively remove noise from sensor data and restore the real signal. This multi-level denoising mechanism can cope with various environmental changes, improve data quality, and provide more reliable data support for fault prediction. (3) Fault prediction and optimized decision-making: By combining long short-term memory (LSTM) networks with deep reinforcement learning (DRL), the system can analyze the time series changes of geothermal well data and make fault predictions. The reinforcement learning model optimizes the decision-making process through real-time feedback, provides intelligent decision-making based on historical data, and improves the accuracy and response speed of prediction. This adaptive prediction mechanism can significantly improve the operational safety of geothermal wells and reduce the probability of sudden failures. (4) Strengthen fault type identification and maintenance decision optimization: Based on support vector machine (SVM) fault type classification, the system can accurately identify different types of faults and select the best response strategy according to the fault type. At the same time, combined with the maintenance decision optimization model of reinforcement learning, the maintenance resources are optimally scheduled through the simulated annealing algorithm, thereby reducing maintenance costs, improving maintenance efficiency, and ensuring the timeliness and accuracy of maintenance tasks; (5) System intelligence and automation: By combining sensor data analysis, intelligent decision-making and resource optimization, it can automatically generate detailed maintenance plans, push them to on-site staff and record the maintenance process, greatly reducing manual intervention and operational errors. At the same time, through the combination of user interfaces and mobile applications, maintenance personnel can grasp the equipment status and maintenance tasks in real time, improving work efficiency and the level of intelligent equipment management. (6) Enhance data integrity and reliability, and ensure data accuracy and consistency: The generation and verification of check marks improves data integrity and reliability, ensuring that geothermal well monitoring data is not tampered with or lost during transmission and storage. Through real-time verification, data errors can be quickly discovered, misjudgment of faults can be reduced, and system response speed and data real-time performance can be optimized. At the same time, check marks provide support for fault warning mechanisms, detect abnormal data in advance and handle potential risks, enhance system security, prevent data from being illegally tampered with, and ensure data security. The solution has good scalability, supports different data formats and monitoring requirements, simplifies large-scale management, and adapts to future technological changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a system architecture diagram of a geothermal well monitoring and maintenance management system based on the Internet of Things proposed by the present invention; Figure 2 This is a method flow chart of a geothermal well monitoring and maintenance management method based on the Internet of Things proposed by the present invention. DETAILED DESCRIPTION

[0016] Embodiment 1, as Figure 1 As shown, the present invention proposes a geothermal well monitoring and maintenance management system based on the Internet of Things, including: a sensor module, a data acquisition and transmission module, a data processing module, a cloud platform and a big data analysis module, and a user interface and application module.

[0017] The sensor module integrates several sensor units for collecting geothermal well monitoring data; Sensor units include but are not limited to: Temperature sensor: monitors the temperature at the wellhead and underground of geothermal wells to evaluate the utilization status of geothermal resources; Pressure sensor: monitors downhole pressure in real time, helps determine the pressure status at the wellhead, and identifies whether there are abnormal pressure changes; Flow meter: measures the wellhead flow rate and determines the flow of geothermal resources; Water quality sensor: detects numerical data, including but not limited to the composition, pH value, salinity, etc. of well water, and evaluates the quality of well water; Vibration sensor: detects the operating status of the equipment and identifies any possible mechanical problems; The data acquisition and transmission module is used to receive data from the sensor and convert it into digital signals, and upload the data to the data processing module using wireless communication technology (including but not limited to LoRa, NB-IoT or 5G); The data processing module pre-processes the received sensor data and transmits the pre-processed sensor data to the cloud platform and the big data analysis module; The cloud platform and big data analysis module are used to store and analyze sensor data, including: Cloud data platform: used to store, process and analyze data from geothermal wells, supporting large-scale data storage and efficient query; Data analysis unit: Analyzes the working status of equipment through machine learning and data mining technology, including but not limited to early warning of equipment failure or performance degradation, supports anomaly detection and trend analysis, and helps formulate maintenance plans; Intelligent decision-making unit: Based on data analysis results, it provides real-time decision support, generates maintenance strategies, and predicts future trends; The user interface and application module provide an operation interface and a visual interface for on-site staff, including but not limited to providing real-time data viewing, fault alarm push, and maintenance log recording functions.

[0018] Embodiment 2, as Figure 2As shown, the geothermal well monitoring and maintenance management method based on the Internet of Things proposed in the present invention is applied to the geothermal well monitoring and maintenance management system based on the Internet of Things proposed in Example 1, and its specific implementation steps are as follows: S1. Install several sensors at key locations of geothermal wells (including but not limited to wellheads and downhole pipelines), including but not limited to temperature sensors, pressure sensors, flow meters, water quality sensors and vibration sensors, to monitor the working status of geothermal wells in real time and continuously collect key data, including wellhead temperature, downhole pressure, geothermal flow, water quality components (including but not limited to pH value, salinity), and equipment vibration; The data collected by the sensor is then transmitted to the data collection and transmission module through wireless communication technology (including but not limited to LoRa, NB-IoT, and 5G).

[0019] S2, the data acquisition and transmission module converts the analog signals collected by the sensor into digital signals, and uploads the data to the data processing module using wireless communication technology (including but not limited to LoRa, NB-IoT or 5G).

[0020] S3, data processing module is based on the denoising method combining adaptive fuzzy clustering and deep learning. By combining adaptive fuzzy clustering and deep convolutional neural network, it adaptively identifies and filters noise and restores the true signal of sensor data. Fuzzy clustering identifies noise clusters by clustering data, while deep convolutional neural network denoises by learning noise features. The denoising process is as follows: S31. Perform cluster analysis on sensor data based on adaptive fuzzy clustering (FCM) and divide the data into several clusters. The process is as follows: S3101. The basic formula of fuzzy clustering is: ; In the formula, μ ij Represents data point x i For cluster c j The membership degree of x i represents the i-th data point, i.e., the original measurement value collected from the sensor; d(x i ,c j ) represents the data point x i To cluster center c j The Euclidean distance; m represents the fuzzy factor, which controls the membership distribution of data points in multiple clusters. In this embodiment, the value is 2; c represents the total number of clusters; c k represents the cluster center of the kth cluster; S3102, clustering the sensor data by a fuzzy clustering algorithm to identify the data points belonging to noise, and after clustering, analyzing the membership of each cluster. If the mean membership of a cluster is lower than a set threshold, the sample corresponding to the cluster is considered to be noise; S32. Use convolutional neural network (CNN) to further repair the noise data. The convolution layer learns the noise characteristics in the input signal, removes the noise, and retains the real signal. The specific implementation process is as follows: S3201, repair the noise data, the convolution operation formula is: ; In the formula, y[i] represents the convolution output; x[i+τ] represents the i+τth data point of the input signal; w[τ] represents the τth weight parameter of the convolution kernel; K represents the size of the convolution kernel; Based on this, through the convolution operation, the filter analysis identifies the noise pattern in the signal, and finally the noise is partially filtered out, retaining the useful information of the signal; S3202, CNN training process uses the back propagation algorithm to adjust the parameters of the convolution kernel by minimizing the loss function (mean square error). The formula of the loss function is: ; Where L(θ) represents the loss function, that is, the error between the predicted value and the true value; y i represents the true value of the i-th data point, that is, the signal after noise repair; represents the predicted value of the i-th data point, which is calculated by the input data and model parameters θ through network forward propagation; N represents the size of the data set, that is, the total number of input data; S3203. After the training is completed, the convolutional neural network model repairs the input noise signal and generates an output that is smoother and closer to the real signal. At this time, the network has learned the spatial and temporal characteristics of the noise and can remove these characteristics to restore the real content of the signal. S33. In order to cope with the ever-changing environment and noise types, an adaptive adjustment mechanism is introduced, which dynamically adjusts the parameters of the convolution kernel to adapt to the new noise pattern. This process is achieved through a dynamic learning rate: ; In the formula, θ t represents the parameters of the model at time step t; θ t-1 represents the parameters of the model at the previous time step t-1; η represents the learning rate, that is, the step size of the model update; Represents the gradient of the loss function, that is, the direction and magnitude of the model parameter adjustment; S34, output the sensor data after denoising i(data i Indicates T i The sensor data acquired in real time at all times) and the verification mark is generated for the denoised sensor data. The generation process is as follows: S3401, sensor data i Convert to binary string hdata i ; S3402, select a random number r∈Z q , taking the random number r as the first-order verification parameter; Where q is the order of the cyclic group G; G is the defined q-order cyclic group; based on this, the bilinear map e is defined: G×G→G T ; G T represents a cyclic group; Z p is the ring of integers modulo q; S3403, calculate the second-order calibration parameter r': ; Where g is the generator of the cyclic group G; α and β represent predefined encoding parameters; α∈Z q ; β∈Z q ; Decoding parameters α' and β' are defined accordingly: α'=g α ; β' = g β ; and make the bilinear mapping e, decoding parameters α' and β' public; S3404, output verification mark Lc i =(r,r'); S35, {verification mark Lc i =(r, r'), sensor data i}Transmitted to the data analysis unit and cloud data platform.

[0021] S4, the data analysis unit combines LSTM time series modeling with reinforcement learning optimization, and uses SVM classification to identify fault types to improve the accuracy of fault prediction. The analysis process is as follows: S41, receiving {verification mark Lc i =(r, r'), sensor data i}, and verify accordingly: S4101, receive the sensor data data i Convert to binary string hdata i ' ; S4102, calculate the first-order check code Ce1: ; Wherein, g is a generator of a predefined cyclic group G; e is a predefined bilinear map; S4103, obtain a predefined second-order check code Ce2: Ce2=e(g, g); S4104, if Ce1=Ce2, then the received sensor data data i Verification passed, that is, the received sensor data data i Complete and compliant; S42. Input the preprocessed data into the LSTM network (Long Short-Term Memory Network). The core of LSTM is four important gating mechanisms: Forget Gate, Input Gate, Candidate Memory, and Output Gate. Its goal is to predict future trends and states based on historical data. At each time step, LSTM updates its state based on the current input and the state at the previous moment, as follows: S4201, Forget Gate: ; In the formula, f t represents the output of the forget gate, which determines how much information from the previous moment is retained in the current state; σ represents the Sigmoid activation function, which limits the output to between 0 and 1; W f represents the weight matrix of the forget gate, which controls the impact of input information on memory; h t-1 represents the hidden state of the previous moment (i.e. the output of the previous moment); x t represents the input (sensor data) at the current moment; b f Represents the bias term, which adjusts the output of the neural network; S4202, Input Gate: ; In the formula, i t Represents the output of the input gate, which controls the impact of the current input information on the memory unit; W i represents the weight matrix of the input gate; b i represents the bias term of the input gate; S4203, Candidate Memory: ; In the formula, represents the candidate memory unit, which is used to generate the current state update; tanh represents the hyperbolic tangent activation function, which ensures that the output of the candidate memory unit remains between -1 and 1; W C The weight matrix representing the candidate memory unit; S4204, Cell State Update: ; In the formula, C t Indicates the unit state at the current moment, which determines the memory of the model; C t-1 Indicates the cell state at the previous moment; * indicates element-by-element multiplication; S4205, Output Gate: ; In the formula, o t Represents the output of the output gate, which determines the hidden state at the current moment (i.e., the output of the network); W o represents the weight matrix of the output gate; b o represents the bias term of the output gate; S4206, Final Output (Hidden State): ; In the formula, h t Represents the hidden state at the current moment, indicating the current prediction result; tanh(C t ) represents the generation of output through the value of the unit state; Based on this: LSTM can effectively remember long-term temporal dependencies and handle temporal changes in geothermal well data by controlling these gates; S43. Combining the advantages of traditional Q-learning methods and deep reinforcement learning (DRL), an intelligent fault prediction and optimization model is constructed. The model introduces deep neural networks (DNN) on the basis of traditional Q-learning, which can process high-dimensional and complex data and continuously optimize the prediction strategy based on real-time feedback. The details are as follows: S4301. The intelligent fault prediction and optimization model uses a deep neural network to approximate the Q function. Specifically, the Q value is updated using the following formula: ; In the formula, s t Indicates the current state, corresponding to the current state of sensor data in the geothermal well system, including but not limited to temperature, pressure, and vibration; a t Indicates the current action, including but not limited to starting an alarm, closing a valve, and starting an emergency stop; θ Q Represents the parameters (weights) of the current neural network, which are optimized through training; r t represents the reward at the current moment, reflecting the effect of the current action, including but not limited to the success or failure of the warning; γ represents the discount factor, which determines the importance of future rewards. It usually takes a value between 0 and 1. The closer it is to 1, the more important the future rewards are; Represents the parameters of the target network, which is used to calculate the target Q value. The target network is updated regularly to maintain stability; a' represents the next state s t+1 One of all the actions taken in The goal of updating the Q value is to optimize the current decision by maximizing future rewards. The specific update process is as follows: ; In the formula, y t Indicates the target Q value calculated according to the target network, which is used to compare with the Q value output by the current network; S4302. In order to update the parameters of the Q network, the mean square error (MSE) is used as the loss function to measure the difference between the current Q value and the target Q value: ; Where D represents the experience pool, which contains the historical state-action-reward quadruple, each of which represents the complete experience of the agent interacting with the environment at a certain moment: ; represents the expected operation on the samples sampled from the experience pool D; L(θ Q ) represents the loss function; The purpose of the loss function is to minimize the difference between the current Q value and the target Q value; S4303, the fault prediction process is as follows: A1. State space s t :The state space is composed of data from several sensors, including but not limited to temperature, pressure, and vibration. The output of the sensor forms a vector s t ; Action Space a t : Define the following actions: (1) a1: No fault, continue monitoring; (2) a2: A failure occurs and the early warning program is activated; (3) a3: A failure occurs and the equipment is shut down for maintenance; A2. The reward function determines the optimization direction of the model. The reward function reflects the effect of fault prediction and encourages the model to make accurate predictions: Positive reward: If the model successfully predicts a fault and takes corresponding measures (including but not limited to starting an early warning), a positive reward is given: r t =+10, if the fault is warned in advance and effectively responded to; Negative reward: If the fault is not discovered in time, resulting in equipment damage, a negative reward will be given: t =-50, if a failure occurs without early warning; Penalty reward: If the system takes unnecessary actions prematurely, including but not limited to premature shutdown, a negative reward is given: r t =-20, if premature fault warning leads to unnecessary downtime; A3: Q value update and model optimization, as follows: (1) State adoption: processed sensor data received by the data analysis unit; (2) Action selection: Select an action based on the current state, and use the ε-greedy strategy to select the action with the largest Q value; (3) Experience replay: The current state, action, reward, and next state are stored in the experience pool. During each training, batches of data are randomly extracted from the experience pool for training to avoid the impact of the correlation of time series data on training. (4) Q value update: according to the target Q value y t The difference between the Q value and the current Q value updates the parameters of the neural network: ; (5) Target network update: At regular intervals, the weights of the current Q network are copied to the target network to maintain the stability of the target Q value: ; S4304: Output the prediction results and corresponding strategies, and transmit the prediction results and corresponding strategies to the intelligent decision-making unit.

[0022] S5. The intelligent decision-making unit classifies and makes decisions on faults based on the prediction results of sensor data, as follows: S51. Use support vector machine (SVM) to identify fault types. The basic idea of ​​SVM is to distinguish different types of data by constructing a hyperplane. Its classification decision function is: ; In the formula, y represents the predicted fault type, which belongs to discrete categories, including but not limited to "overheating fault" and "pressure abnormality"; w represents the weight vector, which controls the inclination of the hyperplane; x represents the input feature vector, which contains sensor data (including but not limited to temperature and pressure); b represents the bias term, which adjusts the classification boundary; sign identifies the sign function, and the output is +1 or -1, which represents the classification results of different categories; S52. Based on the strategy corresponding to the prediction results, a reinforcement learning model is used to select maintenance decisions, including: Whether to perform maintenance (whether to trigger a maintenance decision); What type of maintenance to choose (including but not limited to preventive maintenance, corrective maintenance, predictive maintenance); Determine when maintenance should occur (urgent, scheduled, or deferrable); Resource scheduling (configuration of equipment, personnel and materials); S53. Use simulated annealing algorithm to achieve optimal resource scheduling. The optimization objective function can be expressed as: ; In the formula, C total represents the total maintenance cost; represents the personnel cost required for the ith maintenance task; represents the cost of the equipment required for the i-th task; represents the cost of materials required for the i-th task; S54. Automatically generate a detailed maintenance plan based on the decision model and resource optimization. The plan content includes but is not limited to: Maintenance task list: arrange maintenance tasks according to priority; Personnel and equipment arrangement: arrange corresponding technicians, maintenance equipment, and spare parts; Time scheduling: Arrange specific maintenance time according to the urgency of the task to ensure that the maintenance will not affect the normal operation of the equipment; S55, transmitting the classification results and maintenance plan to the user interface and application module and cloud data platform.

[0023] S6. The user interface and application module visualizes the sensor data, prediction results, classification results and maintenance plan, and automatically pushes them to the mobile application of the on-site staff. The staff arranges specific maintenance work according to the maintenance plan. After the maintenance work is completed, the staff needs to enter the maintenance record in the mobile application, including but not limited to the replaced parts and the detailed process of the maintenance; Based on this, the user interface and application module automatically record and update the health status of the equipment to ensure the accuracy of subsequent monitoring.

[0024] S7. The cloud data platform saves sensor data, prediction results, classification results, maintenance plans, and repair records.

[0025] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A geothermal well monitoring and maintenance management method based on the Internet of Things, characterized in that: The specific implementation steps include the following: S1. Real-time monitoring of the working status of the geothermal well and continuous collection of data on the working status of the geothermal well; S2, aggregate sensor data, i.e., geothermal well working status data; S3, analyze the sensor data through adaptive fuzzy clustering, identify the noise data, and use convolutional neural network to repair the noise data, retain the real signal, and dynamically adjust the convolution kernel parameters to adapt to different noise patterns, optimize the data processing process, and then generate a verification mark; S4. Verify the integrity and compliance of sensor data and input it into the long short-term memory network (LSTM) for time series data processing. Combine deep reinforcement learning with traditional Q learning to build a fault prediction model, update the Q value through a deep neural network, evaluate actions under different states, and optimize the decision-making strategy based on real-time feedback, outputting prediction results and optimization strategies. S5. Use support vector machine to identify fault types, distinguish different fault categories by constructing hyperplanes, and select maintenance strategies based on prediction results, including maintenance decisions, maintenance types, timing, and resource scheduling, and use simulated annealing algorithm to optimize resource scheduling, generate detailed maintenance plans, and output classification results and maintenance plans. S6, visualize the sensor data, prediction results, classification results and maintenance plan, and the staff arranges specific maintenance work according to the maintenance plan and records the maintenance records; S7. Save sensor data, prediction results, classification results, maintenance plans, and repair records.

2. The method for monitoring and maintaining geothermal wells based on the Internet of Things according to claim 1, characterized in that: The process of repairing noisy data is as follows: S21. Perform cluster analysis on sensor data based on adaptive fuzzy clustering and divide the data into several clusters. The process is as follows: S2101. The basic formula of fuzzy clustering is: ； In the formula, μ ij Represents data point x i For cluster c j The membership degree of x i represents the i-th data point; d(x i ,c j ) represents the data point x i To cluster center c j The Euclidean distance of the clusters; m represents the fuzzy factor, which controls the membership distribution of data points in multiple clusters; c represents the total number of clusters; c k represents the cluster center of the kth cluster; S2102, clustering the sensor data by a fuzzy clustering algorithm to identify the data points belonging to noise, and after clustering, analyzing the membership of each cluster. If the mean membership of a cluster is lower than a set threshold, the sample corresponding to the cluster is considered to be noise; S22. Use the convolutional neural network (CNN) to further repair the noise data. The convolutional layer learns the noise characteristics in the input signal, removes the noise, and retains the real signal. Specifically: S2201. Through the convolution operation, the filter analyzes and identifies the noise pattern in the signal, filters out the noise part, and repairs the noise data. The convolution operation formula is: ； In the formula, y[i] represents the convolution output; x[i+τ] represents the i+τth data point of the input signal; w[τ] represents the τth weight parameter of the convolution kernel; K represents the size of the convolution kernel; S2202. The training process of CNN uses the back propagation algorithm to adjust the parameters of the convolution kernel by minimizing the loss function. The formula of the loss function is: ； Where L(θ) represents the loss function, that is, the error between the predicted value and the true value; y i represents the true value of the i-th data point, that is, the signal after noise repair; represents the predicted value of the i-th data point, which is calculated by the input data and parameter θ through the network forward propagation; N represents the size of the data set, that is, the total number of input data; S2203. After the training is completed, the convolutional neural network repairs the input noise signal and restores the true content of the signal.

3. The method for monitoring and maintaining geothermal wells based on the Internet of Things according to claim 2, characterized in that: The adjustment process of dynamically adjusting the convolution kernel parameters to adapt to different noise modes is as follows: An adaptive adjustment mechanism is introduced to dynamically adjust the parameters of the convolution kernel to adapt to new noise patterns. This is achieved through a dynamic learning rate: ； In the formula, θ t represents the parameters of the model at time step t; θ t-1 represents the parameters of the model at the previous time step t-1; η represents the learning rate, that is, the step size of the model update; Represents the gradient of the loss function, that is, the direction and magnitude of the adjustment of the model parameter θ.

4. The method for monitoring and maintaining geothermal wells based on the Internet of Things according to claim 1, characterized in that: The verification mark generation process is as follows: S41, sensor data i Convert to binary string hdata i ; Among them, data i Indicates T i Sensor data acquired in real time at all times; S42, select a random number r∈Z q , taking the random number r as the first-order verification parameter; Where q is the order of the cyclic group G; G is a predefined cyclic group of order q; based on this, the bilinear map e is defined as: G×G→G T ; G T represents a cyclic group; Z p is the ring of integers modulo q; S43, calculate the second-order calibration parameter r': ； Where g is the generator of the cyclic group G; α and β represent predefined encoding parameters; α∈Z q ; β∈Z q ; Decoding parameters α' and β' are defined accordingly: α'=g α ; β' = g β ; and make the bilinear mapping e, decoding parameters α' and β' public; S44, output verification mark Lc i =(r, r').

5. The method for monitoring and maintaining geothermal wells based on the Internet of Things according to claim 4, characterized in that: The verification process for verifying the integrity and compliance of sensor data is as follows: S51, the received sensor data data i Convert to binary string hdata i ' ; S52, calculate the first-order check code Ce1: ; Wherein, g is a generator of a predefined cyclic group G; e is a predefined bilinear map; S53, obtaining a predefined second-order check code Ce2: Ce2=e(g, g); S54, if Ce1=Ce2, it means that the received sensor data data i Verification passed, that is, the received sensor data data i Complete and compliant.

6. The method for monitoring and maintaining geothermal wells based on the Internet of Things according to claim 1, characterized in that: The generation process of prediction results and optimization strategies is as follows: S61. Input the preprocessed data into the LSTM network. At each time step, the LSTM updates its own state according to the current input and the state at the previous moment. The LSTM learns long-term dependencies and processes the time series changes in the geothermal well data by controlling the forget gate, input gate, candidate memory unit, and output gate. S62: Build an intelligent fault prediction and optimization model, combine traditional Q-learning methods and deep reinforcement learning, and continuously optimize the prediction strategy based on real-time feedback, as follows: S6201, the intelligent fault prediction and optimization model uses a deep neural network to approximate the Q function. Specifically, the Q value is updated by the following formula: ； In the formula, s t Indicates the current state, corresponding to the current state of sensor data in the geothermal well system; a t represents the current action taken; θ Q Represents the parameters of the current neural network, i.e., weights, which are optimized through training; r t represents the reward at the current moment, reflecting the effect of the current action; γ represents the discount factor, which determines the importance of future rewards; Represents the parameters of the target network, which is used to calculate the target Q value; a' represents the next state s t+1 One of all the actions taken in The goal of updating the Q value is to optimize the current decision by maximizing future rewards. The specific update process is as follows: ； In the formula, y t Indicates the target Q value calculated according to the target network, which is used to compare with the Q value output by the current network; S6302: Update the parameters of the Q network and use the mean square error as the loss function to measure the difference between the current Q value and the target Q value: ； Where D represents the experience pool, which contains the historical state-action-reward quadruple, each of which represents the complete experience of the agent interacting with the environment at a certain moment: ; represents the expected operation on the samples sampled from the experience pool D; L(θ Q ) represents the loss function; The purpose of the loss function is to minimize the difference between the current Q value and the target Q value; S6303, Fault prediction and optimization model performs fault prediction, uses reward function to evaluate prediction accuracy, gives positive rewards when faults are successfully predicted, and imposes penalties when warnings are not given in time; S6304. Output prediction results and optimization strategies.

7. The method for monitoring and maintaining geothermal wells based on the Internet of Things according to claim 6, characterized in that: The fault prediction process is as follows: S71. Define the state space s t : The state space consists of sensor data, forming a vector s t ; S72. Define action space a t : Define the following actions: a1: No fault, continue monitoring; a2: A fault occurs and the early warning program is started; a3: A failure occurs and the equipment is shut down for maintenance; S73. Define reward function: Positive reward: If the model successfully predicts a failure and takes action, a positive reward is given: r t =+10, that is, if the fault is warned in advance and effectively responded to; Negative reward: If the fault is not discovered in time, resulting in equipment damage, a negative reward will be given: t =-50, that is, if a failure occurs and there is no early warning; Penalty Reward: Give negative rewards if unnecessary actions are taken prematurely: r t =-20, i.e. if premature fault warning leads to unnecessary downtime; S74: Q value update and model optimization, as follows: The state adopts: the processed sensor data received by the data analysis unit; Action selection: Select an action based on the current state, and use the ε-greedy strategy to select the action with the largest Q value; Experience replay: The current state, action, reward, and next state are stored in the experience pool. During each training, batches of data are randomly extracted from the experience pool for training to avoid the impact of the correlation of time series data on training. Q value update: according to the target Q value y t The difference between the current Q value and the current Q value updates the parameters of the neural network: ; Target network update: Set the update time interval to copy the weights of the current Q network to the target network to maintain the stability of the target Q value: .

8. The method for monitoring and maintaining geothermal wells based on the Internet of Things according to claim 1, characterized in that: The optimization objective function of the simulated annealing algorithm is: ； In the formula, C total represents the total maintenance cost; represents the personnel cost required for the ith maintenance task; represents the cost of the equipment required for the i-th task; Represents the cost of materials required for the i-th task.

9. A geothermal well monitoring and maintenance management system based on the Internet of Things, which is used to execute a geothermal well monitoring and maintenance management method based on the Internet of Things according to any one of claims 1 to 8, characterized in that: include: A sensor module, which integrates several sensor units and is used to collect geothermal well monitoring data; The data acquisition and transmission module is used to receive data from the sensor, convert it into digital signals, and upload the data to the data processing module using wireless communication technology; The data processing module is used to pre-process the received sensor data and transmit the pre-processed sensor data to the cloud platform and the big data analysis module; User interface and application module, used to provide an operation interface and a visual interface for field staff; Cloud platform and big data analysis module, used to store and analyze sensor data, including: A cloud data platform for storing, processing and analyzing data from geothermal wells, supporting large-scale data storage and efficient query; Data analysis unit, which is used to analyze the working status of equipment through machine learning and data mining technology, support anomaly detection and trend analysis, and help formulate maintenance plans; Intelligent decision-making unit, used to provide real-time decision support, generate maintenance strategies and predict future trends based on data analysis results.

Images