Energy underground subway station intelligent operation and maintenance method

Abstract

The present application relates to shallow geothermal energy development and utilization technology, especially to a kind of energy underground subway station intelligent operation and maintenance method, with the federal learning operation and maintenance control model constituted by central server-local client, the various local learning model training of 5 modules is carried out on local client, the calculation result is encrypted and uploaded to central processor by global federal learning model, the result obtained by global model is again issued to the local model of each local client, through the repeated iteration between client-central server, until global model reaches stability, finally make different module regulation and control scheme matched with each other.The present application can be used for real-time regulation and early warning;Realize the integrated regulation and control between station user end, unit equipment, energy tunnel, energy bottom plate, energy support pile multiple modules, the control scheme obtained is scientific and reasonable, improve the stability and service life of shallow geothermal energy underground subway station environmental control system.

Description

Technical Field

[0001] This invention relates to shallow geothermal energy development and utilization technology, and in particular to a smart operation and maintenance method for underground subway stations. Background Technology

[0002] The energy consumption of the environmental control system in underground subway stations accounts for as much as 30-40% of the total energy consumption of urban rail transit, indicating significant energy-saving potential. Shallow geothermal energy, as a widely distributed and easily developed renewable low-grade heat energy source, is mainly used for building heating and cooling. Replacing traditional air conditioning systems with shallow geothermal energy can reduce the energy consumption of subway station environmental control systems by about one-third.

[0003] Energy-powered underground subway stations utilize an emerging technology that extracts shallow geothermal energy for indoor temperature control by embedding heat exchange pipes within the station structure to create an energy station structure (such as energy tunnels, energy foundations, and energy support piles). This technology eliminates the need for additional drilling and heat exchange wells, reducing initial investment costs and giving traditional underground structures new functions, aligning with the development concept of integrated underground space development. Traditional operation and maintenance methods not only require significant manpower and financial resources but also suffer from issues such as control lag, poor coordination, and reduced equipment durability. Furthermore, during the construction or operation of energy-powered underground subway station structures, damage, blockage, or bending of the heat exchange pipes can hinder the normal circulation of the heat exchange medium, leading to unstable water pressure in various modules or even the entire system, corrosion and damage to key load-bearing components, and reduced heat exchange efficiency, ultimately causing the environmental control system to malfunction. However, current technologies for the inspection and maintenance of energy-powered underground subway station structures after construction are still immature. Therefore, scientific and rational operation and maintenance and control methods are important support for maximizing the energy-saving advantages of underground subway stations. The heat exchange pipeline maintenance technology during the operation and maintenance phase is a strong guarantee for ensuring the long-term stable operation of the station's environmental control system.

[0004] The Chinese utility model patent application number is CN202120598233.1, entitled "A Composite Heat Pump System Applicable to Subway Stations". It establishes a composite heat pump system consisting of a ground source heat pump system and a water loop heat pump system. In summer, the ground source heat pump system provides cooling for the station's public areas and management rooms, while in winter, the water loop heat pump system provides cooling.

[0005] Chinese invention patent application number CN201210585495.X, entitled "Ground Source Heat Pump Air Conditioning System for Subway Stations", describes a system consisting of a ground source heat pump unit, a buried pipe heat exchanger, a water storage tank, station air conditioning terminals, and building air conditioning terminals.

[0006] Chinese invention patent application number CN202010244228.0, entitled "A Method and Apparatus for Performance Testing and Optimization Control of a Ground Source Heat Pump System," discloses a method for calculating the real-time unit energy efficiency ratio of the system based on collected real-time operating data, thereby further determining the real-time operating status of the system. The system adjustment mode is determined by comparing historical operating data.

[0007] Chinese invention patent application number CN201910043376.3, entitled "A Performance Prediction Method for Ground Source Heat Pump Systems", discloses a ground source heat pump performance prediction method using a decision tree as the data structure, which includes borehole distribution pattern, borehole radius, buried pipe depth, number of buried pipes, lateral spacing of buried pipes, longitudinal spacing of buried pipes, thermal conductivity of the filling material inside the pipe, nominal outer diameter of the U-shaped pipe, and temperature of the soil and rock at the far end.

[0008] Currently, there is only a limited amount of research on ground source heat pump systems for subway stations, lacking research specifically on energy-powered underground subway station systems and the operation and maintenance control of energy station structures (energy support piles, energy tunnels, energy foundation slabs). Machine learning is a scientific technology that enables computers to automatically analyze patterns from a type of data using appropriate algorithms, and then use these patterns to predict unknown data. However, the aforementioned machine learning-based ground source heat pump performance prediction and control methods primarily focus on the control of the heat pump units. In practical engineering, the operation of energy-powered underground subway stations requires coordinated management of the operating states of indoor users, unit equipment, and buried pipes in the energy station structure to achieve a dynamically balanced and mutually compatible operating state. Controlling and managing only the heat pump units will accelerate equipment wear and tear and easily lead to unnecessary energy waste. Existing ground source heat pump control technologies for energy station structures are not suitable for the intelligent operation and maintenance control of energy-powered underground subway stations, and cannot achieve dynamic control of the unit equipment, station users, and energy station structure. Therefore, an easily scalable and simple collaborative control management method is needed for precise control in practical engineering.

[0009] Chinese utility model patent application number CN201721077937.4, entitled "A Permanently Detectable Polyethylene Composite Pipe", discloses a permanently detectable polyethylene composite pipe, comprising a double-layer structure pipe composed of a polyethylene layer and a modified PP coating layer. The pipe body is damaged by detecting the current of the tracer metal strip between the inner and outer layers using a multimeter.

[0010] Chinese invention patent application number CN202111660338.6, entitled "Anti-blockage and dredging system for municipal road engineering and its working method", discloses an anti-blockage and dredging system for municipal road engineering composed of a dredging trolley, a trolley drive mechanism, a visual inspection mechanism, an industrial and mining inspection mechanism and a display mechanism.

[0011] In the aforementioned composite pipeline technology, the metal strips require assembly and connection, which can easily lead to short circuits or open circuits during the construction phase. Furthermore, municipal pipeline dredging systems are limited to large-diameter municipal pipe networks. Existing technologies cannot be applied to the inspection and maintenance of heat exchange pipes within deeply buried, small-diameter energy station structures (energy support piles, energy tunnels, energy base plates). Summary of the Invention

[0012] To address the aforementioned technical problems, this invention provides a smart operation and maintenance method for underground subway stations, including an operation and maintenance control model. The operation and maintenance control model comprises five local client-side local models, including a station user-side control module, a generator equipment-side control module, an energy support pile buried pipe-side control module, an energy tunnel buried pipe-side control module, and an energy foundation slab buried pipe-side control module. The method specifically includes the following steps:

[0013] D1: Collect and organize information on underground subway stations for energy use as training sample data, and establish a local client-side local model. The local client-side local model includes a station user terminal control module, a generator unit terminal control module, an energy support pile buried pipe terminal control module, an energy tunnel buried pipe terminal control module, and an energy foundation slab buried pipe terminal control module. The training sample data includes station user terminal operating parameters and control parameters, generator unit operating parameters and control parameters, energy support pile buried pipe terminal operating parameters and control parameters, energy tunnel buried pipe terminal operating parameters and control parameters, and energy foundation slab buried pipe terminal operating parameters and control parameters. Perform simple data cleaning, analyze the characteristics of the sample data, and allocate them to the corresponding local clients.

[0014] D2: The station user terminal control module, the unit equipment terminal control module, the energy support pile buried pipe terminal control module, the energy tunnel buried pipe terminal control module, and the energy base plate buried pipe terminal control module learn and calculate based on the collected parameters, update the local model of the local client, and calculate the results;

[0015] D3: Encrypt the results of the local model calculations of the five local clients into desensitized parameters in the form of a public key, and upload them to the central server;

[0016] D4: The central server uses its private key to decrypt the encrypted and de-identified parameters uploaded by the five clients. The central server performs the decoding operation of the encrypted and de-identified parameters and performs secure aggregation. Then it updates the global shared model. After each round of model weight update, the central server calculates the error and accuracy of the global model.

[0017] D5: The central server generates a public key for encrypting data transmission using the global shared model and distributes it to each client. Each local client updates the iteration results of other relevant clients based on the global shared model as the new sample data attribute parameters.

[0018] D6: Repeat steps D2 to D5 iteratively until the global model is robust. Finally, each local client calculates the corresponding result based on the global shared model, and calculates the mutually matched control states in the energy underground subway station, including the opening degree of various regulating valves at the station user end, the set temperature of each room in the station, the opening degree of various regulating valves at the unit equipment end, the opening degree of various regulating valves at the energy support pile buried pipe end, the opening degree of various regulating valves at the energy tunnel buried pipe end, and the opening degree of various regulating valves at the energy base plate buried pipe end.

[0019] Preferably, the training sample data mentioned in step D1 can be obtained through numerical simulation calculation, on-site monitoring technology, questionnaire survey technology, image recognition technology, and thermal infrared technology. The data cleaning includes erroneous data, abnormal data, incomplete data, and duplicate data. The data cleaning content includes identifying invalid values, outliers, and missing values; and processing outliers, invalid values, and missing values.

[0020] Preferably, the update calculation of the local model of the station user terminal control module in step D2 includes the following steps: D2.1.1: Decrypt the public key of the encrypted transmission data using the private key, update the relevant clients of the energy support pile buried pipe end control module, energy tunnel buried pipe end control module, energy base plate buried pipe end control module, unit equipment control module, and station user control module according to the global shared model of the central server, and use the iteration result as the attribute parameters of the new sample data, calculate the information gain of all attributes of the training sample according to the information entropy, and sort all attributes according to the information gain;

[0021] D2.1.2: First, find the attributes with information gain higher than the average level from the candidate attributes, and then select the attribute with the highest gain rate from them to predict as the branch attribute of the decision tree;

[0022] D2.1.3: Then, treat each root node as a complete dataset, divide the samples according to the suboptimal attribute, and form leaf nodes by taking samples with the same attribute value as the same sample set. Use the MEP method to prune the decision tree, and iterate in this way to form a decision tree.

[0023] D2.1.4: Compare the station user-end control parameters determined by the decision with the equipment adjustment range. If the calculation result meets the adjustment range, the decision calculation result is used as advanced predictive control information, which is converted into encrypted parameters using a public key and uploaded to the central server for iteration of the globally shared model of federated learning. If the calculation result exceeds the adjustment range, proceed to step D2.1.5.

[0024] D2.1.5: Output the station user terminal control parameters determined by the decision and issue an alarm. Use the equipment control threshold as the advanced predictive control information, convert it into encrypted parameters using the public key, and upload it to the central server for the global shared model iteration of federated learning.

[0025] Preferably, the update calculation of the local model of the unit equipment end control module in step D2 includes the following steps: D2.2.1: Decrypt the public key of the encrypted transmission data using the private key, update the relevant clients of the energy support pile buried pipe end control module, energy tunnel buried pipe end control module, energy base plate buried pipe end control module, unit equipment control module, and station user control module according to the global shared model of the central server, and use the iteration result as the attribute parameter of the new sample data, and use the control valve opening parameter of the unit equipment as the learning target;

[0026] D2.2.2 To resolve the relationship between the independent and dependent variables, a slack variable ξ is introduced. i A nonlinear segmentation support vector classifier considering soft margins is constructed using the penalty coefficient C. The prediction accuracy and stability of the model are represented by the loss function, and its corresponding conditional extremum function is obtained.

[0027] D2.2.3 The above conditional extremum function is transformed into a multivariate function by using the Lagrange function. The partial derivatives of the Lagrange function with respect to the optimization objectives w, b, and ξ are set to 0 to obtain the Lagrange multipliers. The original conditional extremum function is transformed into the dual function, thereby finding the minimum value within the constraint region.

[0028] D2.2.4: Using a "linear kernel" to handle the inner product of mapping functions in dual functions;

[0029] D2.2.5: Optimize the model parameters in the local model of the support vector machine using simulated annealing, including the insensitive loss function ε, the penalty coefficient C, and the hyperparameters γ, λ, a, c, and d in the kernel function;

[0030] D2.2.6: Input the operating status parameters of the target unit equipment, and use the trained support vector machine local model to predict the control parameters of the computer unit equipment and the opening degree of related control valves;

[0031] D2.2.7: Compare the calculation results with the equipment adjustment range. If the calculation results meet the adjustment range, the predicted calculation results are used as advanced prediction control information and converted into encrypted parameters using the public key and uploaded to the central server for iteration of the globally shared model of federated learning. If the calculation results exceed the adjustment range, proceed to step D2.2.8.

[0032] D2.2.8: Output the calculated unit equipment control results and issue an alarm. Use the equipment control threshold as the advanced predictive control information, convert it into encrypted parameters using the public key, and upload it to the central server for the global shared model iteration of federated learning.

[0033] Preferably, the local model update calculation of the energy support pile buried pipe end control module in step D2 includes the following steps:

[0034] D2.3.1: Decrypt the public key of the encrypted data using the private key. Use the unit equipment operating parameters and station user operating parameters updated by the global shared model of the central server as new sample data attribute parameters. Based on the sample data, use the flow rate, pressure, temperature, and liquid level of the manifold, branch pipe, and cooling water in the energy support pile as the sample attribute space set, and use the relevant control valve opening parameters as the learning target.

[0035] D2.3.2: Sample data preprocessing, defining state vector X and target vector Y; performing quality analysis, feature analysis, and cleaning on the sample data; using sensitivity analysis to sort the sample attribute variables according to their correlation, taking the transformed attributes with the highest ranking as elements of the state parameters, where the state vector at time t is expressed as:

[0036] X(t)=[x1(t),x1(t-1),x2(t),x2(t-1),x3(t),x3(t-1),...,x n (t),x n (t-1)];

[0037] D2.3.3: The Manhattan distance is used to calculate the distance between each state vector in the state space, reflecting the degree of correlation between each state vector X and the target vector Y in the state space;

[0038] D2.3.4: Input the sample attribute information of the prediction set. The sample attributes include the temperature, pressure, and flow rate parameters of the cooling water at the buried pipe end of the energy support pile. The sample attributes of the training set are used to form a state vector. The K nearest neighbors of the current vector are searched in the training set to form a K-nearest neighbor prediction model. The K-nearest neighbor prediction model is then used to calculate the predicted valve opening degree corresponding to the prediction set. The expression of the K-nearest neighbor prediction model is: D2.3.5: Evaluation of prediction results for energy support pile control. The prediction results of the prediction set are compared with the labels corresponding to the samples in the prediction set to form a confusion matrix. Different evaluation indicators are calculated based on the confusion matrix to further evaluate the model prediction error and model accuracy. At least one of the following evaluation indicators should be selected for evaluation: sensitivity, accuracy, and precision.

[0039] D2.3.6: Input the basic operating parameters of the energy support piles to be predicted as the prediction set attribute space, and perform the adjustment and control prediction calculation of the energy support piles;

[0040] D2.3.7: Compare the calculation result with the control valve adjustment range. If the calculation result meets the adjustment range, the predicted calculation result is used as advanced predictive control information and converted into encrypted parameters using the public key and uploaded to the central server for iteration of the globally shared model of federated learning. If the calculation result exceeds the adjustment range, proceed to step D2.3.8.

[0041] D2.3.8: The calculated control results of the energy support pile buried pipe end will be output and an alarm will be issued. The control threshold will be used as the advanced prediction control information and converted into encrypted parameters using the public key and uploaded to the central server for the global shared model iteration of federated learning.

[0042] Preferably, the energy tunnel buried pipe end control module described in step D2 establishes a K-nearest neighbor local model of the control parameters of the energy tunnel buried pipe end based on various operating status parameters at different locations of the energy tunnel buried pipe end circulation pipeline, the unit equipment operating status parameters updated by the central server, and the station user end operating status parameters as sample attribute data. It then determines the opening degree of various regulating valves on the cooling water circulation pipeline of the energy tunnel end and uploads them to the central processor with encryption.

[0043] Preferably, the energy base plate buried pipe end control module in step D2 uses the temperature, flow, and pressure status parameters at different locations of the energy base plate buried pipe end pipeline, the unit equipment operation status parameters updated by the central server, and the station user end operation status parameters as sample attribute data to establish a K-nearest neighbor local model of the control parameters of the energy base plate buried pipe end, determine the opening degree of various regulating valves on the cooling water circulation pipeline of the energy base plate end, and encrypt and upload them to the central processor.

[0044] Preferably, step D1 further includes using a fault detection device to detect faults in the heat exchange tubes of the energy support piles, energy tunnel, and energy base plate. The fault detection device includes a double-layer intelligent heat exchange tube 1, a miniature magnetoelectric sensor 3, a cable tray 4, an automatic data acquisition device 5, and a control terminal 6. Multiple magnetic induction rings are evenly spaced on the double-layer intelligent heat exchange tube 1. The control terminal is electrically connected to the miniature magnetoelectric sensor, the automatic data acquisition device, and the cable tray. The double-layer intelligent heat exchange tube 1 includes an inner polyethylene tube 1-A, an outer polyethylene tube 1-B, and magnetic induction rings 2. The ring 2 is heat-fused and embedded in the wall of the outer polyethylene pipe 1-B, with a ring width of 5-8 mm and a spacing of 0.3-0.5 m. The inner wall of the outer polyethylene pipe 1-B is connected to the inner polyethylene pipe 1-A by threads. The miniature magnetoelectric sensor 3 includes a dredging module, a monitoring module, a magnetoelectric sensor 3-6, and a moving module. The dredging module includes a spiral stainless steel wire 3-1, a connecting component 3-2, a gear 3-3, and a series motor 3-5. The monitoring module includes a surround camera 3-4 and a lighting lamp 3-9. The moving module includes a drive wheel 3-7 and a corresponding motor.

[0045] Preferably, the fault detection includes the following steps:

[0046] S1: Before the test begins, turn on the power switch of the data acquisition instrument, and move a magnetic induction ring around the magnetoelectric sensor. When the magnetic induction ring encounters the sensor's sensing point, an audible and visual alarm will be triggered, and the instrument will show an indication, indicating that the data acquisition instrument is working normally.

[0047] S2: During testing, the miniature magnetoelectric sensor 3, containing the unblocking module, is placed into the heat exchange pipe along the pipeline. The pipeline arranged vertically inside the energy support pile is slowly inserted using the gravity of the magnetoelectric sensor itself and the rotation of the cable reel 4. The pipeline laid horizontally inside the energy base plate and the S-shaped pipeline arranged inside the energy tunnel are moved within the pipeline using the moving module of the miniature magnetoelectric sensor. The motors corresponding to the drive wheels 3-7 are turned on to make the miniature sensor move within the pipeline. When the center of the sensor intersects with the magnetic induction ring 2, the instrument emits a buzzer sound accompanied by a light indicator. At the same time, the monitoring module is turned on to perform real-time image monitoring and sound waveform monitoring within the pipeline. The automatic acquisition instrument 5 automatically records the number of each magnetic induction ring measuring point, the time interval between two adjacent measuring points, real-time image and sound waveform information.

[0048] S3: The automatic data acquisition device 5 transmits the automatically recorded data to the control terminal 6. At least one of the following methods can be used to analyze the acquired data and accurately locate the damage point: Isolation Forest algorithm, X-ray digital imaging technology, ultrasonic testing, ultrasonic guided wave, ultrasonic C-scan, ultrasonic phased array, or high-temperature thickness measurement method. The control terminal 6 can preliminarily locate the approximate location of the blockage point based on the analyzed data anomaly time and location. The analysis of real-time data using the Isolation Forest algorithm includes the following steps:

[0049] S3.1: Collect the current, acoustic, and electromagnetic signals transmitted from the unblocking module to the control center as sample attributes to form a sample set, which can be represented as X = (x1, x2, x3, ..., x...). n ), where x n =(x n1 ,x n2 ,x n3 ,...,x nm ), x nm The nth type attribute signal at time m;

[0050] S3.2: Use dimension q as the attribute cut point to segment attribute types, forming different isolated trees. For each isolated tree, use the cut point p as the outlier cut point to segment the attribute values ​​and divide the sample space. Q is a random value belonging to the range (0, n), and p is a value belonging to the range min(x). n ) < p < max(x) n Random values ​​within a given range;

[0051] S3.3: Based on the splitting of attribute cut points and outlier cut points, a hyperplane is formed in the sample space. Samples smaller than the cut point are classified into one class, forming the left node of the isolated tree, and vice versa.

[0052] S3.4: Repeat S3.2 to S3.3 until each isolated tree leaf node contains only one sample, thus completing the training of the isolated tree model;

[0053] S3.5: Take the sample data x at each time step i By traversing and calculating the data in an isolated tree model, the average height h(x) of each sample data can be obtained. i The outlier score of the sample data is calculated using the following formula:

[0054]

[0055] Where h(x) is the height of x in each tree, c(n) is the average path length for a given number of samples n, and E(h(x)) is the expected path length of x in multiple trees;

[0056] When the score S(x,ψ) is close to 1, the signal corresponding to the sample at that moment is considered an anomalous signal;

[0057] When the score S(x,ψ) is approximately equal to 0.5, the signal corresponding to the sample at that moment is considered a suspected anomalous signal;

[0058] When the score S(x,ψ) is close to 0, the sample at that moment is considered a normal signal.

[0059] Preferably, step D1 further includes repairing the faulty area, including the following steps:

[0060] T1: Turn on the surround camera 3-4 and the lighting 3-9, and transmit the data to the control terminal 6 in real time so that the camera is aimed at the blockage;

[0061] T2: Begin the unblocking work. If the blockage is found to be caused by pipe heat fusion, turn on the series motor 3-5 that controls the spiral stainless steel wire 3-1. After the stainless steel wire rotates to the working speed, if it is in a vertical pipe in the energy support pile, continue to slowly lower the sensor cable using the sensor's own weight. If it is in a horizontal pipe in the energy base plate or energy tunnel, use the moving module to control the sensor's movement and slowly lower the cable. The wire will stir the blockage until the blockage is cleared. If the blockage is caused by pipe heat fusion, the stainless steel wire 3-1 and connecting component 3-2 of the unblocking module can be replaced with a heat fusion rod, and the sensor can be placed at the blockage again. Turn on the heating rod to the heat fusion temperature of the polyethylene material, and slowly lower the sensor cable using the sensor's own weight until the blockage is cleared.

[0062] T3: After removing the miniature magnetoelectric sensor 3 from the pipeline, perform pipeline pressure testing and flushing to clean any remaining foreign objects inside the pipeline, thus completing the unblocking process.

[0063] T4: After accurately locating the damage point, an automatic repair agent is used to repair the damaged section. The automatic repair agent consists of polyethylene powder and granules, nylon fiber filaments, ethylene glycol, and adhesive. The automatic repair agent uses nylon fiber filaments with a fineness of 200-300D, and the fiber length is processed to 3-5mm. The automatic repair agent repair process includes the following steps:

[0064] T4.1: After locating the damage point, seal either end of the pipe inlet / outlet according to the pressure test standard, and inject an appropriate amount of automatic repair agent into the pipe through the other pipe opening;

[0065] T4.2: After injecting the repair agent, pressurize the pipe, but the pressure should not exceed the rated pressure that the polyethylene pipe used can withstand, so as to avoid pressure damage to the pipe.

[0066] T4.3: When the pressure gauge reading is stable, the repair of the damaged point can be considered to be basically completed. After the automatic repair agent has fully reached the strength requirements, the pipeline is pressure tested. Under the test pressure, the pressure should be stabilized for 15 to 20 minutes. After stabilization, the pressure drop should be 1% to 3%, and there should be no leakage. This is considered to be the completion of the repair work.

[0067] This invention provides a smart operation and maintenance method for underground subway stations powered by energy sources. It utilizes a control decision prediction model built upon historical data technology, overcoming the lag and reliance on experience inherent in traditional control methods. This model enables real-time regulation and early warning. It achieves integrated control across multiple modules, including station users, generator sets, energy tunnels, energy foundations, and energy support piles. The resulting control scheme is scientifically sound and improves the stability and service life of the environmental control system for shallow geothermal underground subway stations. The fault detection and repair method employs a pipeline inspection device combining a double-layer intelligent heat exchanger tube structure with a micro-magnetic sensor. The operation is simple, highly operable, easy to control, and readily implementable. The double-layer intelligent heat exchange tube structure with an embedded metal ring is simple and easy to prefabricate. The miniature magnetoelectric sensor serves both monitoring and unblocking functions, offering versatility. Traditional sensors are difficult to deploy and cannot accurately obtain anomaly information. The anomaly detection method based on the isolated forest algorithm overcomes the problem of difficult sensor deployment. By training the anomaly information recognition model through the analysis of various historical data, anomalies can be identified in a timely and accurate manner. During the detection process, if a blockage is found in the pipeline, the monitoring module and the unblocking module are immediately activated to unblock the pipeline. If the pipeline is damaged, the inner heat exchange tube can be unscrewed and replaced immediately, improving the efficiency of problem detection and resolution. The automatic repair agent is readily available and economical, easy to use, and provides good repair results, while also strengthening the structure of the heat exchange tube. Attached Figure Description

[0068] Figure 1 This is a schematic diagram of the operation and maintenance control method of an energy underground subway station according to the present invention.

[0069] Figure 2 This is a schematic diagram of a partial model structure of the decision tree of the station user terminal control module in the intelligent operation and maintenance method for underground subway stations of the present invention.

[0070] Figure 3 This is a schematic diagram of a partial model of the decision tree of the station user terminal control module in the intelligent operation and maintenance method for underground subway stations of the present invention.

[0071] Figure 4 This is a schematic diagram of the support vector machine prediction local model structure of the unit equipment control module in the intelligent operation and maintenance method for underground subway stations of the present invention.

[0072] Figure 5 This is a schematic diagram of the flow of the support vector machine prediction local model of the unit equipment control module in the intelligent operation and maintenance method of underground subway stations of the present invention.

[0073] Figure 6 This is a schematic diagram of the support vector machine algorithm improved by simulated annealing in the intelligent operation and maintenance method for underground subway stations of energy sources according to the present invention.

[0074] Figure 7 This is a schematic diagram of the process of the K-nearest neighbor local model of the control module K of the energy support pile buried pipe end control module of the present invention, which is a smart operation and maintenance method for energy underground subway stations.

[0075] Figure 8 This is a schematic diagram of the process of the K-nearest neighbor local model of the energy tunnel buried pipe end control module of the present invention, which is a smart operation and maintenance method for energy underground subway stations.

[0076] Figure 9 This is a schematic diagram of the process of the K-nearest neighbor local model of the energy base plate buried pipe end control module of the intelligent operation and maintenance method of energy underground subway station of the present invention.

[0077] Figure 10 This is a schematic diagram showing the connection of the fault detection device control terminal, automatic data acquisition instrument, and cable storage tray in the intelligent operation and maintenance method for underground subway stations of the present invention.

[0078] Figure 11 This is a schematic diagram of the pipeline layout inside the energy support piles in the intelligent operation and maintenance method for energy underground subway stations of the present invention.

[0079] Figure 12 This is a schematic diagram of the pipeline layout within the energy base plate in an intelligent operation and maintenance method for underground subway stations according to the present invention.

[0080] Figure 13 This is a schematic diagram of the pipeline layout in the energy tunnel in the intelligent operation and maintenance method for an energy underground subway station according to the present invention.

[0081] Figure 14 This is a schematic diagram of the double-layer intelligent heat exchanger pipe structure in the intelligent operation and maintenance method for underground subway stations of the present invention.

[0082] Figure 15 This is a schematic diagram of the cross-sectional structure of the double-layer intelligent heat exchanger tube in the intelligent operation and maintenance method for underground subway stations of the present invention.

[0083] Figure 16 This is a schematic diagram of the structure of a miniature magnetoelectric sensor in an intelligent operation and maintenance method for underground subway stations of energy sources according to the present invention.

[0084] Figure 17 This is a schematic diagram of the cross-section of a miniature magnetoelectric sensor structure AA in an intelligent operation and maintenance method for underground subway stations according to the present invention.

[0085] Figure 18 This is a schematic diagram of the isolated forest abnormal signal detection algorithm structure in the intelligent operation and maintenance method for underground subway stations of the present invention.

[0086] In the diagram: 1-A: Inner polyethylene pipe; 1-B: Outer polyethylene pipe; 2: Magnetic induction ring; 3: Miniature magnetoelectric sensor; 3-1: Spiral stainless steel wire; 3-2: Connecting component; 3-3: Gear; 3-4: Surround camera; 3-5: Series motor; 3-6: Magnetoelectric sensor; 3-7: Drive wheel; 3-8: Cable; 3-9: Lighting lamp; 4: Cable reel; 5: Automatic data acquisition device; 6: Control terminal; 7: Miniature magnetoelectric sensor wiring terminal; 7-1: Miniature magnetoelectric sensor wiring terminal inside the energy base plate buried pipe; 7-2: Miniature magnetoelectric sensor wiring terminal inside the energy support pile buried pipe; 7-3: Miniature magnetoelectric sensor wiring terminal inside the energy tunnel buried pipe. Detailed Implementation

[0087] This invention provides a smart operation and maintenance method for energy-efficient underground subway stations, such as... Figure 1 As shown, the system includes an operation and maintenance control model, which comprises five local client-side local models. These five local client-side local models include a station user-side control module, a generator set equipment-side control module, an energy support pile buried pipe-side control module, an energy tunnel buried pipe-side control module, and an energy foundation slab buried pipe-side control module. The specific steps include:

[0088] D1: Information on underground subway stations using energy sources was collected and organized as training sample data through numerical simulation, on-site monitoring, questionnaire surveys, image recognition, and thermal infrared technology. A local client-side model was established, which included a station user-side control module, a generator unit control module, an energy support pile buried pipe control module, an energy tunnel buried pipe control module, and an energy foundation buried pipe control module. The sample data included the operating parameters and control parameters of the station user-side, generator unit, energy support pile buried pipe, energy tunnel buried pipe, and energy foundation buried pipe. Simple data preprocessing and cleaning were performed, the characteristics of the sample data were analyzed, and the data was allocated to the corresponding local clients.

[0089] The data cleaning described herein targets erroneous data, abnormal data, incomplete data, and duplicate data. The content of data cleaning includes identifying invalid, outlier, and missing values; processing outliers, invalid, and missing values; the identification methods for invalid, outlier, and missing values ​​can be achieved through one of the following: 1. Visually identifying outliers as abnormal or invalid values ​​by drawing a scatter plot that reflects the relationship between two sets of data; 2. When the data follows a normal distribution, the 3σ principle can be used to identify outliers. In this case, values ​​in a set of measurements that deviate from the mean by more than three times the standard deviation are defined as outliers; 3. Statistically calculating the maximum, minimum, median, and upper and lower quartiles of the dataset and drawing a box plot, the upper and lower quartiles of the box plot can be used to identify outliers or invalid values.

[0090] Optionally, invalid, outlier, and missing values ​​can be handled using one of the following methods: 1. Fill in invalid, outlier, and missing values ​​using statistical data features based on the mean, median, and mode; 2. Fit invalid, outlier, and missing values ​​using a regression model or maximum likelihood estimation; 3. Delete them directly.

[0091] The station user terminal operating status parameters include: indoor ambient temperature, outdoor ambient temperature, passenger behavior status, equipment operating status, indoor user terminal manifold water level, differential pressure, manifold flow rate, manifold pressure, manifold temperature, indoor supply water temperature, indoor return water temperature, indoor supply water flow rate, indoor return water flow rate, indoor supply water pressure, and indoor return water pressure; the unit operating status parameters include: condenser working liquid level, condenser inlet temperature, evaporator working liquid level, evaporator inlet temperature, and compressor pressure; the energy support pile buried pipe end operating status parameters include: energy support pile manifold water level, differential pressure, manifold flow rate, and manifold temperature. The parameters for the energy tunnel buried pipe end include: energy support pile manifold water level, differential pressure, manifold flow rate, manifold pressure, manifold temperature, energy tunnel supply and return water temperature, energy tunnel return water temperature, energy tunnel supply and return water flow rate, and energy tunnel supply and return water pressure; the parameters for the energy base plate buried pipe end include: energy base plate manifold water level, differential pressure, manifold flow rate, manifold pressure, manifold temperature, energy base plate supply and return water temperature, and energy base plate return water temperature. The station user-end control parameters include: set temperature of the fan coil unit, opening degree of the indoor return water control valve, opening degree of the indoor supply water control valve, opening degree of the indoor user-end water collector manifold control valve, opening degree of the water collector branch pipe control valve, opening degree of the indoor user-end water distributor manifold control valve, and opening degree of the water distributor branch pipe control valve; the unit control parameters include: opening degree of the makeup water pump control valve, opening degree of the cooling circulation pump control valve, opening degree of the chilled water circulation pump control valve, and opening degree of the expansion valve; the energy support pile buried pipe end control status parameters include: opening degree of the energy support pile water collector manifold control valve. The opening degree of the control valve for the branch pipe of the water collector, the opening degree of the control valve for the water distributor of the energy support pile, the opening degree of the control valve for the branch pipe of the water distributor, the opening degree of the return water control valve for the energy support pile, the opening degree of the water supply control valve, the opening degree of the control valve for the water distributor of the energy tunnel, the opening degree of the control valve for the branch pipe of the water collector, the opening degree of the control valve for the branch pipe of the water distributor, the opening degree of the return water control valve for the energy tunnel, the opening degree of the water supply control valve, the opening degree of the control valve for the water distributor of the energy base plate, the opening degree of the control valve for the branch pipe of the water collector, the opening degree of the control valve for the branch pipe of the water distributor, the opening degree of the return water control valve for the energy base plate, and the opening degree of the water supply control valve.

[0092] D2: The station user terminal control module, the unit equipment terminal control module, the energy support pile buried pipe terminal control module, the energy tunnel buried pipe terminal control module, and the energy base plate buried pipe terminal control module learn and calculate based on the collected parameters, update the local model of the local client, and calculate the results;

[0093] Local Client 1: Station User Terminal Control Module, such as Figure 2As shown, based on the station user terminal operation status parameters, unit operation status parameters, energy support pile buried pipe operation status parameters, energy tunnel buried pipe operation status parameters, and energy foundation slab buried pipe operation status parameters as sample attribute data, a C4.5 decision tree local model of station user terminal operation and maintenance control parameters is established. This model determines the control parameters of the terminal fan coil units and the valve openings of the chilled water circulation pipelines, and then encrypts and uploads these parameters to the central processing unit. Figure 3 The C4.5 decision tree flowchart is provided for the station user terminal's operation control parameters. The decision tree can also be generated based on the ID3 algorithm and the CART algorithm. The specific steps of the s-th round of this client are as follows:

[0094] D2.1.1: Decrypt the public key of the encrypted data using the private key, update the relevant clients of the energy support pile buried pipe end control module, energy tunnel buried pipe end control module, energy base plate buried pipe end control module, unit equipment control module, and station user control module according to the global shared model of the central server, and use the iteration results as the attribute parameters of the new sample data. Calculate the information gain of all attributes of the training samples according to the information entropy, and sort all attributes according to the information gain; the calculation method of information entropy is as follows;

[0095] The total information entropy of a given sample S:

[0096] Attribute A has K distinct attribute values, therefore S is divided into K subsets. Calculate the information entropy of attribute A for a given sample S.

[0097] Calculate the information gain IG(S,A) of S partitioned according to attribute A: IG(S,A) = H(S) - H A (S),

[0098] The splitting information E(A) of the computed attribute. And information gain G(A),

[0099] D2.1.2: The larger the number of attribute values, the greater the splitting information, thus offsetting the impact of the number of attribute values. First, find the attribute with information gain higher than the average level from the candidate attributes, and then select the attribute with the highest gain rate as the branch attribute of the decision tree.

[0100] D2.1.3: Then treat each root node as a complete dataset, divide the samples according to the suboptimal attribute, and form leaf nodes by taking samples with the same attribute value as the same sample set. Use the MEP method to prune the decision tree, and iterate to form a decision tree. The REP method and PEP method can also be used to prune the decision tree.

[0101] D2.1.4: Compare the station user-end control parameters determined by the decision with the equipment adjustment range. If the calculation result meets the adjustment range, the decision calculation result is used as advanced predictive control information, which is converted into encrypted parameters using a public key and uploaded to the central server for iteration of the globally shared model of federated learning. If the calculation result exceeds the adjustment range, proceed to step D2.1.5.

[0102] D2.1.5: Output the station user terminal control parameter results determined by the decision and issue an alarm. Use the equipment control threshold as the advanced predictive control information, convert it into encrypted parameters using the public key, and upload it to the central server for the global shared model iteration of federated learning.

[0103] Local Client 2: Unit Equipment Control Module. Based on the station user terminal operating parameters, energy support pile buried pipe operating status parameters, energy tunnel buried pipe operating status parameters, energy foundation buried pipe operating status parameters updated iteratively by the central server, and the unit equipment terminal operating parameters of the local client, including the operating status parameters of the heat pump unit, cooling circulation pump, makeup water pump, and chilled water circulation pump, a support vector machine local model for unit equipment terminal control is established. The control scheme is predicted, calculated, and encrypted before being uploaded to the central processing unit. Figure 4 This is a support vector machine predictive model structure diagram for the unit equipment control module. It enables corresponding control adjustments based on the operating status of various unit equipment within the equipment room, adapting to changes at the station user end and the buried pipe ends of the three types of energy station structures. Figure 5 The flowchart of the support vector machine prediction model for the unit equipment control module is shown below. The specific steps of the model in the s-th round are as follows:

[0104] D2.2.1: Decrypt the public key of the encrypted data using the private key, update the relevant clients of the energy support pile buried pipe end control module, energy tunnel buried pipe end control module, energy foundation plate buried pipe end control module, unit equipment control module, and station user control module according to the global shared model of the central server, and use the iteration results as the attribute parameters of the new sample data to establish, as follows: Figure 5 The support vector machine local model of the unit equipment control module shown uses the unit equipment operating status parameters, including: condenser working liquid level, condenser inlet temperature, condenser flow rate, evaporator working liquid level, evaporator inlet temperature, evaporator flow rate, and compressor pressure, as sample attributes X. i This forms sample data X = {X1, X2, ..., X...} containing multiple features. n The relevant control valve opening parameters are used as the learning target y = {y1, y2, ... y}. n};

[0105] D2.2.2 To resolve the relationship between the independent and dependent variables, a slack variable ξ is introduced. iA nonlinear segmentation support vector classifier considering soft margins is constructed using the penalty coefficient C. The prediction accuracy and stability of the model are represented by the loss function L = max(0,|z|-ε), and its corresponding conditional extremum function is obtained.

[0106]

[0107] D2.2.3 The above conditional extremum function is transformed into a multivariate function by using the Lagrange function. The partial derivatives of the Lagrange function with respect to the optimization objectives w, b, and ξ are set to 0 to obtain the Lagrange multipliers. The original conditional extremum function is transformed into the dual function, thereby finding the minimum value within the constraint region.

[0108] D2.2.4: The inner product φ(X) of the mapping function in the dual function i ) T φ(X j The kernel used is a linear kernel: φ(X) i ) T φ(X j )=k(X i ,X j ) = X i T X j Process it;

[0109] D2.2.5: As Figure 6 The simulated annealing method is used to optimize the model parameters in the local model of a support vector machine, including the insensitive loss function ε, the penalty coefficient C, and the hyperparameters γ, λ, a, c, and d in the kernel function. The specific steps are as follows:

[0110] D2.2.5.1: Randomly generate an initial parameter set for interactive verification. Record the error value EEP as the current annealing system state E0, the initial temperature T0, and the annealing end temperature as T1.

[0111] D2.2.5.2: According to the perturbation algorithm m i '=m i +s·(μ-0.5)(B i -A i The parameters are perturbed to form a new parameter set, and the current annealing system state E is obtained through interactive verification. n Calculate ΔE = E n -E n-1 ;

[0112] D2.2.5.3: If ΔE < 0, accept the new parameter set and jump to step D2.2.5.5; otherwise, accept the corresponding parameter set according to the Metropolis criterion exp(ΔE / KT) - μ > 0 and jump to step D2.2.5.5; if none of the above conditions are met, reject the critical state and execute step D2.2.5.4.

[0113] D2.2.5.4: If the parameter set is rejected, return to step D2.2.5.2, re-perturb to generate a new parameter set, and perform interactive verification until the parameter set acceptance condition in step D2.2.5.3 is met;

[0114] D2.2.5.5: Set the end temperature to T1 as the algorithm exit point, and the global maximum number of EEP calculations to N. Annealing stops when T1 or N is reached. The cross-validation error E at this critical state is then accepted. n It should be the lowest E n The corresponding parameters should be the optimal prediction parameters; otherwise, return to step D2.2.5.2.

[0115] D2.2.6: Input the operating status parameters of the target unit equipment, and use the trained support vector machine local model to predict the control parameters of the computer unit equipment and the opening degree of related control valves;

[0116] D2.2.7: Compare the calculation results with the equipment adjustment range. If the calculation results meet the adjustment range, the predicted calculation results are used as advanced prediction control information and converted into encrypted parameters using the public key and uploaded to the central server for iteration of the globally shared model of federated learning. If the calculation results exceed the adjustment range, proceed to step D2.2.8.

[0117] D2.2.8: Output the calculated unit equipment control results and issue an alarm. Use the equipment control threshold as the advanced predictive control information, convert it into encrypted parameters using the public key, and upload it to the central server for the global shared model iteration of federated learning.

[0118] Local Client 3: Energy Support Pile Buried Pipe End Control Module. Based on various operating status parameters of the energy support piles, the unit equipment updated by the central server, and the operating status parameters of the user terminals within the station as the sample attribute space, the module sorts the attributes according to the selected distance metric, takes the top K attributes to establish a K-nearest neighbor local model of the control parameters of the energy support pile buried pipe end, determines the opening degree of various regulating valves on the cooling water circulation pipeline of the energy support pile end, and uploads them to the central processor with encryption. Figure 7 The flowchart for the K-nearest neighbor prediction model of the control module for the buried pipe end of the energy support pile is as follows:

[0119] D2.3.1: Decrypt the public key of the encrypted data using the private key. Use the unit equipment operating parameters and station user operating parameters updated by the global shared model of the central server as new sample data attribute parameters. Based on the sample data, use the flow rate, pressure, temperature, and liquid level of the manifold, branch pipe, and cooling water in the energy support pile as the sample attribute space set, and use the relevant control valve opening parameters as the learning target.

[0120] D2.3.2: Sample data preprocessing involves identifying outliers and invalid values ​​in the samples using box plots, correcting the identified outliers and invalid values ​​based on the median of their respective attribute values, and defining state vector X and target vector Y. Quality analysis, feature analysis, and cleaning are performed on the sample data. Sensitivity analysis can be used to rank the sample attribute variables according to their correlation, and the transformed attributes at the top of the ranking are taken as elements of the state parameters. The state vector at time t can be represented as:

[0121] X(t)=[x1(t),x1(t-1),x2(t),x2(t-1),x3(t),x3(t-1),...,x n (t),x n (t-1)]

[0122] D2.3.3: Using Manhattan distance Calculate the distance between each state vector X and the target vector Y in the state space to reflect the degree of correlation between each state vector in the state space. It can also be calculated using metrics such as Euclidean distance, Chebyshev distance, Minkowski distance, standardized Euclidean distance, Mahalanobis distance, cosine of the included angle, and combined distance.

[0123] D2.3.4: Input the sample attribute information of the prediction set. The sample attributes include the temperature, pressure, and flow rate parameters of the cooling water at the end of the energy support pile buried pipe. The sample attributes of the training set are used to form a state vector. The K nearest neighbors of the current vector are searched in the training set to form a K-nearest neighbor prediction model. The K-nearest neighbor prediction model is then used to calculate the predicted valve opening degree corresponding to the prediction set. The expression of the K-nearest neighbor prediction model is:

[0124] D2.3.5: Evaluation of Energy Support Pile Control Prediction Results. The prediction results of the prediction set are compared with the labels corresponding to the samples in the prediction set to form a confusion matrix. Based on the confusion matrix, different evaluation indicators are calculated to further evaluate the model's prediction error and accuracy. At least one of the following evaluation indicators can be selected: sensitivity, accuracy, and precision.

[0125] Sensitivity: The closer its value is to 1, the better the model fit.

[0126] Accuracy: The closer its value is to 1, the stronger the model's predictive ability;

[0127] The closer its value is to 1, the higher the model's prediction accuracy.

[0128] Table 1 Confusion Matrix

[0129]

[0130] D2.3.6: Input the basic operating parameters of the energy support piles to be predicted as the prediction set attribute space, and perform prediction calculations for the regulation and control of the end cooling water circulation loop;

[0131] D2.3.7: Compare the calculation result with the control valve adjustment range. If the calculation result meets the adjustment range, the predicted calculation result is used as advanced predictive control information and converted into encrypted parameters using the public key and uploaded to the central server for iteration of the globally shared model of federated learning. If the calculation result exceeds the adjustment range, proceed to step D2.3.8.

[0132] D2.3.8: The calculated control results of the energy support pile buried pipe end will be output and an alarm will be issued. The control threshold will be used as the advanced prediction control information and converted into encrypted parameters using the public key and uploaded to the central server for the global shared model iteration of federated learning.

[0133] Local Client 4: The energy tunnel buried pipe end control module uses the temperature, pressure, and flow parameters of the cooling water in the energy tunnel, the unit equipment updated by the central server, and the operating status parameters of the user terminals within the station as the sample attribute space. It sorts the attributes according to the selected distance metric and takes the top K attributes to establish a... Figure 8 The K-nearest neighbor local model of the control parameters at the buried pipe end of the energy tunnel is used to determine the opening degree of various regulating valves on the cooling water circulation pipeline at the energy tunnel end and upload them to the central processing unit with encryption. The sample data mainly includes: water level, differential pressure, flow rate, pressure, and temperature of the manifold at the buried pipe end of the energy tunnel, and cooling water temperature, flow rate, and water pressure inside the energy tunnel. For specific steps, please refer to the control module for the buried pipe end of the energy support pile.

[0134] Local Client 5: The control module for the buried pipe end of the energy base plate uses the temperature, pressure, and flow parameters of the cooling water in the energy base plate, the unit equipment updated by the central server, and the operating status parameters of the user terminals within the station as the sample attribute space. It sorts the attributes according to the selected distance metric and takes the top K attributes to establish a... Figure 9The K-nearest neighbor local model of the control parameters at the buried pipe end of the energy base plate is used to determine the opening degree of various regulating valves on the cooling water circulation pipeline at the energy base plate end and upload them to the central processing unit with encryption. The sample data mainly includes: water level, differential pressure, flow rate, pressure, and temperature of the manifold at the buried pipe end of the energy base plate, and cooling water temperature, flow rate, and water pressure inside the energy base plate. For specific steps, please refer to the control module of the buried pipe end of the energy support pile.

[0135] D3: The results of local model calculations for five local clients—the energy support pile buried pipe end control module, the energy tunnel buried pipe end control module, the energy floor buried pipe end control module, the unit equipment control module, and the station user terminal control module—are encrypted into de-identified parameters using a public key and uploaded to the central server. The relevant data for client i is represented as follows: Can be encrypted as

[0136] D4: The central server uses its private key to decrypt the encrypted and anonymized parameters uploaded by the five clients. The central server can then...

[0137] Decryption of encrypted and anonymized parameters is performed. The decrypted parameters of all client models are aggregated through federated learning based on knowledge distillation. The globally shared model of M clients is then updated according to the following formula: Among them W n+1 For the global sharing model in the nth round, For the nth round, the client sub-model weights of client i are uploaded to the server. After each round of model weight updates, the central server calculates the error and accuracy of the globally shared model. The central server can also update the globally shared model using synchronous methods such as gradient averaging and federated averaging.

[0138] D5: The central server generates a public key for encrypting data transmission using the global shared model and distributes it to each client. Each local client updates the iteration results of other relevant clients based on the global shared model as the new sample data attribute parameters.

[0139] D6: Repeat steps D2 to D5 iteratively until the global model is robust. Finally, the client calculates the corresponding results based on the global model. The system calculates the matching control states in the underground energy subway station, including the indoor fan coil unit temperature setting, chilled water return control valve opening, chilled water supply control valve opening, chilled water collector manifold control valve opening, collector branch control valve opening, chilled water distributor manifold control valve opening, distributor branch control valve opening, makeup water pump control valve opening, chilled water pump control valve opening, cooling pump control valve opening, expansion valve opening, buried pipe end collector manifold control valve opening, collector branch control valve opening, buried pipe end distributor manifold control valve opening, distributor branch control valve opening, and the supply and return water control valve openings for the three energy station structures (energy support piles, energy tunnels, and energy base plates).

[0140] like Figures 10-17 As shown, step D1 also includes using a fault detection device to detect faults in the heat exchange tubes of the energy support piles, energy tunnels, and energy base plate. If a heat exchange tube malfunctions, the accuracy of the collected sample data cannot be guaranteed, and the next step cannot be performed. The fault detection device includes a double-layer intelligent heat exchange tube 1, a miniature magnetoelectric sensor 3, a cable tray 4, an automatic data acquisition device 5, and a control terminal 6. Multiple magnetic induction rings are evenly spaced on the double-layer intelligent heat exchange tube 1. The control terminal is electrically connected to the miniature magnetoelectric sensor, the automatic data acquisition device, and the cable tray, respectively. The double-layer intelligent heat exchange tube 1 includes an inner polyethylene tube 1-A, an outer polyethylene tube 1-B, and magnetic induction rings 2. The ring 2 is heat-fused and embedded inside the wall of the outer polyethylene pipe 1-B, with a ring width of 5-8 mm and a spacing of 0.3-0.5 m. The inner wall of the outer polyethylene pipe 1-B is connected to the inner polyethylene pipe 1-A by threads. The miniature magnetoelectric sensor 3 includes a dredging module, a monitoring module, a magnetoelectric sensor 3-6, and a moving module. The dredging module includes a spiral stainless steel wire 3-1, a connecting component 3-2, a gear 3-3, and a series-wound motor 3-5. The monitoring module includes a surround camera 3-4 and a lighting lamp 3-9. The moving module includes a drive wheel 3-7 and a corresponding motor. The miniature magnetoelectric sensor 3 is applied to heat exchange pipes buried in energy support piles, energy base plates, and energy tunnel structures. During maintenance, the control cable is connected to the miniature magnetoelectric sensor terminal 7-1 in the energy base plate buried pipe, the miniature magnetoelectric sensor terminal 7-2 in the energy support pile buried pipe, and the miniature magnetoelectric sensor terminal 7-3 in the energy tunnel structure buried pipe.

[0141] The detection principle is as follows: When an electrical signal is applied to the magnetoelectric sensor, the electromagnetic frequency will change as the sensor passes through the magnetic ring. The electrical signal is transmitted to the automatic data acquisition instrument through the connected cable. The acquisition instrument automatically records the time when the sensor passes through each magnetic ring in sequence and transmits this information to the control terminal. If there is no electrical signal transmission to the next measurement point for a long time after passing through a certain measurement point, it is determined that there may be a blockage between the two measurement points.

[0142] The fault detection includes the following steps:

[0143] S1: Before the test begins, turn on the power switch of the data acquisition instrument, and move a magnetic induction ring around the magnetoelectric sensor. When the magnetic induction ring encounters the sensor's sensing point, an audible and visual alarm will be triggered, and the instrument will show an indication, indicating that the data acquisition instrument is working normally.

[0144] S2: During testing, the miniature magnetoelectric sensor 3 is inserted into the heat exchange tube along the pipeline. For vertically arranged pipelines within the energy support pile, the sensor 3 is slowly inserted using its own gravity and the rotation of the cable reel 4. For horizontally laid pipelines within the energy base plate and S-shaped pipelines within the energy tunnel, the miniature magnetoelectric sensor's moving module activates the motor corresponding to the drive wheel 3-7, causing the sensor to move within the pipeline. When the sensor's center intersects with the magnetic induction ring, the instrument emits a beep, accompanied by a light indicator. Simultaneously, the surround camera 3-4 and the lighting 3-9 are activated, transmitting data in real-time to the control terminal 6, ensuring the camera is aimed at the blockage. The automatic data acquisition instrument 5 automatically records the number of each magnetic induction ring measurement point, the time interval between adjacent measurement points, real-time image data, and acoustic waveform information.

[0145] S3: The automatic data acquisition instrument 5 transmits the automatically recorded data to the control terminal 6. The control terminal 6 can use the isolated forest algorithm to analyze the acquired data and accurately locate the damage point. Based on the analyzed abnormal time and location of the data, the approximate location of the blockage point can be preliminarily located. X-ray digital imaging (DR) technology, ultrasonic detection, ultrasonic guided wave, ultrasonic C-scan, ultrasonic phased array or high temperature thickness measurement method can also be used to analyze the acquired data and accurately locate the damage point.

[0146] The Isolation Forest algorithm is used to analyze real-time data. An isolation tree is built in the attribute space of the sample set to isolate attributes. Within the same attribute space, outliers are relatively distant from the attribute values ​​of most samples. Through model training, outliers are separated earlier, while normal values ​​are located in deeper child nodes of the isolation tree. The algorithm structure is as follows: Figure 18 As shown, the specific steps are as follows:

[0147] S3.1: Organize the current signal, sound wave signal, and electromagnetic signal transmitted from the unblocking module to the control center as sample attributes to form a sample set, which can be represented as X=(x1,x2,x3,...,xn ), where x n =(x n1 ,x n2 ,x n3 ,...x nm ), x nm Let n be the attribute signal of the nth class at time m.

[0148] S3.2: Use dimension q as the attribute cut point to segment attribute types, forming different isolated trees. For each isolated tree, use cut point p as the outlier cut point to segment attribute values, thus splitting the sample space. Q is a random value belonging to the range (0, n), and p is a value belonging to the range min(x). n ) < p < max(x) n () random values ​​within the range.

[0149] S3.3: Based on the splitting of attribute cut points and outlier cut points, a hyperplane is formed in the sample space. Samples smaller than the cut point are classified into one class, forming the left node of the isolated tree, and vice versa.

[0150] S3.4: Repeat S3.2 to S3.3 until each isolated tree leaf node contains only one sample, thus completing the training of the isolated tree model.

[0151] S3.5: Take the sample data x at each time step i By traversing and calculating the data in an isolated tree model, the average height h(x) of each sample data can be obtained. i ), outlier scores of sample data Identify abnormal signals based on the table below.

[0152] Table 2

[0153] S(x,ψ) value Signal evaluation 1 Abnormal signals 0.5 Suspected abnormal signal 0 Normal signal

[0154] Where h(x) is the height of x in each tree, c(n) is the average path length for a given number of samples n, and E(h(x)) is the expected path length of x in multiple trees;

[0155] When the score S(x,ψ) is close to 1, the signal corresponding to the sample at that moment is considered an anomalous signal;

[0156] When the score S(x,ψ) is approximately equal to 0.5, the signal corresponding to the sample at that moment is considered a suspected anomalous signal;

[0157] When the score S(x,ψ) is close to 0, the sample at that moment is considered a normal signal.

[0158] The ultrasonic phased array method is used to analyze real-time data. Specifically, the time delay of the excitation and reception pulses of each array element in the array transducer is controlled to change the phase relationship when the array element receives the sound wave from a certain point in the object. The focus point and the direction of the sound beam are changed to synthesize the phased array beam, thereby scanning the information.

[0159] Ultrasonic guided wave testing is used to analyze real-time data. Specifically, several sensor probes are placed at multiple locations within the pipeline to form a sensor array that detects the condition of the pipeline wall. As the information sensed by the sensors passes through the guided wave and is reflected, the damage to the pipeline wall is detected. Subsequent analysis and calculation of the relevant data identify the problems in the pipeline wall. To generate an appropriate waveform for typical pipe wall thicknesses, a much lower frequency than conventional ultrasonic testing is required. Guided waves typically operate at frequencies of 60–100 kHz. Therefore, the detection sensitivity of guided waves for individual defects is relatively low compared to ultrasonic testing, which typically uses frequencies in the MHz range. However, guided wave testing has the advantage of propagating over long distances of 20–30 meters with minimal attenuation, allowing for large-scale testing by fixing a pulse-echo array at a single location.

[0160] Ultrasonic C-scan detection is used to analyze real-time data. The specific method involves deep scanning of the pipeline. The basic working principle is that ultrasonic waves are generated inside the component being inspected by a reflective probe. The probe then transmits the received information about the defect location back as ultrasonic waves. The software displays various indicators of the reflected waves, thereby determining the specific location and size of the pipeline defect.

[0161] Step D1 also includes repairing the faulty area, including the following steps:

[0162] T1: Turn on the surround camera 3-4 and the lighting 3-9, and transmit the data to the control terminal 6 in real time so that the camera is aimed at the blockage;

[0163] T2: Start the unblocking work. If the blockage is found to be caused by pipe heat fusion, turn on the series motor 3-5 that controls the spiral stainless steel wire 3-1. After the stainless steel wire rotates to the working speed, continue to use the sensor's own weight to slowly lower the sensor cable. The wire stirs the blockage until the blockage is cleared. If the blockage is caused by pipe heat fusion, the stainless steel wire 3-1 and connecting component 3-2 of the unblocking module can be replaced with a heat fusion rod. Place the sensor back on the blockage, turn on the heating rod to the heat fusion temperature of the polyethylene material, and use the sensor's own weight to slowly lower the sensor cable until the blockage is cleared.

[0164] T3: After removing the miniature magnetoelectric sensor 3 from the pipeline, perform pipeline pressure testing and flushing to clean any remaining foreign objects inside the pipeline, thus completing the unblocking process.

[0165] T4: After accurately locating the damage point, an automatic repair agent is used to repair the damaged section. The automatic repair agent comprises polyethylene powder and granules, nylon fiber filaments, ethylene glycol, and a binder. The automatic repair agent uses nylon fiber filaments with a fineness of 200-300D, processed to a length of 3-5mm to ensure the mixture does not clump inside the pipe, thus reducing the repair agent's effectiveness. The nylon fiber filaments are intertwined with the polyethylene powder, rubber powder, and granules, making the polyethylene powder difficult to detach and participating in filling the holes in the heat exchange pipe. Because the pipe wall has a double-layer structure, the fiber filaments penetrate both sides of the holes, firmly fixing the polyethylene microparticles. The automatic repair agent repair process includes the following steps:

[0166] T4.1: After locating the damage point, seal either end of the pipe inlet / outlet according to the pressure test standard, and inject an appropriate amount of automatic repair agent into the pipe through the other pipe opening;

[0167] T4.2: After injecting the repair agent, pressurize the pipe, but the pressure should not exceed the rated pressure that the polyethylene pipe used can withstand, so as to avoid pressure damage to the pipe.

[0168] T4.3: When the pressure gauge reading stabilizes, the repair of the damaged point can be considered basically completed. After the automatic repair agent has fully reached the strength requirements, the pipeline should be pressure tested. Under the test pressure, the pressure should be stabilized for 15 to 20 minutes. After stabilization, the pressure drop should be 1% to 3%, and there should be no leakage. This is considered as the repair work being completed.

[0169] This invention provides a smart operation and maintenance method for underground subway stations powered by energy sources. It utilizes a control decision prediction model built upon historical data technology, overcoming the lag and reliance on experience inherent in traditional control methods. This model enables real-time regulation and early warning. It achieves integrated control across multiple modules, including station users, generator sets, energy tunnels, energy foundations, and energy support piles. The resulting control scheme is scientifically sound and improves the stability and service life of the environmental control system for shallow geothermal underground subway stations. The fault detection and repair method employs a pipeline inspection device combining a double-layer intelligent heat exchanger tube structure with a micro-magnetic sensor. The operation is simple, highly operable, easy to control, and readily implementable. The double-layer intelligent heat exchange tube structure with an embedded metal ring is simple and easy to prefabricate. The miniature magnetoelectric sensor serves both monitoring and unblocking functions, offering versatility. Traditional sensors are difficult to deploy and cannot accurately obtain anomaly information. The anomaly detection method based on the isolated forest algorithm overcomes the problem of difficult sensor deployment. By training the anomaly information recognition model through the analysis of various historical data, anomalies can be identified in a timely and accurate manner. During the detection process, if a blockage is found in the pipeline, the monitoring module and the unblocking module are immediately activated to unblock the pipeline. If the pipeline is damaged, the inner heat exchange tube can be unscrewed and replaced immediately, improving the efficiency of problem detection and resolution. The automatic repair agent is readily available and economical, easy to use, and provides good repair results, while also strengthening the structure of the heat exchange tube.

[0170] This invention has been described through embodiments. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of this invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of this invention.

Claims

1. A smart operation and maintenance method for underground subway stations, characterized in that, Includes the following steps: D1: Collect and organize information on underground subway stations for energy use as training sample data, and establish a local client-side local model. The local client-side local model includes a station user terminal control module, a generator unit terminal control module, an energy support pile buried pipe terminal control module, an energy tunnel buried pipe terminal control module, and an energy foundation slab buried pipe terminal control module. The training sample data includes station user terminal operating parameters and control parameters, generator unit operating parameters and control parameters, energy support pile buried pipe terminal operating parameters and control parameters, energy tunnel buried pipe terminal operating parameters and control parameters, and energy foundation slab buried pipe terminal operating parameters and control parameters. Perform simple data cleaning, analyze the characteristics of the sample data, and allocate them to the corresponding local clients. D2: The station user terminal control module, the unit equipment terminal control module, the energy support pile buried pipe terminal control module, the energy tunnel buried pipe terminal control module, and the energy base plate buried pipe terminal control module learn and calculate based on the collected parameters, update the local model of the local client, and calculate the results; D3: Encrypt the results of the local model calculations of the five local clients into desensitized parameters in the form of a public key, and upload them to the central server; D4: The central server uses its private key to decrypt the encrypted and de-identified parameters uploaded by the five clients. The central server performs the decoding operation of the encrypted and de-identified parameters and performs secure aggregation. Then it updates the global shared model. After each round of model weight update, the central server calculates the error and accuracy of the global model. D5: The central server generates a public key for encrypting data transmission using the global shared model and distributes it to each client. Each local client updates the iteration results of other relevant clients based on the global shared model as the new sample data attribute parameters. D6: Repeat steps D2~D5 iteratively until the global model is robust. Finally, each local client calculates the corresponding result based on the global shared model, and calculates the mutually matched control states in the energy underground subway station, including the opening degree of various regulating valves at the station user end, the set temperature of each room in the station, the opening degree of various regulating valves at the unit equipment end, the opening degree of various regulating valves at the energy support pile buried pipe end, the opening degree of various regulating valves at the energy tunnel buried pipe end, and the opening degree of various regulating valves at the energy base plate buried pipe end.

2. The intelligent operation and maintenance method for an underground subway station according to claim 1, characterized in that, The training sample data mentioned in step D1 can be obtained through numerical simulation calculation, on-site monitoring technology, questionnaire survey technology, image recognition technology, and thermal infrared technology. The data cleaning includes erroneous data, abnormal data, incomplete data, and duplicate data. The content of data cleaning includes identifying invalid values, outliers, and missing values; and processing outliers, invalid values, and missing values.

3. The intelligent operation and maintenance method for an underground subway station according to claim 1, characterized in that, The update calculation of the local model of the station user terminal control module described in step D2 includes the following steps: D2.1.1: Decrypt the public key of the encrypted data using the private key, update the relevant clients of the energy support pile buried pipe end control module, energy tunnel buried pipe end control module, energy base plate buried pipe end control module, unit equipment control module, and station user control module according to the global sharing model of the central server, and use the iteration results as the attribute parameters of the new sample data. Calculate the information gain of all attributes of the training sample according to the information entropy, and sort all attributes according to the information gain. D2.1.2: First, find the attributes with information gain higher than the average level from the candidate attributes, and then select the attribute with the highest gain rate from them to predict as the branch attribute of the decision tree; D2.1.3: Then, treat each root node as a complete dataset, divide the samples according to the suboptimal attribute, and form leaf nodes by taking samples with the same attribute value as the same sample set. Use the MEP method to prune the decision tree, and iterate in this way to form a decision tree. D2.1.4: Compare the station user-end control parameters determined by the decision with the equipment adjustment range. If the calculation result meets the adjustment range, the decision calculation result is used as advanced predictive control information, which is converted into encrypted parameters using a public key and uploaded to the central server for iteration of the globally shared model of federated learning. If the calculation result exceeds the adjustment range, proceed to step D2.1.

5. D2.1.5: Output the station user terminal control parameters determined by the decision and issue an alarm. Use the equipment control threshold as the advanced predictive control information, convert it into encrypted parameters using the public key, and upload it to the central server for the global shared model iteration of federated learning.

4. The intelligent operation and maintenance method for an underground subway station according to claim 1, characterized in that, The update calculation of the local model of the unit equipment control module mentioned in step D2 includes the following steps: D2.2.1: Decrypt the public key of the encrypted data using the private key, update the relevant clients of the energy support pile buried pipe end control module, energy tunnel buried pipe end control module, energy base plate buried pipe end control module, unit equipment control module, and station user control module according to the global shared model of the central server, and use the iteration results as the attribute parameters of the new sample data, and use the control valve opening parameters of the unit equipment as the learning target. D2.2.2 To resolve the relationship between the independent and dependent variables, slack variables are introduced. and penalty coefficient A nonlinear segmentation support vector classifier considering soft margins is constructed. The prediction accuracy and stability of the model are represented by the loss function, and its corresponding conditional extremum function is obtained. D2.2.3 The above-mentioned conditional extremum function is transformed into a multivariate function for solution using the Lagrangian function, and the Lagrangian function is set to the objective function. , , The partial derivatives are 0, so we obtain the Lagrange multipliers, which transform the original conditional extremum function into the dual function, thereby finding the minimum value within the constraint region; D2.2.4: Using a "linear kernel" to handle the inner product of mapping functions in dual functions; D2.2.5: Optimize model parameters in the local model of a support vector machine using simulated annealing, including the insensitive loss function. Penalty coefficient Hyperparameters in kernel functions , , , , ; D2.2.6: Input the operating status parameters of the target unit equipment, and use the trained support vector machine local model to predict the control parameters of the computer unit equipment and the opening degree of related control valves; D2.2.7: Compare the calculation results with the equipment adjustment range. If the calculation results meet the adjustment range, the predicted calculation results are used as advanced prediction control information and converted into encrypted parameters using the public key and uploaded to the central server for iteration of the globally shared model of federated learning. If the calculation results exceed the adjustment range, proceed to step D2.2.

8. D2.2.8: Output the calculated unit equipment control results and issue an alarm. Use the equipment control threshold as the advanced predictive control information, convert it into encrypted parameters using the public key, and upload it to the central server for the global shared model iteration of federated learning.

5. The intelligent operation and maintenance method for an underground subway station according to claim 1, characterized in that, The local model update calculation of the energy support pile buried pipe end control module mentioned in step D2 includes the following steps: D2.3.1: Decrypt the public key of the encrypted data using the private key. Use the unit equipment operating parameters and station user operating parameters updated by the global shared model of the central server as new sample data attribute parameters. Based on the sample data, use the flow rate, pressure, temperature, and liquid level of the manifold, branch pipe, and cooling water in the energy support pile as the sample attribute space set, and use the relevant control valve opening parameters as the learning target. D2.3.2: Sample data preprocessing, defining state vector X and target vector Y; performing quality analysis, feature analysis, and cleaning on the sample data; using sensitivity analysis to sort the sample attribute variables according to their correlation, taking the transformed attributes with the highest ranking as elements of the state parameters, where the state vector at time t is expressed as: ； D2.3.3: The Manhattan distance is used to calculate the distance between each state vector in the state space, reflecting the degree of correlation between each state vector X and the target vector Y in the state space; D2.3.4: Input the sample attribute information of the prediction set. The sample attributes include the temperature, pressure, and flow rate parameters of the cooling water at the buried pipe end of the energy support pile. The sample attributes of the training set are used to form a state vector. The K nearest neighbors of the current vector are searched in the training set to form a K-nearest neighbor prediction model. The K-nearest neighbor prediction model is then used to calculate the predicted valve opening degree corresponding to the prediction set. The expression of the K-nearest neighbor prediction model is: ; D2.3.5: Evaluation of prediction results for energy support pile control. The prediction results of the prediction set are compared with the labels corresponding to the samples in the prediction set to form a confusion matrix. Different evaluation indicators are calculated based on the confusion matrix to further evaluate the model prediction error and model accuracy. At least one of the following evaluation indicators should be selected for evaluation: sensitivity, accuracy, and precision. D2.3.6: Input the basic operating parameters of the energy support piles to be predicted as the prediction set attribute space, and perform the adjustment and control prediction calculation of the energy support piles; D2.3.7: Compare the calculation result with the control valve adjustment range. If the calculation result meets the adjustment range, the predicted calculation result is used as advanced predictive control information and converted into encrypted parameters using the public key and uploaded to the central server for iteration of the globally shared model of federated learning. If the calculation result exceeds the adjustment range, proceed to step D2.3.

8. D2.3.8: The calculated control results of the energy support pile buried pipe end will be output and an alarm will be issued. The control threshold will be used as the advanced prediction control information and converted into encrypted parameters using the public key and uploaded to the central server for the global shared model iteration of federated learning.

6. The intelligent operation and maintenance method for an underground subway station according to claim 1, characterized in that, The energy tunnel buried pipe end control module described in step D2 uses various operating status parameters at different locations of the energy tunnel buried pipe end circulation pipeline, the unit equipment operating status parameters updated by the central server, and the station user end operating status parameters as sample attribute data to establish a K-nearest neighbor local model of the energy tunnel buried pipe end control parameters, determine the opening degree of various regulating valves on the cooling water circulation pipeline of the energy tunnel end, and encrypt and upload them to the central processor.

7. The intelligent operation and maintenance method for an underground subway station according to claim 1, characterized in that, The energy base plate buried pipe end control module mentioned in step D2 uses the temperature, flow, and pressure status parameters at different locations of the energy base plate buried pipe end pipeline, the unit equipment operation status parameters updated by the central server, and the station user end operation status parameters as sample attribute data to establish a K-nearest neighbor local model of the control parameters of the energy base plate buried pipe end, determine the opening degree of various regulating valves on the cooling water circulation pipeline of the energy base plate end, and encrypt and upload them to the central processor.

8. A smart operation and maintenance method for an underground subway station based on energy resources, as described in claim 1, is characterized in that... Step D1 also includes using a fault detection device to detect faults in the heat exchange tubes of the energy support piles, energy tunnels, and energy base plate. The fault detection device includes a double-layer intelligent heat exchange tube (1), a miniature magnetoelectric sensor (3), a cable tray (4), an automatic data acquisition instrument (5), and a control terminal (6). The double-layer intelligent heat exchange tube (1) is provided with multiple magnetic induction rings at equal intervals. The control terminal is electrically connected to the miniature magnetoelectric sensor, the automatic data acquisition instrument, and the cable tray, respectively. The double-layer intelligent heat exchange tube (1) includes an inner polyethylene tube (1-A), an outer polyethylene tube (1-B), and magnetic induction rings (2). The outer polyethylene pipe (1-B) is hot-melted and embedded in the pipe wall, with a ring width of 5-8 mm and a spacing of 0.3-0.5 m. The inner wall of the outer polyethylene pipe (1-B) is connected to the inner polyethylene pipe (1-A) by threads. The micro magnetoelectric sensor (3) includes a dredging module, a monitoring module, a magnetoelectric sensor (3-6), and a moving module. The dredging module includes a spiral stainless steel wire (3-1), a connecting component (3-2), a gear (3-3), and a series motor (3-5). The monitoring module includes a surround camera (3-4) and a lighting lamp (3-9). The moving module includes a drive wheel (3-7) and a corresponding motor.

9. A smart operation and maintenance method for an underground subway station according to claim 8, characterized in that, The fault detection includes the following steps: S1: Before the test begins, turn on the power switch of the data acquisition instrument, and move a magnetic induction ring around the magnetoelectric sensor. When the magnetic induction ring encounters the sensor's sensing point, an audible and visual alarm will be triggered, and the instrument will show an indication, indicating that the data acquisition instrument is working normally. S2: During testing, the micro magnetoelectric sensor (3) with the unblocking module is placed into the heat exchange pipe along the pipeline. The pipeline arranged vertically in the energy support pile is slowly placed in by the gravity of the magnetoelectric sensor itself and the rotation of the cable tray (4). The pipeline laid horizontally in the energy base plate and the pipeline arranged in an S-shape in the energy tunnel are moved in the pipeline by the moving module of the micro magnetoelectric sensor. The motor corresponding to the active wheel (3-7) is turned on to make the micro sensor move in the pipeline. When the center of the sensor intersects with the magnetic induction ring (2), the instrument emits a beeping sound and is accompanied by light indication. At the same time, the monitoring module is turned on to monitor the image and sound waveform in the pipeline in real time. The automatic acquisition instrument (5) automatically records the number of each magnetic induction ring measurement point and the time interval between two adjacent measurement points, as well as real-time image and sound waveform information. S3: The automatic data acquisition instrument (5) transmits the automatically recorded data to the control terminal (6). At least one of the following methods can be used to analyze the acquired data and accurately locate the damage point: isolated forest algorithm, X-ray digital imaging technology, ultrasonic detection, ultrasonic guided wave, ultrasonic C-scan, ultrasonic phased array or high temperature thickness measurement method. The control terminal (6) can preliminarily locate the approximate location of the blockage point based on the abnormal time and location of the analyzed data. The analysis of real-time data using the isolated forest algorithm includes the following steps: S3.1: The current, acoustic, and electromagnetic signals transmitted from the unblocking module to the control center are collected as sample attributes to form a sample set, which can be represented as follows: ,in , The nth type attribute signal at time m; S3.2: Use dimension q as the attribute cut point to segment attribute types, forming different isolated trees. For each isolated tree, use the cut point p as the outlier cut point to segment the attribute values ​​and partition the sample space. Q is a random value belonging to the range (0, n), and p is a random value belonging to the range (0, n). Random values ​​within a range; S3.3: Based on the splitting of attribute cut points and outlier cut points, a hyperplane is formed in the sample space. Samples smaller than the cut point are classified into one class, forming the left node of the isolated tree, and vice versa. S3.4: Repeat S3.2~S3.3 until each isolated tree leaf node contains only one sample, thus completing the training of the isolated tree model; S3.5: Sample data at each time step The data is traversed and calculated using an isolated tree model to obtain the average height of each sample. The outlier score of the sample data is calculated using the following formula: ； in for At the height of each tree, For a given number of samples The average path length over time. for Expected path length in multiple trees; When the score When the value equals 1, the signal corresponding to the sample at that moment is considered an abnormal signal; When the score When the value is equal to 0.5, the signal corresponding to the sample at that moment is considered a suspected abnormal signal; When the score When the value equals 0, the sample at that moment is considered a normal signal.

10. A smart operation and maintenance method for an underground subway station based on energy resources, as described in claim 9, is characterized in that... Step D1 also includes repairing the faulty area, including the following steps: T1: Turn on the surround camera (3-4) and the lighting (3-9) and transmit the data to the control terminal (6) in real time to make the camera point at the blockage; T2: Start the unblocking work. If the blockage is found to be caused by pipe heat fusion, turn on the series motor (3-5) that controls the spiral stainless steel wire (3-1). After the stainless steel wire rotates to the working speed, if it is in the vertical pipeline in the energy support pile, continue to slowly lower the sensor cable using the sensor's own weight. If it is in the horizontal pipeline in the energy base plate or energy tunnel, use the moving module to control the sensor's movement and slowly lower the cable. The wire stirs the blockage until the blockage is cleared. If the blockage is caused by pipe heat fusion, the stainless steel wire (3-1) and connecting component (3-2) of the unblocking module can be replaced with a heat fusion rod and the sensor can be placed at the blockage again. Turn on the heating rod to the heat fusion temperature of the polyethylene material and slowly lower the sensor cable using the sensor's own weight until the blockage is cleared. T3: After removing the miniature magnetoelectric sensor (3) from the pipeline, perform pipeline pressure testing and flushing to clean the foreign objects remaining in the pipeline and complete the unblocking work; T4: After accurately locating the damage point, an automatic repair agent is used to repair the damaged section. The automatic repair agent consists of polyethylene powder and granules, nylon fiber filaments, ethylene glycol, and adhesive. The automatic repair agent uses nylon fiber filaments with a fineness of 200~300D, with the fiber length processed to 3~5mm. The automatic repair agent repair process includes the following steps: T4.1: After locating the damage point, seal either end of the pipe inlet / outlet according to the pressure test standard, and inject an appropriate amount of automatic repair agent into the pipe through the other pipe opening; T4.2: After injecting the repair agent, pressurize the pipe, but the pressure should not exceed the rated pressure that the polyethylene pipe used can withstand, so as to avoid pressure damage to the pipe. T4.3: When the pressure gauge reading is stable, the repair of the damaged point is complete; after the automatic repair agent has fully reached the strength requirements, the pipeline is pressure tested. Under the test pressure, the pressure should be stabilized for 15 to 20 minutes. After stabilization, the pressure drop should be 1% to 3%, and there should be no leakage to be considered as the repair work is complete.

Images